Two weeks ago I wrote that GPU rental prices were the most honest signal in AI infrastructure. Since then, three things have happened that materially strengthen the thesis, and I want to be direct about what they mean for anyone positioned in this space. This piece is a follow-up, but it is designed to stand on its own. If you take one thing from this letter, let it be this: the cost of building new AI infrastructure is rising faster than the market appreciates, and that has immediate, actionable consequences for how you should think about the companies that already own it.

What This Means for Your Portfolio

I am front-loading the actionable framework because I know many of you read these letters to sharpen your own process. Here are the four things I would be watching right now.

First, track the replacement cost curve. The cost of building a next-generation GPU rack has moved from roughly $6.5 million in January to quotes in the low-to-mid sevens today. High-bandwidth memory costs are projected to rise 40 percent for 2027 shipments. When replacement costs rise, the economic useful life of existing hardware extends. Every data centre that is already built, already energised, and already generating revenue becomes more valuable, not because anything about the asset changed, but because the cost of replicating it just went up. Look for companies with large installed GPU fleets and near-term capacity additions. They are the direct beneficiaries of this dynamic.

Second, follow the rental data. One-year H100 contract prices have now been revised upward twice in consecutive model updates. The latest forecast calls for $2.75 per GPU hour by December 2026, up from $2.44 just weeks ago. Rental prices are a real-time clearing price for physical compute. They cannot be fabricated. When they rise, margins expand mechanically for every provider with capacity coming off contract. The market structure matters here: most GPU rental volume is transacted on contracts of one year or longer, not on-demand spot pricing. On-demand listings only adjust after the contract market has already moved. The contract data leads. And the contract data is saying that providers can now negotiate for higher prepayments, longer tenors, and more favourable start dates. Time is on their side. If rental prices plateau at current levels, the economics are still materially better than what most models assume. The only scenario that undermines the thesis is a meaningful decline in rental rates, and the supply chain data suggests that decline is unlikely before late 2027 at the earliest.

Third, distinguish between demand risk and engineering risk. The market has historically priced AI infrastructure as though demand might evaporate. But $625 billion in signed Microsoft backlog and $240 billion in Google backlog are not demand that can evaporate. It is committed capital. The real risks are execution risks: can providers build fast enough, can they secure memory and components, can they energise on schedule? These are engineering problems, and they favour companies that are already built, already powered, and already contracted. A farmer does not abandon his land because one season is dry. He assesses the aquifer, checks the soil, and plants again.

Fourth, watch the power pipeline. In a world where every hyperscaler is supply-constrained and new grid connections take four or more years, the companies that control energised power capacity hold a structural advantage that no amount of capital expenditure can quickly overcome. Power is the new moat. The chips will change, the models will evolve, the customers will rotate. But the power stays. If you find this analysis useful, I encourage you to subscribe to follow the thesis as it develops. The data will continue to update, and I intend to write every time the picture changes materially.

The Three Data Points That Changed This Week

Now let me walk through the evidence in detail, because each of these data points reinforces the others.

The first is the rental price revision. Industry data from the GPU contract market shows H100 one-year rental prices continuing to climb, with the latest forecasts revised upward for the second consecutive update. The new target of $2.75 per GPU hour by December 2026 represents a 7 percent increase from the prior model just weeks earlier. This is not a forecast built on assumptions about demand growth. It is constructed from contracted transaction data across over one hundred market participants, validated by actual deals. Two consecutive upward revisions in a matter of weeks. The direction of travel is unambiguous.

The second is the cost escalation across the entire hardware stack. High-bandwidth memory, the specialised memory that sits alongside every AI processor, is now projected to rise roughly 40 percent for 2027 deliveries. HBM represents a significant portion of the chip bill of materials, and the increase flows directly into GPU package costs, server pricing, and rack-level capital expenditure. Nvidia’s Rubin racks, which were being quoted at $6 to $6.5 million in January, are now drawing quotes in the low-to-mid sevens. For Rubin GPU packages specifically, this translates to roughly $7,000 in additional cost per unit. The blended increase is modestly lower than the headline 40 percent due to shifting supplier mix assumptions, but the direction is clear: the replacement cost of AI infrastructure is rising, not falling.

The third is what happened at Anthropic. The company’s annualised revenue crossed $30 billion in April. Up from $9 billion at the end of 2025. Up from $1 billion in January of last year. That trajectory represents one of the fastest revenue ramps in the history of enterprise software. Salesforce took approximately twenty years to reach the same figure. More than 1,000 enterprise customers now spend over $1 million annually, a number that doubled in under two months. Claude Code alone generated over $2.5 billion in run-rate revenue by February. Altimeter Capital’s Brad Gerstner said publicly that he would not be surprised if Anthropic exits 2026 at $80 to $100 billion in annualised revenue. I do not know if that number proves accurate. But I know what it implies about the demand for compute. Every dollar of AI revenue requires physical infrastructure underneath it. Servers, power, cooling, interconnection. And Anthropic is one company among many. OpenAI sits at $24 billion in run-rate revenue. The demand for the infrastructure that supports these businesses is not theoretical. It is contractual and it is accelerating.

None of these are forecasts. The rental revision is based on contracted transaction data. The HBM cost increase is based on supplier pricing negotiations. The Anthropic revenue figure was confirmed by Bloomberg. I am not building an argument on projections. I am building it on what has already happened.

The Replacement Cost Thesis

This is the dynamic that I believe has been most underappreciated. The conversation about GPU infrastructure has been dominated by one concern: that new supply will flood the market and compress margins. The data tells a different story. Not only are rental prices for existing hardware rising, but the cost to build the next generation of that hardware is rising in parallel.

When the replacement cost of an asset rises, the economic useful life of the existing installed base extends. An H100 that was deployed two years ago does not become less valuable when its replacement costs more. It becomes more valuable, because the hurdle rate for displacing it has just gone up. This is not different from a building that appreciates when construction costs rise. The asset you already own is repriced upward by the market’s inability to cheaply replicate it.

There is a second-order effect that deserves attention. When replacement costs rise, the useful life assumptions embedded in financial models should extend. A GPU that costs more to replace will be operated longer. A fleet that generates cash flow for six years instead of four changes the net present value calculation substantially. The industry data confirms this logic explicitly: rising rental rates improve return on invested capital by expanding margins on already-deployed fleets, while simultaneously extending the economic useful life of existing GPUs, meaning invested capital generates cash flows for longer before requiring reinvestment.

The memory shortage driving these cost increases is not a temporary disruption. SK Hynix, Samsung, and Micron have effectively sold out their HBM capacity through 2026. All three manufacturers are reallocating wafer capacity from commodity DRAM toward HBM, which consumes approximately three times the wafer space per gigabyte. This is a zero-sum reallocation. Every wafer dedicated to an HBM stack for an AI accelerator is a wafer denied to the DRAM modules used in laptops, smartphones, and conventional servers. Conventional DRAM contract prices are projected to rise 55 to 60 percent in early 2026, with server DRAM climbing more than 60 percent. The memory manufacturers have made their choice: they are building for AI margins, not commodity volumes. Relief is not expected until new mega-fabrication facilities reach volume production in late 2027 at the earliest. This is not a supply chain hiccup. It is a structural reallocation of the world’s memory manufacturing capacity toward AI, and it has consequences that will persist for years.

The Demand Validation

I want to spend a moment on the Anthropic number because it deserves careful attention. Roughly 80 percent of their revenue comes from enterprise customers. These are not casual subscriptions that churn when novelty fades. They are multi-year API contracts, cloud provider integrations, and signed enterprise commitments. Eight of the Fortune 10 are now Claude customers. Business subscriptions to Claude Code have quadrupled since the beginning of 2026.

Connect this back to the infrastructure thesis. Every token served requires a GPU. Every enterprise customer running agentic workflows requires sustained, high-concurrency compute. Anthropic has signed a 3.5-gigawatt compute deal with Google and Broadcom, with capacity expected to come online beginning in 2027. The fact that one of the fastest-growing companies in the world is locking in multi-gigawatt capacity two years in advance tells you everything about the supply-demand imbalance.

And Anthropic is one company. OpenAI disclosed $2 billion in monthly revenue. Microsoft’s Azure backlog stands at $625 billion. Google’s cloud backlog surged 55 percent in a single quarter. These are not forecasts. They are contracts.

I want to address the counterargument directly, because I take it seriously. The bear case says these revenue numbers are unsustainable, that AI spending will cool, that enterprises will pull back once the novelty fades. I understand the instinct. Rapid growth invites scepticism. But when 1,000 businesses are each spending over $1 million annually, when that number is doubling every two months, when the revenue is embedding itself into code repositories and business-critical workflows, you are not looking at novelty-driven demand. You are looking at enterprise adoption that compounds. The kind of demand that does not reverse easily.

The Power Ceiling Hardens

In my prior piece, I described power as the new moat. The latest industry modelling reinforces this by introducing behind-the-meter gas turbine options as an alternative to grid power. This is not an academic exercise. It reflects the fact that behind-the-meter strategies are gaining real traction in the United States because grid capacity simply cannot keep up with AI deployment timelines.

Grid electricity costs currently sit around $0.087 per kilowatt hour for data centre operators. Behind-the-meter gas turbine power can be generated at roughly $0.064 per kilowatt hour under base assumptions, but actual costs swing dramatically by region and season, from sub-$4 per MMBTU in oversupplied markets like ERCOT to over $12 in constrained Northeast winter conditions. The geopolitical dimension adds further uncertainty. Energy price volatility driven by the conflict in Iran creates planning risk for any operator dependent on gas-fired generation.

For an infrastructure provider with locked-in renewable power at $0.033 per kilowatt hour, this volatility is irrelevant. A company sitting on nearly 3 gigawatts of secured grid-connected power at that cost is operating at roughly 40 percent of the cost that a new entrant would face trying to self-generate with gas turbines. That is not a narrow advantage. It is a structural moat that deepens with every passing quarter, and it is insulated from a category of geopolitical and commodity risk that new entrants must absorb entirely.

The Instruments Are Confirming

In recent weeks, the market has begun to catch up. Several infrastructure providers have re-rated 30 to 35 percent as investors start to recognise the dynamics described above. The question is whether the re-rating is complete, and I do not believe it is. The rental price revisions, the replacement cost increases, and the demand validation from the leading AI labs all suggest that the fundamental picture has improved faster than the stocks have moved, even after the recent run. When rental prices are revised upward twice in three weeks and memory costs are forecast to rise 40 percent for the next generation of hardware, the financial models that drove the prior valuations need to be rebuilt. Most have not been.

I return to the image of the instruments in the cockpit. Two weeks ago, the market’s inner ear was telling it that AI infrastructure was tilting toward a bust while the instruments said otherwise. The instruments were right. The rental data confirmed it. The cost data confirmed it. The revenue data confirmed it. The market has begun to trust the gauges, but the altitude indicator is still reading higher than where the aircraft is flying.

The risks remain real, and I take them seriously. GPUs depreciate. Technology evolves rapidly. Execution on construction timelines is never guaranteed. Geopolitical disruptions to energy markets add another layer of unpredictability. These are engineering risks, and they deserve rigorous analysis. But they are not demand risks. And engineering risks favour companies with strong balance sheets, purpose-built infrastructure, physical assets that retain value beyond any single generation of silicon, and locked-in power costs that insulate them from commodity volatility. The companies that control energised capacity today are not just ahead on the deployment timeline. They are structurally insulated from the cost pressures that every new entrant must absorb.

The cost of a shovel just went up. The gold is still real. And the ground has not moved.

I do not know what the market will do next quarter. I never have. But I know what the rental rates are doing, I know what the replacement costs are doing, I know what the revenue of the largest AI companies is doing, and I know the difference between a thesis built on hope and a thesis built on arithmetic. Ours is built on arithmetic. That is enough.

Neel Khokhani

Founder and CEO, Epochal Corporation

@neel_epochal

My son Maison is eight. A few weeks ago, he watched me reading about data centres and asked me a question I have not been able to shake.

“Dad, where does the internet live?”

I told him it lives in very large buildings filled with computers. He thought about this, then asked: “What happens when you run your computer for a really long time?” He knew the answer. “It gets hot.” Now imagine a million computers running at the same time, all day, every day, forever. “That would be really, really hot.” Yes it would. “So who keeps them cold?”

I paused. Because the truth is, I have spent months studying that exact question. And the more I studied it, the more I realised that the answer is one of the most important and most overlooked investments in the entire AI buildout.

He looked at me and said: “Dad, if nobody keeps them cold, do the computers break?”

Yes, Maison. They do. And when they break, it costs roughly $9,000 per minute. For a hyperscaler running a frontier AI training cluster, the number is far higher. A single thermal failure in a rack filled with Vera Rubin GPUs could destroy millions of dollars of silicon in seconds. The cooling system is not an amenity. It is the difference between a functioning data centre and a very expensive oven.

What Vera Rubin Actually Requires

Five days ago, Jensen Huang stood on stage at GTC and said he now sees at least $1 trillion in purchase orders for Nvidia’s Blackwell and Vera Rubin platforms through 2027. Last year, that number was $500 billion. It doubled in twelve months.

But the number is not what matters most. What matters is what sits inside the rack.

Each Vera Rubin GPU draws approximately 2,300 watts, a 64 percent increase over Blackwell at 1,400 watts. A single NVL72 rack consumes up to 230 kilowatts at peak power. That is the electrical load of 175 American homes in a cabinet the size of a refrigerator.

Vera Rubin’s compute trays are entirely fanless. Airflow requirements drop 80 percent. Liquid cooling flow nearly doubles. This is not air cooling with liquid assist. This is liquid cooling, period.

Nvidia made this choice because 2,300 watts per chip cannot be removed by air. The physics are non-negotiable. You are removing the heat output of a space heater from an object the size of a dinner plate, for 72 such objects simultaneously, without a single thermal excursion.

Rubin Ultra, on the roadmap for late 2027, draws 3,600 watts per GPU. A full Kyber rack of 576 GPUs will consume 600 kilowatts. The electricity of nearly 500 American homes in a single cabinet.

At that density, the building is no longer a data centre.

It is a cooling system that happens to contain computers.

Blackwell to Vera Rubin to Rubin Ultra to Feynman. Each generation hotter. The demand for cooling does not flatten until chips stop getting hotter. Thermodynamics guarantees this curve has no visible ceiling.

The Consensus Is Wrong

Here is what the market believes: cooling is a commodity input. The hyperscalers will buy it from whichever large manufacturer offers the best price. Carrier, Trane, Schneider, Vertiv, they all make cooling equipment. The market is competitive. Margins will compress. There is nothing special here.

This view is wrong, and it is wrong for a specific reason.

Vera Rubin does not need a cooling product. It needs a cooling solution engineered from scratch for a rack geometry, chip architecture, and power density that has never existed before. It needs a thermal path designed for 72 fanless GPU packages drawing 2,300 watts each, with liquid delivered at precise temperature, pressure, and flow rate, maintaining full redundancy in an enclosure where a single failure can destroy millions of dollars of hardware in minutes.

The giants of the cooling industry make excellent standardised products. But a standardised product cannot solve a bespoke problem. This is not a problem you solve with a catalogue. This is a problem you solve with engineers.

The Bespoke Manufacturer

We own a position in a company that I believe sits at the centre of this problem. I will not name it, because we are still building the position. But I will describe the business in enough detail that a serious reader can evaluate the thesis independently.

The company is a manufacturer of highly engineered, custom-built thermal management systems. Several years ago, it acquired a specialist in high-performance cooling for hyperscale data centres, with deep expertise in both airside and direct-to-chip liquid cooling. The co-founder of that acquired business is now the CEO of the entire company. His background is structural engineering and high-performance thermal systems. The board has made its bet on where the future lies.

The data centre cooling subsidiary’s revenue more than doubled in the past year. Backlog surged 140 percent to over $1.3 billion, with a book-to-bill of 2.4. The revenue run rate is approaching $550 million, with a target of $1 billion within three to four years. The company builds equipment designed to last 25 to 30 years, compared to the industry average of 15 to 20. It sells directly to building owners who optimise for total cost of ownership, not the lowest sticker price.

And the one we own is on the very short list of companies that can engineer what Vera Rubin demands.

I will be writing about the other side of this physical layer, the companies that build the ground beneath the data centres before the first server ever arrives, in my next letter.

Cooling Is Not a Cost. It Is Revenue.

Nvidia itself frames this in terms that should change how investors think.

In its technical documentation for Vera Rubin, the company describes “parasitic energy,” the roughly 30 percent of total data centre power lost to conversion, distribution, and cooling before it reaches the GPUs. At scale, that represents billions in wasted revenue. Every watt spent on cooling instead of compute is a watt that could have generated tokens. Every token not generated is revenue not earned.

A more efficient cooling system directly increases the tokens a facility produces. The cooling manufacturer is not selling a commodity. It is selling economic capacity. When GPU rental rates are surging 15 to 20 percent year to date and compute is sold out through year-end, every additional token is real money.

The Thermal Lock-In

When a hyperscaler designs a cluster around Vera Rubin, the cooling is co-engineered from the earliest design stages. Nvidia’s documentation describes Vera Rubin as requiring “nearly double” the coolant flow of the prior generation. New cold plates, new manifolds, new quick-disconnect fittings, new coolant distribution units, all engineered for this specific platform.

Once specified into a deployment, switching requires re-engineering the entire thermal path. When the cluster costs hundreds of millions and the customer needs it operational on schedule, the qualified partner is not replaceable.

What Happens After the Data Centres Are Built

A thoughtful reader should ask: once the cooling systems are installed, does the demand stop?

It does not.

GPU generations turn over every 18 to 24 months. A system designed for 1,400 watts per GPU cannot serve a chip drawing 2,300 watts. Every generation of silicon creates a new generation of cooling demand inside the same building. The installed base also generates recurring revenue through maintenance and replacement of components under continuous thermal stress.

And the buildout itself is not a single event. The world operates roughly 80 gigawatts of data centre capacity today. That doubles by 2028 and grows at 14 percent annually into the 2030s. The hyperscalers are not building one wave and stopping.

The demand for cooling is not a one-time purchase. It is a compounding obligation.

The Simple Version

Every chip needs to be cooled. The hotter the chip, the harder the cooling. The harder the cooling, the fewer companies can do it. The fewer companies that can do it, the more valuable those companies become.

Vera Rubin is the hottest chip ever built. The one after it will be hotter. And the one after that, hotter still.

That is the entire thesis.

This is the first of two letters on the physical layer of AI infrastructure. The next will examine the companies that build the ground beneath the data centres, before the first server ever arrives. Subscribe so you do not miss it.

Neel Khokhani

Founder and CEO, Epochal Corporation

@neel_epochal

Dear Partners,

I recently spent time reviewing transcripts from Data Center World London 2026. Not the keynotes or the panel discussions, but the expo floor conversations. More than thirty vendors across power generation, electrical distribution, cooling, and modular construction, talking candidly about what they can deliver, what they cannot, and what is sold out for years.

What follows is what I think matters for how we think about our holdings and the structural reality of this industry over the next several years.

From Equipment to Electrons

Eighteen months ago, the bottleneck in data center construction was equipment. Transformers carried multi-year lead times. Switchgear was backordered. Cooling systems were difficult to source. That era is ending. Low-voltage switchgear now ships in weeks. Cooling equipment carries lead times that are workable for any serious project. The equipment supply chain has largely normalised.

The constraint has moved. It now sits squarely on power. Reciprocating engines are sold out through 2028. Industrial gas turbines cannot be delivered before late 2030 or early 2031. The bottleneck on the turbine side is precision blade manufacturing, a process that does not scale quickly. Key transformer components carry three to five year lead times. Grid interconnections in constrained markets take twelve months or more before construction even begins.

The table below summarises the current order-to-delivery landscape as I understand it from the vendor conversations.

The question that governed data center procurement for the past two years, “can you get the equipment?”, has been replaced by a more fundamental one: “can you get the power?”

The Foundation You Cannot Pour Twice

There is a reason that the most expensive part of any skyscraper is not the steel or the glass. It is the foundation. The piling, the bedrock anchoring, the underground work that no one photographs. It takes the longest, costs the most, and once it is done, it cannot be replicated next door without starting from scratch. Everything visible above ground depends entirely on what was built below it.

Power is the foundation of the data center industry. You can source servers, order cooling, hire electricians, and pour concrete. But without power, none of it computes a single operation. And unlike equipment, power cannot be manufactured on demand. It is tethered to geography, to grid interconnection, to regulatory approval, and to physical generation assets that take years to build.

Companies that already control power, whether through existing grid interconnections, operational generation assets, or binding long-duration contracts with hyperscaler prepayments and parent guarantees, hold something that cannot be replicated on a reasonable timeline. Their advantage is not commercial. It is temporal. They have what others need and cannot quickly obtain. That is a moat measured not in intellectual property or brand loyalty, but in physics and time.

When Speed Costs More Than Steel

One of the more striking patterns in the vendor conversations was how thoroughly speed has displaced price as the dominant procurement criterion. The industry acknowledges that wet cooling systems offer roughly 30 percent lower capital costs and significantly better efficiency than air-cooled alternatives. The economics are not close. And yet, air-cooled systems dominate first-phase deployments because they compress timelines. Developers are choosing the more expensive option deliberately, planning to retrofit later, because delay costs more than inefficiency.

Prefabricated construction tells the same story. Over 90 percent of new builds from one major manufacturer are now prefabricated. Equipment pricing is roughly equivalent to traditional approaches, but containerised delivery compresses what used to be multi-year timelines into months.

When an industry starts optimising for speed over cost, it tells you something about the demand environment. No one rushes to build capacity they are not confident will be used. The urgency itself is the signal.

The Supply Side No One Reads

The financial media covers data centers through the lens of demand. Will the hyperscalers keep spending? Will AI justify the investment? These are reasonable questions. But they miss the supply side almost entirely.

Even if demand were to plateau at current levels, the supply of new capacity would remain constrained for years. Power generation is sold out through the end of the decade. Turbine blades cannot be manufactured faster. Transformer bushings carry multi-year lead times. This is not a story about whether demand will grow. It is a story about what happens when demand has already outrun the physical capacity to supply it.

I often think about what separates good investments from great ones. A good investment is cheap relative to what it earns. A great investment is cheap relative to what it earns, and structurally difficult to compete against. When your advantage is protected by the laws of physics and procurement timelines rather than by a patent or a brand, it tends to be more durable than the market gives it credit for.

That is where I believe our holdings are positioned. Not because we were clever about timing, but because we were disciplined about what we required before deploying capital. Real assets. Sound balance sheets. Operational infrastructure with binding contracts that would take years to replicate. The supply chain data confirms what the thesis anticipated: time is on our side.

The Romans had a tradition when they built an arch. The engineer who designed it was required to stand beneath it when the scaffolding was removed. We apply a version of that test to everything we own. And when I look at the supply chain data from this industry, I see something reassuring. The scaffolding is coming off, and the arches are holding.

Regards,

Neel

Since October 2025, GPU rental prices have been climbing. In January, the increases were gradual. By mid to late February, they went vertical. Year to date, rental rates for AI compute infrastructure are up 15 to 20 percent. The trajectory suggests a cumulative increase of roughly 40 percent through 2026. AI labs are accepting renewal price hikes of 40 percent to hold onto existing clusters. On demand capacity is sold out. Lead times for new deployments extend into the middle of the year.

This was not supposed to happen. The consensus view was that new supply would push prices down. The opposite has occurred.

The scarce resource in a gold rush is never the ambition. Ambition is abundant. The scarce resource is the shovel.

Right now, Microsoft has billions of dollars of AI chips sitting in warehouses. Not because nobody wants them. Because the electrical grid cannot absorb them. Satya Nadella said it himself: the constraint is not compute supply. It is the ability to get the builds done fast enough, close to power. The company has an $80 billion backlog of Azure orders it cannot fulfil. Not because demand softened. Because physics intervened.

Meanwhile, Google was forced to cut its own chip production targets. Not because customers cancelled. Because the foundry that makes every advanced AI chip on earth does not have enough capacity. Nvidia has locked up more than half of that foundry’s advanced packaging lines through 2027. Everyone else is fighting for the remainder.

The market is debating whether demand for AI infrastructure is real. The rental rates have already answered the question. The market just has not looked up yet.

I understand the instinct to distrust this. When prices move this fast, the natural reaction is suspicion. I share that instinct in most contexts. But I have a reference point that most financial analysts do not.

Before I managed capital, I built a pilot training business. Not a software company. A physical infrastructure business. Hangars, simulators, aircraft, fuel systems, runway access. I learned early that the person who owns the airfield is in a different position than the person who owns the plane. Planes come and go. Models change. Operators rotate. But the airfield, the physical capacity to receive them, endures. And when demand for flight hours exceeds available airfield capacity, the airfield owner does not need to negotiate. The scarcity does the negotiating for him.

That is the position AI infrastructure owners occupy right now. The chips will evolve. The models will change. The customers will rotate between architectures. But the rack space, the power, the cooling, the interconnection, these are the airfield. And the airfield is full, with a waiting list that stretches into next year.

$650 Billion and It Is Not Enough

In 2026, the four largest hyperscalers are on track to spend roughly $650 billion in capital expenditure. Amazon alone has committed approximately $200 billion. Alphabet has guided to between $175 billion and $185 billion. Meta sits at $115 billion to $135 billion. Microsoft is tracking toward $145 billion annualised. Their combined spend will exceed four times what the entire US energy sector deploys in a year. Amazon’s capex alone is larger than the entire US energy sector’s annual spend.

And it is not enough. Microsoft disclosed a remaining performance obligation of $625 billion. Azure grew 39 percent year over year while being explicitly capacity constrained. Google’s cloud backlog surged 55 percent in a single quarter to over $240 billion. Every hyperscaler has said the same thing on earnings calls: we are supply constrained, not demand constrained.

These are the most profitable enterprises in history. Microsoft generated $25.7 billion in free cash flow in a single quarter while deploying $37.5 billion in capex. Their liabilities to assets ratio actually declined to 48 percent in late 2025, near decade lows. Roughly 75 percent of aggregate capex in 2026 is directed at AI infrastructure. Even if AI revenues grew more slowly than expected, the traditional cloud business alone would sustain enormous demand for the capacity being built. The downside case is not zero demand. The downside case is somewhat slower growth in a market that is already larger than most countries’ GDP.

Goldman Sachs noted that AI capex relative to GDP remains well below the peaks seen in prior technology investment cycles. During the late 1990s telecom buildout, equivalent spending would have required roughly $700 billion in 2026 terms. And that cycle, for all the excess it produced, laid the physical foundation of the modern internet. The current cycle is arguably more durable because the demand signals are more concentrated, more contractual, and more visible than anything the telecom boom produced.

I will state my view plainly: hyperscaler capex in 2027 will approach one trillion dollars. Not because these companies want to make headlines, but because the backlog demands it, the competitive dynamics require it, and the balance sheets can support it. Microsoft’s remaining performance obligations grew 51 percent year over year. Google’s backlog surged 55 percent in a single quarter. If each of the five major hyperscalers increases spending by 40 to 50 percent in 2027, which is below the growth rate they achieved from 2024 to 2025, the aggregate figure approaches $900 billion to one trillion. And this excludes sovereign wealth funds, Chinese technology companies, and the emerging class of AI focused capital allocators.

The $650 billion is not evidence of exuberance. It is evidence of rationing.

Building Ahead of Demand vs Building Behind It

This is the distinction that matters most, and the one the market is getting wrong.

There are two kinds of infrastructure booms. In the first, companies build ahead of demand. They lay cable across oceans, pour concrete into speculative towers, and bet that the customers will eventually arrive. The fibre optic boom of the late 1990s was this kind. Companies built massive capacity on the assumption that internet traffic would fill it. For years, it did not. Utilisation rates sat in the single digits. The revenue was speculative. The bust was brutal.

The AI infrastructure cycle is the second kind. This is a build behind demand. The customers are already here. The contracts are already signed. Microsoft has $625 billion in remaining performance obligations. Google cut its own TPU production targets not because demand fell, but because it could not secure enough packaging capacity from TSMC. The hyperscalers are not building and hoping. They are building and still falling behind.

The difference is not a nuance. It is the entire difference between a speculative bubble and a supply constrained expansion. A bubble collapses when demand fails to appear. A supply constrained expansion resolves when capacity catches up. In one case, the asset becomes stranded. In the other, it becomes more valuable during the constraint period and normalises when supply equilibrates. The failure mode is entirely different.

Every time someone draws a comparison to the dot com bust, ask them one question: did Cisco have $625 billion in signed backlog when the music stopped? Did the fibre optic companies have customers willing to accept 40 percent price hikes to retain capacity? The answer to both is no. The structural conditions are not comparable. The comparison is convenient, but it is wrong.

This does not mean the spending is riskless. I want to be clear about that. GPUs depreciate. Technology evolves rapidly. Useful lives may prove shorter than assumed. Regulatory environments will shift. And the market will, at times, price these assets as though the demand were imaginary, even while the contracts sit in plain sight on the balance sheet. These are real considerations, and they deserve rigorous analysis. But they are engineering risks, not demand risks. And engineering risks are precisely the kind that favour companies with strong balance sheets, long duration contracts, and physical assets that retain value beyond any single generation of silicon. A farmer does not abandon his land because one season is dry. He assesses the aquifer, checks the soil, and plants again.

One Foundry, One Node, Every Chip

The reason the hyperscalers cannot spend more is specific and technical, and it has a name.

Every major AI accelerator launching in 2026 requires fabrication on a single process node at a single foundry. TSMC in Taiwan. The N3 family. Nvidia’s next generation, Google’s TPUs, Amazon’s Trainium, AMD’s Instinct. All of them. Competing for the same wafers from the same production lines. This convergence is not a temporary scheduling conflict. It is a structural collision. The entire AI industry is attempting to pass through a single gate at the same time.

There is one bakery in town that makes the bread everyone needs. It runs at full capacity every day. There are enough customers to buy three times the output. More ovens are being built, but they will not be ready for eighteen months. The bread is allocated, not sold. The customers who secured their orders earliest get served. Everyone else waits.

Nvidia has secured more than half of TSMC’s advanced packaging capacity through 2027. Google was forced to cut its 2026 TPU production target from four million to three million units because of limited packaging access. The advanced packaging process known as CoWoS, which integrates processors with high bandwidth memory on a silicon interposer, remains a separate and equally binding bottleneck. TSMC has been scaling this from roughly 35,000 wafers per month in late 2024 toward 130,000 by the end of 2026. That sounds like rapid growth. It is. But demand has grown faster. Every additional wafer is consumed before it comes online. High bandwidth memory from SK Hynix, Samsung, and Micron is fully allocated through 2026 with double digit price increases already underway.

Here is a number that puts the fragility of this supply chain in perspective. AI chips represented less than 0.2 percent of global wafer starts in 2024, yet already generated roughly 20 percent of semiconductor revenue. That extreme concentration on a single space explains why these shortages feel different from anything that came before. Today’s AI accelerators require leading edge logic, exotic memory stacks, and advanced packaging that cannot be expanded quickly. The bottleneck is not a temporary disruption. It is an architectural fact.

Behind the silicon wall sits the power wall. Data centres require enormous electricity. AI workloads require materially more than traditional cloud computing. The US grid was not built for this. Connection timelines stretch beyond four years in major markets. Northern Virginia and Texas are turning away new projects because they have exhausted available power capacity. Microsoft’s processors sit in warehouses waiting for the grid to catch up. Building a data centre now takes over three years. Securing utility interconnections takes longer. These are physical facts that do not compress in response to capital expenditure. For infrastructure providers that already have power, already have energised sites, and already have tenants, this is not a problem. It is a moat that widens with every month of delay for everyone else.

Why Rental Prices Are the Signal That Matters Most

I opened this letter with the rental rate data because I believe it is the single most important signal in AI infrastructure investing right now. More important than earnings beats, more important than capex guidance, more important than analyst sentiment. Let me explain why.

Rental prices are a real time clearing price for physical compute. They cannot be manufactured by press releases or inflated by accounting. When a GPU hour trades at a higher price than it did six months ago, that is the market telling you, in cash, that demand exceeds supply at the current price. There is no way to fake that signal.

What makes the current pricing environment especially striking is that the increases are no longer explained by rising input costs alone. Server component prices have risen, reflecting memory shortages and tariff pressures. But rental prices are now increasing well beyond what those input costs would justify with a constant margin. The implication is clear: margins for infrastructure owners are expanding, not compressing. And they are expanding at a pace that the market has not priced in.

From what I can observe, it has become a genuine challenge to procure not only next generation Blackwell capacity but even the prior generation Hopper capacity. At the higher performance tiers, availability is spoken for through August or September. At lower tiers, through May or June. The constraint is data centre space and GPU shipments. The market is tight across the entire stack. Anyone who needs compute and does not already have it under contract is facing a difficult few quarters.

The reason is structural. AI consumption has changed. The first wave was conversational: a question, an answer, a predictable amount of compute. The current wave is agentic. AI systems running multi step workflows autonomously, planning, executing, verifying, iterating at high concurrency. A single agentic session can consume more tokens than dozens of traditional conversations. The adoption curve for agentic coding tools echoes the original ChatGPT moment, with some tools already accounting for a meaningful share of public code commits. Media generation tools producing billions of images per month add another demand layer that did not exist eighteen months ago. Video generation, where each output requires processing enormous numbers of frames and pixels, is even more compute intensive. Users iterate rapidly to refine results, multiplying consumption with each attempt. The demand curve is not flattening. It is steepening.

For infrastructure owners, the margin story is transformative. Project level operating margins, already strong at roughly 40 to 45 percent, are expanding toward 50 percent and above on renewal. When an existing compute fleet reprices into the current rental environment, the EBIT increase on that capacity can approach 40 percent. Every tranche of capacity that comes off an older contract and re-enters the market generates materially higher cash flow. This is not a story of survival. It is a story of strengthening economics in a supply constrained environment. And the providers that have capacity coming off contract in the current pricing environment are the direct beneficiaries, even as the market punishes their stock prices on sentiment alone.

The Market Has It Backwards

Here is what I find remarkable. Investor sentiment toward AI infrastructure providers has turned increasingly negative at the exact moment the underlying economics have strengthened the most. Rental rates are surging. Margins are expanding. Capacity is sold out. Backlogs are growing. And the stocks are falling.

The market is pricing these companies as though $625 billion in signed backlog does not exist. That is not scepticism. That is innumeracy.

I have seen this pattern before. The physical economy and the financial economy diverge for a season, sometimes for several seasons. Sentiment leads price in the short term. Fundamentals lead price in the long term. And the reconciliation, when it comes, tends to be swift and uncomfortable for those on the wrong side.

I learned something crossing the Bass Strait at night in a single engine aircraft that has never left me. When you are suspended over black ocean with no visible horizon, your inner ear will lie to you. It will tell you the world is tilting when the instruments say you are level. The pilots who survive are the ones who trust the instruments over their own senses. The ones who follow the feeling into the darkness do not come back.

The market’s inner ear is telling it that AI infrastructure is tilting toward a bust. The instruments say otherwise. The contracts are binding. The rental rates are surging. The capacity is sold out. The backlog is growing. I trust the instruments.

The risks are real, and I take them seriously. GPUs depreciate. Technology evolves. Useful lives may prove shorter than assumed. Regulatory environments will shift. These are engineering risks, and they deserve careful analysis. But they are not demand risks. And engineering risks favour companies with binding contracts, strong counterparties, and physical assets that retain value beyond any single chip generation. A farmer does not abandon his land because one season is dry. He assesses the aquifer, checks the soil, and plants again.

The tenants are the most creditworthy companies in the history of capitalism. The contracts are binding, multi year, and prepaid. The utilisation is at or near one hundred percent. The rental rates are surging. And the supply chain cannot deliver fast enough to satisfy what the hyperscalers have already committed to purchase.

In most market environments, the primary risk for infrastructure assets is demand. Will the tenants show up? Will the contracts be renewed? Will the utilisation hold? In the current cycle, those questions have been answered. The risk has inverted. The risk is not that demand evaporates. The risk is that the supply chain cannot deliver fast enough. For those who own the physical assets, this is the position of the landlord in a city where every building is full, where the waiting list is growing, where new construction will take years, and where the tenants are Fortune 5 companies who have signed long term leases at rates the market did not expect.

The world wants to build the infrastructure for the age of intelligence. The shovels are in short supply. The gold is real. And we own the ground where both converge.

I do not know what the market will do next quarter. I never have. But I know what the contracts say, I know what the rental rates are doing, and I know the difference between a thesis built on hope and a thesis built on arithmetic. Ours is built on arithmetic. That is enough.

Neel Khokhani

Founder and CEO, Epochal Corporation

@neel_epochal

Dear Partners,

There is an old parable about a man who buys a house based on a photograph. The photograph shows a beautiful facade, a manicured lawn, a blue sky. He signs the papers. He moves in. On the first night, the roof leaks, the foundation is cracked, and the pipes are rusted. The photograph was real. The house was real. But the photograph showed only what the seller wanted him to see.

I have been thinking about this parable a great deal lately, because it captures something important about what is happening in the energy infrastructure buildout right now, and because it explains both why we avoided one side of this trade and why we chose the other.

The Promise of Load

This past utilities earnings season has been extraordinary, at least on paper. American Electric Power doubled its contracted load outlook from 28 GW to 56 GW. FirstEnergy increased its five-year capital expenditure plan by $8 billion. DTE by $6.5 billion. Entergy by $6 billion. The numbers are enormous. The narrative is compelling. The story is that data centers will consume unprecedented amounts of electricity, and that anyone who owns the pipes and wires feeding that demand will prosper for decades.

Markets love a clean narrative. Clean narratives feel safe. They give investors permission to pay premium valuations for what appears to be certainty. But I have learned, sometimes painfully, that the distance between a narrative and a fact can be very wide. And the job of a careful investor is not to accept the narrative. The job is to read the contract.

So I read the contracts.

What the Contracts Actually Say

The architecture behind these utility load projections is more fragile than most investors realise. Let me walk you through the structure, because I believe understanding it is essential to understanding where we have placed our capital and why.

When a hyperscaler like a major cloud provider wants to bring a large data center onto a utility’s grid, the process moves through several stages. The first stage is an engineering study, called a Substation Engineering Letter of Authority. The customer pays a flat fee of $1 to $5 million, regardless of how much power the project will eventually require. If the customer walks away at this point, that fee is all they lose.

The second stage authorises actual substation construction. This costs the customer roughly $17 to $35 million, depending on the voltage class and configuration. These costs are flat. They do not scale with the megawatts contracted.

The third and final stage is the Electric Service Agreement, a bilateral take-or-pay contract that requires approval from a state public utility commission. Once the meter is set and the generation assets come online, the customer commits to paying for at least 80 to 85 percent of contracted capacity, regardless of how much they actually use. This sounds binding, and it is, but only after the infrastructure is physically built and operational. Before that point, the customer’s total sunk cost is $26 to $40 million. On a $1.1 billion reference project, that is roughly 2.4 to 3.6 percent. On the supply side, the numbers are even thinner. An independent power producer entering the PJM interconnection queue posts about $2 million for a 500 MW project, with development costs bringing the total to $7.5 to $12.5 million. That is less than 1.1 percent of total project cost. In ERCOT, which is Texas, the supply side commitment can be as low as $1 to $2 million. Under 0.2 percent.

Let me put this plainly. The contracts that underpin tens of billions in projected utility capital expenditure create near-optional financial obligations on both sides. The customer can walk away for the cost of a rounding error. The generator can walk away for even less.

The Cracks

This is not a theoretical concern. The cracks have already appeared.

AEP built 750 MW of generation capacity in Indiana, West Virginia, and Kentucky in anticipation of data center contracts that never materialised. The company applied to FERC for a waiver to sell that stranded capacity into PJM’s incremental capacity auction. FERC rejected the request in February 2026. AEP now holds generation capacity with no offsetting revenue and no resolution mechanism.

Consider what happened there. The utility built the supply. The demand never showed up. And because the contracting structure placed no proportional take-or-pay obligation on the customer at the generation level, the utility absorbed the entire loss. It is like a farmer who ploughs a hundred acres, buys the seed, hires the labour, and then discovers the buyer never actually committed to purchasing the harvest.

Then there is the dispute between PacifiCorp and Amazon. PacifiCorp allegedly delivered significantly less power than contracted to one Amazon data center campus and zero power to a second, while refusing to complete contracts for two additional campuses with obligations dating to 2021. Amazon claims it invested $39.2 billion in Oregon development on the assumption that contracted delivery would occur. The case is proceeding to discovery. It is the first real test of whether a utility faces material consequences for not performing under an Electric Service Agreement.

The answer remains unresolved. Which tells you something about the certainty embedded in these projections.

The Photograph and the House

This is why I return to the parable of the photograph.

The load projections are real in the sense that they represent genuine demand intent. Hyperscalers want this power. They are willing to pay for it. The secular trend toward artificial intelligence, toward the digitisation of everything, is not imaginary. But demand intent is not the same as a binding financial commitment, and a signed letter of authority is not the same as a delivered megawatt.

Utilities are disclosing load projections derived from contracts that, at the point of signing, carry sunk costs of 2 to 4 percent of project value on the demand side and under 1 percent on the supply side. The market is pricing many utility equities as though these projections represent certainty. They do not. They represent the most optimistic scenario within a distribution of outcomes that includes delays, cancellations, and stranded capital.

The distinction matters. Not because the energy buildout will not happen. Parts of it almost certainly will. It matters because the distance between the narrative and the contractual reality is where risk lives. And where risk lives quietly, loss compounds just as surely as value does elsewhere.

This is exactly why we avoided the utility side of this trade. But it is also why we looked carefully at the other side of the table.

The Other Side of the Table

There is a critical difference between a utility projecting load based on letters of authority and an infrastructure company that has already signed binding, long-duration contracts directly with the hyperscalers themselves. That difference is the reason we own a basket of digital infrastructure companies, and it deserves a careful explanation.

When I look at our holdings in this space, the companies formerly rooted in bitcoin mining that have pivoted to high-performance computing infrastructure, I see something structurally different from the utility story I just described. The difference is not subtle. It is contractual.

Consider the contrast. A utility like AEP discloses 56 GW of projected load in ERCOT, much of it based on letters of authority where the customer’s walk-away cost is flat and immaterial relative to the project. The utility bears the risk of building generation and transmission infrastructure against what are, in financial terms, soft commitments. If the customer does not show up, the utility is left holding stranded assets. That is precisely what happened with AEP’s 750 MW.

Our infrastructure holdings sit on the opposite side of that dynamic. They are not projecting demand based on preliminary letters. They have executed contracts with counterparties like Microsoft, Amazon, and Google, contracts that in several cases involve billions of dollars in prepayments, 10 to 15 year terms, and explicit guarantees from the parent entities of the tenants. These are not options. These are obligations.

Let me be specific about what gives me comfort.

The first source of comfort is the nature of the counterparty commitment. When a hyperscaler signs a 15-year lease with an infrastructure provider and the parent company, Amazon.com itself, guarantees the base rent and expenses, that is a fundamentally different financial instrument than a letter of authority that costs $5 million to exit. When Google backstops over a billion dollars of a tenant’s lease obligations, that is not a soft signal of interest. That is capital at risk. The hyperscalers are not window-shopping with these infrastructure providers the way they might be with certain utility load requests. They are writing cheques that bind them.

The second source of comfort is the physical asset base. These companies own land, power capacity, substations, fibre connections, and data center shells. The assets exist independent of any single contract. If a tenant were to default tomorrow, the infrastructure would still be standing, the power would still be connected, and the sites would still be among the most valuable parcels of energy-connected real estate in the country. In an environment where securing grid-connected power capacity is the primary bottleneck in artificial intelligence infrastructure, the land and the megawatts are the scarce resource. These companies have already secured them. That is not a narrative. It is a physical fact.

The third source of comfort is the direction of the secular trend. I am not someone who invests on the basis of macro narratives alone. But I also do not ignore structural realities. Every major technology company on earth is racing to build computing capacity for artificial intelligence. The demand for power-dense, purpose-built data centers is not speculative. It is observable in the capital expenditure plans of companies that collectively spend hundreds of billions of dollars per year. Our infrastructure holdings are positioned directly in the path of that spending, with signed contracts that convert that macro trend into specific, measurable revenue obligations.

The Honest Accounting

I would not be doing my job as your fiduciary if I presented only the comfort and ignored the risks. Intellectual honesty requires the full picture.

The most important risk is that these companies have taken on significant debt to fund their buildouts. This is a departure from our usual preference, and I want to be transparent about why I have accepted it in this case and where my concern remains.

Several of our infrastructure holdings have issued billions of dollars in convertible notes and project-level bonds to finance the construction of their data center campuses. The debt is real and it is large. In some cases, the debt-to-equity ratios have risen to levels I would normally avoid. This introduces the very thing I have always warned about: a clock. Debt requires servicing regardless of whether construction is on schedule, regardless of whether energisation dates slip, regardless of whether the first dollar of contracted revenue arrives on time.

I have accepted this risk for a specific reason. The debt in these structures is not speculative leverage. It is project finance, matched against contracted revenue streams with creditworthy counterparties. In several cases, the construction is substantially funded by customer prepayments, meaning the hyperscaler is financing its own future capacity through the infrastructure provider. This is different from a company borrowing to fund an idea. This is a company borrowing against a signed contract with a Fortune 10 counterparty.

But I do not want to rationalise away the risk. If construction timelines slip, these companies will carry interest expense for longer before revenue begins. And anyone who has studied the realities of large-scale energy construction knows that timelines slip routinely. Gas turbines carry 18 to 24 month lead times. Large power transformers are backordered globally. Permitting, labour, and supply chain friction are not exceptions to the process. They are the process. The infrastructure providers are not immune to the physics of building things in the real world. Energisation dates that move from April to October, or from 2026 to 2027, are not just footnotes. They are periods during which capital is deployed, debt is accumulating, and revenue is zero.

I also want to be honest about what has changed in these businesses. Some of these companies were, until recently, bitcoin miners with simple balance sheets and no debt. The pivot to high-performance computing infrastructure has transformed them into capital-intensive developers with complex financing structures, multiple project-level entities, and in at least one case, a material financial restatement. The simplicity that I valued when we first invested in some of these names has given way to complexity. Complexity is not inherently bad, but it requires more vigilance. It requires us to read every filing more carefully, to track every construction milestone, and to be honest with ourselves about whether the original thesis still holds.

My current assessment is that it does, but with conditions. The contracted revenue base, the quality of the counterparties, the scarcity of the physical assets, and the near-term catalysts in the form of energisation dates and revenue commencement, all support continued ownership. But I am watching the balance sheets closely. If the debt grows without proportional progress on construction and revenue, I will reassess. Conviction is not the same as stubbornness.

Why a Basket, Not a Single Bet

There is one more dimension to our approach that I want to explain. We own a basket of infrastructure companies, not a single concentrated position. This is deliberate.

Each company in the basket has a different contract structure, a different set of counterparties, different sites at different stages of construction, and different risk profiles. One has Microsoft. Another has Amazon and Google. Their campuses are in different states, connected to different grids, with different energisation timelines. By owning the basket, we diversify away the idiosyncratic risk, the risk that a single site is delayed, a single permit is held up, or a single counterparty renegotiates terms, while maintaining full exposure to the structural trend.

This is the same logic a farmer uses when he plants multiple fields in different parts of the valley. If frost hits one field, the others survive. If a new irrigation channel opens near one field, that field benefits disproportionately. The portfolio is designed to capture the harvest from wherever it arrives first, while ensuring that no single disappointment can damage the whole.

A Tale of Two Farmers

I want to make the distinction between the utility trade and our infrastructure holdings as vivid as I can. The clearest way I know how is through a parable.

Imagine two farmers standing on opposite sides of the same valley, both looking at the same weather forecast. The forecast says rain is coming. Both believe it. Both want to profit from the growing season ahead.

The first farmer does not own his land. He leases it from a landlord under an agreement that either side can exit for a modest penalty. He does not have a well or an irrigation system. He is entirely dependent on the rain arriving when the forecast says it will. He has no signed buyer for his crop. He has a letter of intent from a grocer who has expressed interest but has committed nothing beyond a small deposit. Encouraged by the forecast, the first farmer borrows heavily to buy the best equipment, hires a full crew, and ploughs every acre he can reach. His entire plan depends on three things happening in sequence: the rain arriving on time, the crop growing on schedule, and the grocer honouring a commitment that was never truly binding.

The second farmer owns his land outright. The soil has been tested. There is a well on the property that produces water regardless of whether it rains. He has a signed contract with a large food distributor, one of the most creditworthy buyers in the country, who has prepaid a portion of the purchase price and committed to buying the harvest for the next ten years. The distributor cannot walk away without forfeiting significant capital. The second farmer has also taken on debt to build a new barn and improve his irrigation, more debt than he would normally carry. But the debt is secured against the land, the water, and the signed purchase agreement. It is not speculative. It is structural.

Now ask yourself: which farmer do you want to be?

Both are exposed to the same macro environment. Both believe in the same growing season. But the first farmer is farming the forecast. The second farmer is farming the land.

The utility investors I described earlier in this letter are, in many cases, farming the forecast. They are buying equities priced on load projections derived from contracts where both sides can walk away for pennies on the dollar. The projections may prove correct. The rain may come. But if it does not, or if it comes late, there is no well, no signed buyer, and no prepayment to fall back on.

Our infrastructure holdings are the second farmer. They own the land. They own the power capacity, the grid connections, the fibre, the acreage. These assets exist and hold value independent of any single contract. They have signed agreements with the largest and most creditworthy technology companies on earth, agreements that involve billions in prepayments, parent company guarantees, and 10 to 15 year terms. The buyers cannot walk away cheaply. The founders and executives own meaningful equity in these businesses, meaning their personal wealth is planted in the same soil as ours. And the harvest is not a distant hope. The energisation dates, the construction milestones, the commencement of contracted revenue, all sit within a 12 to 24 month window. The crop is nearly ready.

The market, meanwhile, still sees many of these companies through the lens of what they used to be: bitcoin miners, volatile, speculative, impossible to value. Most institutional investors have not yet updated their models to reflect the transformation. The analyst coverage is thin. The old label lingers. This is exactly the kind of quiet corner, the overlooked field at the edge of the valley, where we have historically found our best opportunities.

I will not pretend this investment is without compromise. The second farmer has taken on more debt than I would normally accept. The balance sheets have grown more complex. In at least one case, the financial reporting has required restatement. These are real concerns and I have explained earlier in this letter why I have accepted them and where I would draw the line. Honest investing means knowing which of your own rules you are bending, why you are bending them, and at what point the bend becomes a break.

The Discipline of Seeing Both Sides

The energy infrastructure buildout is the most important capital expenditure cycle in a generation. It will create enormous value and it will create enormous waste. The utility contracting analysis I shared earlier is a reminder that not every participant in this cycle will be rewarded, and that the structure of the commitments matters more than the size of the announcements.

We are not investing in announcements. We are investing in assets, in contracts, and in the physical scarcity of power-connected land. We are investing on the side of the table where the hyperscalers have committed real capital, not the side where they have signed non-binding letters.

But I do not want to overstate our certainty. The thesis depends on execution. It depends on construction timelines, on supply chains, on the continued willingness of hyperscalers to spend at this pace. The contracting fragility I described earlier in this letter is a sober reminder that the energy upcycle is not a straight line. There will be delays. There may be cancellations. The question is not whether disruptions will occur. It is whether our holdings are positioned to survive them and still deliver on the contracted revenue that underpins our valuation.

I believe they are. But I hold that belief with the humility of someone who knows that the market has a way of teaching lessons to anyone who confuses conviction with certainty.

The Patience to See Clearly

I want to close with something personal. When I was younger, I used to feel a pull toward the exciting trade, the hot sector, the narrative that everyone was talking about. It took me years to understand that the pull itself is the danger. The crowd’s enthusiasm is not a signal. It is noise dressed in conviction.

The greatest investors I have studied, Munger, Buffett, Klarman, Li Lu, share one quality above all others. It is not intelligence, though they have that. It is not access, though they have that too. It is the willingness to look foolish in the short run for the sake of being right in the long run. It is the discipline to say no when everyone else is saying yes, and to wait when everyone else is acting.

That is what I try to do with your capital. I try to separate the photograph from the house. I try to invest in things that are real and provable. And when I make a deliberate exception to our usual rules, as I have with the debt profiles in our infrastructure basket, I try to do so with eyes open, with clear reasoning, and with a willingness to reverse course if the facts change.

The energy buildout may well unfold as projected. I believe significant parts of it will. But I will not invest your capital on the basis of hope. I will invest it on the basis of contracts, assets, and principles that can survive the inevitable disappointments along the way.

Thank you for trusting me with that responsibility. I carry it with the same care today as the day you first placed it in my hands.

With warm regards and steadfast dedication,

Neel Khokhani

There is an old story about a man who discovers oil on his property. Neighbours arrive. Speculators arrive. Someone builds a derrick. Someone else builds a pipeline. Within months, an entire town springs up around the well. Everyone is talking about the oil. But almost nobody is asking the right question: who is buying it, and how much do they need?

The artificial intelligence infrastructure cycle has reached a similar moment. Headlines are dominated by the sheer scale of spending. Hyperscalers are pouring hundreds of billions into data centres, GPUs, and power capacity. The numbers are staggering and they are accelerating. But the conversation is stuck on the supply side. How much are they building? Is it too much? Will there be a bust?

These are natural questions. They are also the wrong ones. The right question is simpler and far more important: what is the demand backstop? What economic force, already in motion, guarantees that this infrastructure will be used?

The answer is not a technology trend. It is a labour market fact. There are approximately $3.7 trillion in annual wages paid to knowledge workers in the United States alone. These are the salaries of managers, analysts, accountants, lawyers, software engineers, marketers, and administrators whose work product is fundamentally digital. Their output lives on screens. Their tools are keyboards. Their deliverables are documents, spreadsheets, code, and decisions.

That $3.7 trillion is the demand backstop. It is the gravitational force that makes this infrastructure cycle structurally different from every previous one. And almost nobody is framing it this way.

The Wage Bill Is the Signal

Let me explain the arithmetic, because it is disarmingly simple.

The Bureau of Labour Statistics publishes detailed occupational data every year. When you isolate the categories of work most exposed to AI, the numbers are large and concrete. Management occupations alone account for roughly $1.55 trillion in annual wages across nearly 11 million workers. Business and financial operations add another $970 billion. Computer and mathematical roles contribute $607 billion. Legal professions add $175 billion. That subtotal is already $3.3 trillion, and it excludes the vast majority of administrative, design, and engineering roles that are also highly exposed.

The $3.7 trillion figure is not a forecast. It is not a projection based on optimistic adoption curves. It is the current annual cost of cognitive labour in the American economy. It exists today, right now, on the payrolls of every corporation in the country.

Now consider what happens when even a small fraction of that wage bill converts into compute demand. At 3 percent, you get $111 billion in annual AI infrastructure spending. At 5 percent, you get $185 billion. These are not aspirational numbers. They are arithmetic.

Think of it the way a farmer thinks about irrigation. You do not ask whether a river will flow. You measure the watershed. The watershed here is $3.7 trillion in wages already being paid to do work that AI can increasingly augment or perform. The river will flow because the water is already there. The only question is how quickly and through which channels.

Why the Bears Are Asking the Wrong Question

The sceptics have a framework, and it is not unreasonable on its surface. They look at the revenue currently generated by AI products and compare it to the capital being deployed. The gap is enormous. One prominent venture firm estimated that end users would need to generate over $600 billion in annual revenue to justify current infrastructure levels. By that measure, you would need thousands of products the size of ChatGPT just to break even.

This analysis is precise and wrong. It makes the classic mistake of measuring a new technology by the revenue model of the old one. It assumes the return on AI infrastructure must come from AI-native subscription revenue, from chatbot fees and API calls. That is like measuring the return on electricity by counting lightbulb sales while ignoring the factories, refrigerators, and cities that electricity would eventually power.

The real return on AI infrastructure will not come primarily from selling AI as a product. It will come from displacing, augmenting, and restructuring $3.7 trillion in existing labour costs. The demand signal is not new revenue. It is the redirection of old spending.

A law firm that pays $500 per hour for associate research does not need a new AI product to generate revenue. It needs an AI tool that does that research in four minutes instead of four hours. The saving is not recorded as AI revenue. It is recorded as higher margins, fewer hires, or greater throughput. The compute cost is buried inside the firm’s technology budget, invisible to anyone counting chatbot subscriptions.

This is why the bear case feels compelling but misses the structural point. It is measuring the wrong river.

The Supply Side Tells You Everything

If there is one qualitative data point that matters more than any financial model, it is this: not a single major hyperscaler reports excess capacity. Not one.

Google’s chief financial officer said publicly that the company exited 2025 with more demand than available capacity. Google Cloud’s backlog surged past $240 billion, a 55 percent increase in a single quarter. Amazon Web Services reported $200 billion in total backlog and added nearly 4 gigawatts of capacity in twelve months. Microsoft disclosed $80 billion in unfulfilled Azure AI orders, constrained not by demand but by the physical limits of power availability. Oracle’s remaining performance obligations hit $455 billion, up 359 percent year over year.

Read those numbers again. These are not speculative. They are contractual commitments, signed and waiting for infrastructure that does not yet exist. The binding constraint is not whether customers want the compute. It is whether the hyperscalers can build fast enough to deliver it.

Think of a port city during a trade boom. Ships are queuing outside the harbour. Warehouses are full. The cranes run day and night and still the backlog grows. You can debate whether the boom will last forever. But you cannot credibly argue that the port is overbuilt while ships are still waiting to dock.

The Capex Acceleration Is Unprecedented

The combined capital expenditure of the five largest hyperscalers, Amazon, Alphabet, Microsoft, Meta, and Oracle, has tripled in two years. In 2023, the big four spent approximately $155 billion. In 2024, that figure rose to $256 billion. For 2025, the combined total reached approximately $443 billion. Projections for 2026 range from $602 billion to $690 billion.

To put this in context, Goldman Sachs estimates that cumulative hyperscaler capital expenditure from 2025 through 2027 will exceed $1.15 trillion. That is more than double the $477 billion spent in the prior three years. And analyst estimates have proven too low for two consecutive years, with actual spending exceeding initial consensus by more than 30 percentage points each time.

Yet even at these levels, AI-related capital expenditure represents only about 0.8 percent of GDP. The telecom build-out of the late 1990s peaked at 1.0 to 1.2 percent of GDP. Reaching that level would require approximately $700 billion in a single year. We are not there yet.

I find this comparison instructive. The 1990s telecom build-out is widely regarded as having created excess capacity. But the infrastructure it left behind, the fibre-optic backbone of the internet, powered the next two decades of digital commerce. Even the “overbuild” turned out to be an underbuild when measured against actual long-term demand. The AI infrastructure cycle has the same structural character, except the demand backstop is more visible and more immediate than it was for broadband in 1998.

Power Is the Real Bottleneck

There is a natural governor on this cycle that prevents the kind of unconstrained overbuild that defined past technology booms. That governor is electricity.

The International Energy Agency estimates that global data centre electricity consumption reached 415 terawatt hours in 2024, roughly 1.5 percent of all electricity generated worldwide. By 2030, that figure is projected to reach 945 terawatt hours, equivalent to the total electricity consumption of Japan.

In the United States, data centre grid demand is expected to more than double from 62 gigawatts in 2025 to 134 gigawatts by 2030. The response from hyperscalers has been extraordinary. Microsoft signed a 20-year agreement to restart a unit at Three Mile Island. Amazon secured a 17-year power purchase agreement for nearly 2 gigawatts from the Susquehanna nuclear plant. Meta committed to a 20-year agreement for over a gigawatt of nuclear capacity.

When companies sign 20-year nuclear power agreements, they are not speculating. They are making irreversible commitments that only make sense if they believe demand will persist for decades. A farmer who builds an irrigation canal is not gambling on next season’s rain. He is investing in the permanent fertility of his land.

The Conversion Is Already Beginning

The theoretical case for AI exposure of knowledge work is well established. A study from OpenAI and the University of Pennsylvania found that roughly 19 percent of American workers have more than half their tasks exposed to AI. Applied to the total US wage bill, the highly exposed segment equates to approximately $3.4 to $4.0 trillion, consistent with the $3.7 trillion figure.

But theory is not what interests me. What interests me is what is already happening on the ground.

Anthropic published research in March 2026 that draws a crucial distinction between theoretical exposure and observed exposure. In computer and mathematical occupations, theoretical AI exposure is 94 percent. Observed exposure, meaning what is actually being automated today, is 33 percent. That gap between 94 and 33 is not a failure of the technology. It is the conversion opportunity. It is the distance between where we are and where the economics are pulling us.

The Dallas Federal Reserve added another layer in February 2026. Employment in the most AI-exposed sectors has already declined by 1 percent since late 2022. That decline falls disproportionately on workers under 25. It is not mass layoffs. It is the quiet cessation of hiring at the entry level, exactly the roles most susceptible to AI augmentation. For workers over 30, employment in these same sectors is growing. AI is not eliminating experienced judgment. It is substituting for the routine cognitive work that used to require junior staff.

This pattern is important because it reveals how the conversion actually works. It is not dramatic. It is not front-page news. It is a gradual reallocation of spending, one hiring decision at a time, one software subscription at a time, one workflow automation at a time. The $3.7 trillion does not convert overnight. It converts the way compound interest builds, slowly and then remarkably.

Where the Infrastructure Pure-Plays Sit

For those of you who follow our portfolio, you know we have held positions in companies that sit directly in the path of this demand. I want to discuss three infrastructure businesses and explain why I believe they occupy a position similar to owning toll roads at the beginning of a continental trade route.

IREN (formerly Iris Energy). We have discussed IREN in previous letters, and the thesis has only strengthened. The company tripled its operational data centre capacity to 810 megawatts during 2025 and grew revenue 168 percent year over year to $501 million. Total grid-connected power now stands at approximately 2.9 gigawatts, of which less than 20 percent is currently utilised. That is not a weakness. It is a reservoir. IREN has pivoted aggressively from bitcoin mining toward AI compute, secured status as an NVIDIA preferred partner, and is building a 75-megawatt liquid-cooled facility designed for next-generation Blackwell GPUs at 200 kilowatts per rack. The company is targeting $500 million in annualised AI cloud revenue by the first quarter of 2026. When you own the land and the power in a supply-constrained market, you do not need to predict the future with precision. You simply need to be standing in the right place when the demand arrives.

CoreWeave (CRWV). CoreWeave went public in March 2025 and has rapidly become the most visible pure-play AI infrastructure company in the market. Revenue reached $5.13 billion in 2025, up 168 percent year over year, with 2026 guidance of $12 to $13 billion. The revenue backlog stands at $66.8 billion, roughly four times where it began the year. Major contracts include OpenAI, Meta, and Microsoft. The company operates 850 megawatts of capacity with 3.1 gigawatts contracted, and was the first cloud provider to offer NVIDIA’s GB200 NVL72 systems. CoreWeave is not selling a dream. It is selling compute to the largest and most sophisticated AI buyers in the world, under long-term contracts, at industrial scale.

Cipher Digital (CIFR). Cipher is the cleanest example of a business transforming itself from bitcoin mining into AI infrastructure. The company formally exited mining in February 2026 and has committed entirely to high-performance computing. Total contracted HPC revenue stands at $9.3 billion with projected average annualised net operating income of $669 million over ten years. The anchor contract is a 15-year lease with Amazon Web Services worth approximately $5.5 billion covering 300 megawatts. An additional $3.8 billion contract with a major cloud provider covers another 300 megawatts at a facility in Texas. The total development pipeline spans 3.4 gigawatts across eight sites. In the space of a year, Cipher has transformed from a commodity mining operation into a contracted infrastructure platform with decade-long revenue visibility. The stock has reflected this, rising over 300 percent in six months.

The Philosophy Behind the Position

I want to step back from the data for a moment and explain why this thesis appeals to me philosophically, not just analytically.

Throughout my career, I have been drawn to situations where the underlying asset is real and the market is distracted by the wrong narrative. In the early days of Soar Aviation, no one believed you could crowdfund an aircraft. The narrative was about the novelty of the funding mechanism. But what mattered was the physical asset and the structural demand for pilot training. The narrative was noise. The asset was signal.

The AI infrastructure cycle presents the same dynamic at a vastly larger scale. The narrative is about whether AI companies can generate enough subscription revenue to justify the build-out. But that is the wrong frame. The signal is the $3.7 trillion wage bill that already exists, the physical infrastructure being built to serve it, and the contractual backlogs that prove demand is not speculative.

Our investment principles have not changed. We still look for what is asset-backed, what is cheap relative to the underlying value, what pays us to wait, what the crowd has neglected, and what is run by people who have their own capital at risk. The AI infrastructure companies we own meet these criteria in the same way that IREN met them when we first invested at $5 per share. The numbers are larger now, but the logic is identical.

There is a passage in Munger’s work that I return to often. He argues that the key to successful investing is not superior intelligence. It is a superior framework, applied with consistency, over a long period of time. The framework we use is designed to identify situations where the physical world is moving in one direction while the financial narrative is stuck debating something else entirely. The $3.7 trillion demand backstop is exactly that kind of situation.

What I Am Watching

I am not naive about the risks. Every infrastructure cycle carries the possibility of excess. The question is always whether the demand is real enough and durable enough to absorb the supply. Here, the evidence is overwhelmingly in favour.

But I am watching three things carefully. First, the pace of enterprise adoption. It is one thing for 78 percent of companies to say they use AI in some function. It is another for that usage to translate into sustained compute spending. The gap between experimentation and integration is where hype cycles die. Second, the power constraints. If permitting and grid expansion lag too far behind, even the strongest demand cannot be served, and the infrastructure companies we own will see their timelines elongated. Third, the health of the hyperscaler balance sheets. These companies are spending 90 percent of operating cash flow on capital expenditure. That is sustainable as long as revenue growth continues. If it stalls, the capex commitments become a burden rather than an asset.

None of these risks are disqualifying. They are the kinds of uncertainties that create opportunity for investors who are willing to do the work and hold the positions through volatility. The farmer does not abandon his fields because of a weather forecast. He checks his irrigation, inspects his fences, and trusts the soil.

Final Thoughts

The $3.7 trillion demand backstop is not a prediction. It is a measurement of the world as it exists today. Knowledge workers are expensive. Their work is increasingly digital. AI can perform meaningful portions of that work at a fraction of the cost. The infrastructure required to deliver that AI is being built at an unprecedented pace, and it is still not being built fast enough.

I do not know which AI model will be dominant in five years. I do not know which application layer company will capture the most value. Those are important questions, but they are not our questions. Our question is simpler: who owns the physical infrastructure that every model, every application, and every enterprise will need? That is where we want to be.

It is the same logic that has guided us from the beginning. Own what is real. Buy it when others are distracted. Be patient enough to let the world catch up to the arithmetic. And never confuse the noise of the crowd with the signal in the soil.

Thank you for your continued trust and partnership. I will keep investing with the same discipline, the same principles, and the same quiet conviction that the best outcomes come not from chasing what is exciting, but from owning what is essential.

With warm regards and steadfast dedication,

Neel Khokhani

There are moments in capital markets when the story becomes so loud that it drowns out the structure underneath it.

Artificial intelligence. Hyperscale demand. GPU shortages. Sovereign compute. Power constraints.

The commentary is endless.

What interests me is quieter.

Who is paying for the infrastructure?

On what terms?

Who carries the risk?

Who carries the clock?

And most importantly, who owns the residual value when the noise fades?

In Iris Energy, the most important development is not the headline growth narrative. It is the architecture of the capital stack.

Let us look at it carefully.

Microsoft has prepaid approximately $1.94 billion at a zero percent cost. In parallel, roughly $3.6 billion of debt has been secured directly against the GPUs at a cost below 6 percent. Combined, that represents roughly $5.5 to $5.6 billion of capital raised at a blended cost below 4 percent, covering approximately 95 percent of the roughly $5.8 billion GPU acquisition cost.

Pause there.

The company is financing one of the largest GPU fleets in the public markets with near zero equity dilution and at a cost of capital that many industrial companies would envy.

That is not speculation. That is structure.

And structure determines outcomes far more than sentiment does.

The Difference Between Story and Structure

It is easy to tell a compelling story about AI demand. It is much harder to build a capital structure that survives volatility.

When I evaluate an investment, I ask a simple question. If the narrative disappears tomorrow, does the balance sheet still make sense?

In IREN’s case, we are not looking at unsecured venture-style capital chasing a dream. We are looking at a prepaid offtake agreement from one of the strongest counterparties in the world, coupled with asset-level debt secured against GPUs that have tangible, monetizable value.

The distinction matters.

A prepayment from Microsoft at zero percent is not merely cheap funding. It is validation of demand. It converts counterparty risk into structured capital. It removes equity from the first layer of risk.

Debt secured by GPUs at sub 6 percent is not reckless leverage. It is asset-backed financing against equipment with established market value, resale markets, and productivity tied to long-term compute contracts.

This is not aspirational funding.

It is engineered funding.

Funding at Scale Without Losing Control

Consider the math.

Approximately $5.8 billion of GPUs.

Approximately $5.5 to $5.6 billion funded through prepayment and secured debt.

That leaves minimal equity required at the GPU layer.

Now expand the lens.

When Horizon 1 through Horizon 4 are live and de-risked, the data center assets themselves become financeable. If the company can refinance approximately 80 to 85 percent loan-to-cost on a roughly $3 billion data center build, that represents roughly $2.4 to $2.5 billion of new debt capacity.

That debt is not incremental risk layered onto speculation.

It is refinancing against stabilized, cash-flowing infrastructure.

And the capital released can be recycled into the next project.

• Fund.

• Deliver.

• De-risk.

• Refinance.

• Recycle.

That is the playbook.

The result is powerful.

On the full roughly $8.8 billion Microsoft stack, combining GPUs and data centers, equity required could be in the range of approximately $800 million.

Under 10 percent of total capex.

If that math holds, then shareholders are participating in an $8.8 billion infrastructure build with single-digit equity contribution.

That is operating leverage, but disciplined operating leverage.

Why This Matters More Than the Share Price

It is tempting to focus on where the stock trades.

That is not the core question.

The core question is whether intrinsic value per share can grow faster than dilution.

This capital stack design suggests that growth is being financed primarily through structured, low-cost capital rather than repeated equity issuance.

That changes the compounding equation.

If a company must issue 30 or 40 percent new equity every time it grows, the shareholder participates in growth but not in ownership stability.

If a company can fund 90 percent of its capex externally at attractive rates, equity becomes a minority contributor to scale but retains majority participation in upside.

This is how infrastructure businesses compound.

• Build at scale.

• Stabilize.

• Refinance.

• Extract equity.

• Rebuild.

It is the same cycle used in commercial real estate, renewable energy, and transport infrastructure.

But here, the underlying asset is compute.

Asset-Backed in a Digital World

One of our core principles has always been asset-backed reality.

In a digital economy, that phrase sounds almost antiquated. Yet the physics of capital still apply.

A data center is concrete, steel, transformers, cooling systems, fiber routes, and long-term power agreements.

A GPU cluster is physical silicon with resale value and global demand.

Power contracts are contractual rights.

These are not abstract concepts.

They are productive assets with economic life.

When we first invested in IREN years ago, the market framed it narrowly as a bitcoin miner. The underlying assets were undervalued because the narrative was constrained.

Today the narrative has expanded to high performance compute and AI. The assets, however, are the same species. They are energy-intensive compute infrastructure.

The difference is recognition.

Cheap Capital as a Strategic Weapon

Capital cost is a competitive advantage.

If you can borrow at a blended sub 4 percent cost while competitors raise equity at high implied cost or borrow at double-digit rates, you win by arithmetic.

Over time, cost of capital compounds just like revenue.

Every percentage point difference in funding cost translates into retained earnings.

The Microsoft prepayment is particularly elegant. Zero percent funding tied to revenue-generating compute. It reduces weighted average cost of capital while increasing capacity.

It also aligns incentives.

Microsoft needs capacity.

IREN needs capital.

Structure aligns both.

Near-Term Catalyst and De-Risking

Another of our principles is the presence of a near-term catalyst within one to two years.

In this case, the catalysts are tangible.

Horizon phases going live.

Cash flows stabilizing.

Refinancing at asset level.

Each completed phase moves risk from development to operation.

Markets price development at a discount. They price stabilized cash flow differently.

The movement from build to operate is the bridge that allows value to surface.

Market Neglect and Misunderstanding

Despite the scale, there remains skepticism.

Some view IREN through its bitcoin history.

Some question whether hyperscale demand persists.

Some focus only on volatility.

That skepticism is not an obstacle. It is a condition for opportunity.

If the market were fully convinced, the equity would likely trade at a materially higher multiple of book and forward cash flow.

We do not need universal agreement.

We need asymmetry.

• Insider Alignment and Ownership

• High insider ownership matters in capital-intensive businesses.

• Management must think in decades, not quarters.

• Capital recycling requires discipline.

• Asset-level financing requires credibility with lenders.

Long-term value creation requires resisting the urge to chase fashionable narratives.

When management’s net worth is meaningfully tied to equity, incentives align with ours.

The Role of Debt

We prefer no debt where possible.

But when debt is directly secured against productive assets, at modest cost, and tied to contracted cash flows, it becomes a tool rather than a threat.

The distinction is crucial.

Debt tied to speculation is dangerous.

Debt tied to stabilized infrastructure can be rational.

In this case, GPU-secured debt and potential data center refinancing are linked to revenue-generating assets.

The clock exists, but it is matched with cash flow.

Paid to Wait

As phases come online, the company transitions from build mode to yield mode.

Cash generation matters.

Net income matters.

Free cash flow matters.

We do not invest in perpetual promises.

The thesis requires that infrastructure translates into earnings.

So far, the structure suggests it can.

Noise Versus Signal

There will be quarters where revenue surprises. There will be headlines about power prices, GPU supply, macro liquidity.

Those are fluctuations.

The signal is the capital stack.

If the company can consistently fund 90 percent of growth externally at attractive rates and recycle equity through refinancing, intrinsic value per share can compound meaningfully.

If that discipline breaks, we reassess.

Investing Is Engineering

I often think of investing not as prediction, but as engineering.

An engineer does not ask whether a bridge will feel strong. He calculates load-bearing capacity, stress tolerance, and redundancy.

In IREN’s case, the engineering question is simple.

Can this capital structure support the scale being built?

At present, the answer appears increasingly affirmative.

• Minimal dilution.

• Low blended cost.

• Asset-backed debt.

• Counterparty prepayment.

• Refinance optionality.

• Recycling loop.

• Fund.

• Deliver.

• De-risk.

• Refinance.

• Recycle.

That loop, repeated carefully, is how infrastructure becomes enduring equity value.

Closing Thoughts

We do not invest because a theme is fashionable.

We invest when structure, valuation, assets, and catalysts align.

In IREN, the most important story is not the excitement around AI.

It is that roughly $8.8 billion of infrastructure can potentially be built with under 10 percent equity.

That is leverage, but disciplined leverage.

That is growth, but engineered growth.

And if executed properly, that is compounding without chronic dilution.

As always, we will monitor execution rather than headlines.

Capital deserves patience, but it also deserves scrutiny.

The work is quiet.

The math is clear.

The structure matters more than the noise.

With discipline and conviction,

Neel Khokhani

Dear Partners

When I was growing up in Vasai, north of Mumbai, there was a cricket ground a short walk from where we lived. It was not a proper ground. No pavilion. No grass worth mentioning. Just a strip of rolled earth that passed for a pitch and an outfield that swallowed the ball if it didn’t quite reach the middle of the bat.

In the early hours, before anyone had played, the pitch looked forgiving. The surface was flat. The ball came on nicely. Shot-making felt easy. By mid-afternoon, after hours of play under the sun, the same pitch slowed dramatically. The ball stopped carrying. Mistimed shots died in the dust. Batsmen who went after it early often found themselves walking back sooner than expected.

Nothing sudden had happened. The ground had not changed its nature. What changed was whether the player had the patience to let conditions reveal themselves.

Markets work the same way. Most investments look easy at the beginning. Prices cooperate. Narratives make sense. Confidence comes cheaply. It is only later, when conditions deteriorate and the score stops moving, that judgment is tested.

Gautam Baid, in The Joys of Compounding, describes long-term success not as a function of repeated brilliance, but as the result of minimizing error and interference over time. Compounding, in that sense, is not something you chase. It is what remains when you stop getting in your own way.

That idea has shaped how I think about capital allocation more than any forecast. It moves the focus away from prediction and toward endurance. Away from excitement and toward survivability.

This letter is about what that looks like in practice.

This looked straightforward, until it wasn’t

We began building our position in Iris Energy (IREN) in 2022 at approximately $7 per share. The thesis was simple. The company owned power and data infrastructure that could not be easily replicated. It carried no debt. And the market was valuing those assets as if they were permanently impaired.

There was nothing clever about the setup. It was arithmetic and structure.

Then the game changed.

The stock did not drift lower politely. It fell sharply. As sentiment around bitcoin mining deteriorated, Iris Energy became a stand-in for everything investors wanted to avoid. Price declines fed narratives. Narratives accelerated selling. Each reinforced the other.

At around $4 per share, we continued to add. Not because the price had fallen, but because nothing important had broken. The balance sheet remained unlevered. The assets remained intact. The company still had time.

Then the stock fell again.

When the scoreboard stops moving

At its lowest point, Iris Energy (IREN) traded below one dollar. At approximately 98 cents, we acquired an additional 100,000 shares at one point we had approximately 2.8m shares of IREN.

By then, the challenge had stopped being analytical. It was psychological.

When a stock trades at that level, price starts to feel like information. You begin to wonder whether the market knows something you don’t. Whether your original assumptions were flawed. Whether discipline has quietly turned into stubbornness.

This is where capital allocation stops being about intelligence and becomes about judgment.

We forced ourselves to return to the same questions we had asked at $7 and at $4, even though doing so offered no emotional comfort.

What assets exist today?
What liabilities exist today?
What conditions would permanently impair those assets?
What conditions would force dilution or liquidation?
What has changed, structurally, since our last review?

The answers were not reassuring, but they were consistent. Price had collapsed. Structure had not.

There was no certainty of recovery. Survival was probable, not guaranteed. But at those prices, the market was assuming near-total failure. The only question that mattered was whether the business could stay standing long enough for perception to change.

Just don’t get out

Buying at 98 cents was not an act of confidence in the usual sense. It was an act of restraint.

On that Vasai cricket ground, the players who lasted were not the ones trying to force boundaries once the pitch slowed. They were the ones who focused on staying in. Defending well. Letting time work.

That is what this moment required. Not aggression. Not optimism. Just refusing to throw the wicket away when conditions were at their worst.

This is the part of compounding that rarely appears in investor letters. The long stretch where nothing validates you. No headlines. No momentum. No sense that patience will be rewarded. Only the discipline to keep returning to first principles.

Same ground, different conditions

When Iris Energy later re-rated, it did not feel like vindication. It felt like recognition finally catching up to what had already been true.

The infrastructure did not suddenly improve. The balance sheet did not suddenly strengthen. The optionality did not appear overnight.

What changed was perception.

From that lowest point, with Iris Energy later touching an intraday high near $50, the position appreciated by over 5,000 percent from the 98-cent level. That number is not important because it is large. It is important because it captures the distance between a price that assumed near-total failure and one that began to acknowledge survivability.

Moves like that are not created by timing. They come from enduring periods where ownership feels unjustified and isolated, and where interference is most tempting.

By the time outcomes look obvious, the work that mattered has long since been done.

Nobody wants to bat here

Core Scientific tested the same discipline in a different way.

Where Iris Energy challenged patience through price collapse, Core Scientific challenged judgment through association. Bankruptcy is a word that shuts down analysis. Many investors treat it as a permanent stain rather than a restructuring of liabilities.

This is often a mistake.

Bankruptcy reorganizes the balance sheet. It does not erase physical reality. The relevant question is not what failed, but what remains.

After restructuring, Core Scientific retained substantial infrastructure, power access, operational capability, and scale. What changed was the capital structure. Debt was reduced. Equity was reset. The business emerged lighter.

The market remained anchored to the past. Narrative dominated economics. Few were willing to step onto that part of the ground.

Once again, the opportunity was not in flair. It was in defense. In being willing to bat where others refused to stand.

Most of the game happens between the highlights

One of the most persistent misconceptions in investing is that progress requires constant action. Markets reward responsiveness and punish stillness, even when stillness is the rational choice.

In reality, most of the value in a portfolio is created during periods when very little appears to be happening. Long stretches where the score barely moves. Short bursts where recognition finally arrives.

The danger is not missing the highlights. It is getting out in the quiet overs because patience wears thin.

You don’t win by swinging at everything

Compounding only works if the base survives. Anything that turns temporary adversity into permanent loss breaks the chain.

This is why we avoid fragile structures. Why we are cautious with leverage. Why we are skeptical of strategies that require perfect timing.

Even strong businesses can be destroyed by being early if they cannot endure disappointment. Time only rewards what can stay in the game.

Leave the ground with wickets in hand

Managing capital is not about stimulation. It is about stewardship.

Our responsibility is not to maximize short-term results. It is to protect capital and allow it to compound across cycles.

That requires accepting boredom. Enduring misunderstanding. Resisting the urge to act simply to feel involved.

On that cricket ground in Vasai, the players who walked off at the end of the day were rarely the most spectacular. They were the ones who understood conditions, respected risk, and preserved optionality.

Investing works the same way.

Thank you for trusting me with your capital. I will continue to manage it with discipline, humility, and respect for the long game.

With steady conviction,
Neel

Over the years I’ve noticed that most investment mistakes don’t come from misunderstanding businesses. They come from misunderstanding what matters in a business. People spend an enormous amount of time debating products, markets, and narratives, and far less time thinking about the things that are difficult to change.

That imbalance usually shows up in the price.

Technology is a good example. It looks dynamic, fast-moving, and endlessly exciting. But much of what people call technology today is surprisingly modular. Compute can be purchased. Servers can be financed. Software can be copied. Over time, those things tend to earn something close to the cost of capital.

What behaves very differently are the constraints underneath. Power is local. Permits are slow. Grids are political. Distance still matters. Latency obeys physics. These things don’t care about quarterly earnings calls.

Two of our holdings illustrate this better than most, even though they’re rarely discussed in the same breath.

One is Iris Energy (IREN). The other is Duos Technologies (DUOT). On the surface, they look like very different animals. One is associated with bitcoin and high-performance computing. The other with machine vision and edge infrastructure. But that surface view misses the point.

What IREN really owns is not mining rigs or AI servers. It owns long-duration access to cheap, renewable power, secured at scale, in places where that power would otherwise be underutilised. Bitcoin mining was simply the first sensible way to monetise that position. High-performance computing is the second. Neither is the foundation.

That’s why I’ve never thought of Iris’s evolution as a pivot. Nothing fundamental changed. The company didn’t suddenly acquire a new advantage. The world simply began to value something Iris already had.

The balance sheet mattered here in a quiet way. No debt meant no clock. And no clock meant time. When you remove time pressure, volatility becomes a nuisance instead of a threat. That’s an underappreciated form of optionality.

Duos operates on a similar logic, just in a different dimension. Most people encounter Duos through its vision systems and AI software. That’s understandable, because software is easy to talk about. But again, the deeper advantage isn’t the application. It’s where the application runs.

Latency is a constraint every bit as real as power. In a lot of industrial and safety-critical settings, you can’t ship data halfway across the country and wait for an answer. Milliseconds matter. Sometimes failure simply isn’t an option. Hyperscale data centres are very good at many things, but they’re not especially good at being close.

Duos is quietly building edge infrastructure in places where proximity is essential and alternatives are poor. The vision software makes sense because the edge exists, not the other way around. The software is how the position is monetised. The position itself is physical.

In both cases, what’s really happening is simple. These businesses own bottlenecks that others have no choice but to rent. Most competitors pay market prices for power or proximity. IREN and Duos internalise those costs at structural discounts. Over time, that difference compounds, even if it’s not obvious at the beginning.

Markets tend to miss this because they’re very good at pricing what’s visible today and not very good at pricing what will be necessary tomorrow. Capacity that isn’t fully utilised looks inefficient. Capital intensity looks risky. Optionality that hasn’t shown up in revenue yet is easy to ignore.

That’s usually when the opportunity exists.

A skeptic could argue that scale will eventually solve these problems. Maybe. But scale doesn’t repeal physics, and it doesn’t make permitting faster. Power grids remain local. Distance still creates latency. Centralisation fixes some issues and creates others.

Another skeptic might say this all depends on demand continuing to grow. That’s fair. But neither of these businesses depends on a single rosy future. They’ve already earned money in less accommodating environments. Bitcoin mining wasn’t hypothetical. Rail inspection isn’t speculative. These aren’t stories waiting for belief.

I often think this style of investing resembles farming more than forecasting. The land matters more than the crop. Crops change. Seasons change. Prices change. Good land stays good land. Most people spend their time arguing about yields. The real work happens earlier, quietly, in choosing where to stand.

By the time the field looks attractive, it’s usually already owned.

I don’t think of IREN or Duos as technology bets. I think of them as ownership of things that don’t move quickly in a world that increasingly wants everything at once. Power isn’t getting less constrained. Latency isn’t becoming irrelevant. As those pressures build, the quiet owners of bottlenecks tend to do fine without needing to explain themselves too often.

That’s the kind of position I’m comfortable holding.

With patience and discipline,
Neel Khokhani

Dear Partners,

In investing, as in life, the best outcomes accrue to those who can hold a calm perspective. The quote above captures something I have found to be deeply practical. Durable outcomes cannot be rushed. This past year has been strong for our family portfolio, but this letter is not a victory lap. It is a reflection on how I think, how I invest, and what it means to carry the responsibility of being your fiduciary.

I often return to the same image. A farmer who understands the land does not confuse motion with progress. He does not pace the field each morning demanding proof that the seeds are working. He tends the soil, protects what is fragile, and respects the seasons. In the same way, I try to protect our capital from the modern habit of constant reaction. Markets are full of noise. The real work is quieter.

Charlie Munger put it plainly: “The big money is not in the buying or the selling, but in the waiting.”
That line sounds simple, almost too simple, until you live through the years when waiting feels uncomfortable. I have learned that most investment mistakes are not caused by ignorance. They are caused by impatience.

The Farmer’s Mindset in Investing

In our view, successful investing resembles sustainable farming. We prepare the soil by studying businesses until we understand what can break them and what can strengthen them. We plant capital only when the ground is fertile, meaning the assets are real, the balance sheet is sound, and the valuation gives us breathing room. Then we allow time to do what time does best.

A farmer cannot tug on seedlings to make them grow faster. If he tries, he destroys the roots. Likewise, investors can destroy compounding by interfering too often. Most of the value creation happens quietly, below the surface, while nothing looks impressive on the outside.

That is why I pay so little attention to daily commentary. It is often entertaining, sometimes intelligent, but rarely useful. It can make you feel informed while quietly eroding judgment.

Seeds of Conviction: Three Case Studies

To make this philosophy tangible, I want to share three examples from recent years. Not as a scoreboard, but as illustrations of how decisions are actually made, and why patience can look irrational before it looks wise.

Iris Energy (IREN): Harvesting a Margin of Safety

Our investment in Iris Energy began when the stock had fallen toward roughly $5. That entry point did not come from optimism. It came from arithmetic.

At the time, the market was valuing the entire business at a level that did not reflect the underlying asset base. In simple terms, the sum of its infrastructure and capacity appeared worth materially more than the market capitalisation. That is what margin of safety looks like in the real world. It is not a feeling. It is a cushion.

The second pillar was balance sheet strength. Iris had no debt. That matters more than many people realise. Debt introduces a clock. A clock turns volatility into danger. Without debt, the company had time, and time is the rarest form of optionality.

When we started building the position in 2022, the stock was already down roughly 70 percent from the IPO, trading around $8. It continued falling after that. That period was not comfortable. But the foundation remained intact, which is why the position could be held with conviction rather than hope.

Later, the company began pivoting from being seen purely as a bitcoin miner toward high performance computing. I do not view that pivot as the creation of value. I view it as the recognition of value that already existed in the infrastructure. When markets finally reframe a business, the price can move quickly. The real work happens long before the reframe.

Today, IREN trades around $39. I continue to hold it. Not because I am emotionally attached, but because the original reasons for ownership were grounded in assets, balance sheet resilience, and a pathway for the market to recognise what was already there.

Core Scientific (CORZ): Resilience Unearthed from Ruin

Core Scientific is a different kind of lesson. It is about mispriced survival.

We invested after the company emerged from Chapter 11 in 2024, around $4 per share. Many investors could not look beyond the word bankruptcy. But bankruptcy often restructures liabilities, not assets. It can remove the burden that prevented the underlying engine from operating.

What I cared about was what remained standing after the storm. Infrastructure. Power access. Operational capability. A reset capital structure. In other words, survivability.

Today the stock trades around $15, and we continue to hold it. After a stellar year, it is tempting to treat gains as proof of being right. I try to treat them as proof that the process is working, and as a reminder to remain humble. The market can reward you and still teach you nothing if you let it.

Fluence Energy (FLNC): Knowing When the Harvest Is Ready

Fluence was a lesson in catalysts and disciplined exits.

We initiated the position around $8 when sentiment toward the sector was barren. There was no excitement, which is often the point. The business had a credible path to improved economics and a near term catalyst that could make that improvement visible.

As the stock moved toward $20, the valuation began reflecting what the market had previously ignored. At that point, the margin of safety narrowed. The crop was ripe. We sold.

A farmer who refuses to harvest because he likes his crops is not wise. He is sentimental. Capital deserves discipline, not attachment.

The Principles That Guide Capital at the Family Office

At the family office level, the job is not to win arguments or impress anyone with activity. The job is to compound capital while avoiding permanent mistakes. That requires filters that last longer than headlines.

Here are the principles I return to repeatedly:

Asset-backed
I want something real underneath the price. Like farmland that can be worked, even if the season is difficult.

Cheap valuation
I want to buy with room for error. Like buying a sturdy tool at a fair price, not a shiny one at an inflated price.

High yield when available, paid to wait
I love being paid while time does its work. Like renting out a property while the neighbourhood improves.

Market neglect
I often prefer what is ignored. The best orchards are not always beside the main road.

High insider ownership
I want management to feel the same outcomes as owners. Like a ship captain who sleeps on the same deck he steers.

No debt, or very modest debt
I want time on our side. Debt shortens time. Time is where compounding lives.

Net income matters
I want reality, not promises. A business should be able to feed itself, not constantly ask for more seed.

A catalyst within one to two years
I want a reason value can surface. Not a dream, a plausible mechanism. Like a bridge opening that finally connects an overlooked town to the rest of the map.

These are not rules for perfection. They are rules for survival. Munger said it best when he spoke about trying to be consistently not stupid. That is a deeper philosophy than it sounds. Avoiding big errors, over and over again, is how you earn the right to compound over decades.

Anchored in Conviction, Looking Ahead

After a strong year, many investors feel pressure to change the formula. I do not. The temptation after success is often the same as the temptation after fear. It is the urge to do something.

The portfolio remains anchored in conviction holdings like IREN and CORZ. These are not positions we hold because they have gone up. They are positions we hold because the original underwriting still stands, and because the businesses continue to meet our standards. I view that continuity as the true source of compounding.

There will be years when results are less pleasant. That is normal. The market has seasons. The only thing I control is process, temperament, and prudence. I will keep returning to those.

Stewardship and Peace of Mind

Being a fiduciary is not simply managing money. It is managing responsibility. It is holding a long horizon on behalf of others, even when the world is shouting about the next hour.

I want you to understand me as an investor in a simple way. I do not chase excitement. I do not confuse movement with progress. I protect capital first, and I pursue returns only through principles that can survive a cycle.

If I do that well, then returns become the byproduct, not the obsession.

Thank you for the trust you place in this approach. I do not take it lightly. I will continue to invest with discipline, humility, and a commitment to protecting both your capital and your peace of mind.

With warm regards and steadfast dedication,

Neel

When I was a child growing up in a small village in India, there was a local shop that used one of those old-fashioned weighing scales with metal weights. From a distance the scale always looked balanced. The shopkeeper moved with such quiet confidence that nobody questioned it. The vegetables went on one side, the weight on the other and the beam settled in that familiar, reassuring way.

But if you stood close, you could see something else.
The needle never quite aligned with the proper mark. The balance looked convincing, but the measurement was slightly inflated. Most people never noticed because the movement of the scale felt normal and the shopkeeper looked like he knew exactly what he was doing.

The market today feels similar to that scale.
On the surface everything looks steady. Credit spreads are calm. Companies sound confident. Volatility barely moves. From a distance you could believe the system is healthy and in good order.

But when you check the reading that matters, which is the price of the market relative to the actual output of the economy, the number is quietly stretched. The Buffett Indicator captures that. The surface is calm, but the weight is off.

This is not a forecast of panic. It is simply recognition that you can overpay without realising it, especially when everything looks normal.

A Simple Way to Think About Valuation

I often return to a basic picture.
A farm, a harvest and a buyer.

A farm produces a fixed amount of wheat each year. You can admire the land and believe in its potential, but the crop is the crop. If you pay a sensible multiple of the harvest, you are likely to earn a sensible return. If you convince yourself the farm is extraordinary and pay fifty or one hundred times the harvest, nothing about the farm changes. Only your return does.

This is exactly how valuation works at the market level.
When price runs too far ahead of output, future returns narrow.

Where the Indicator Sits Today

It helps to attach real numbers to this rather than keep it abstract.
As of late 2025, the total value of the United States stock market sits at roughly 220 to 223 percent of GDP. In simple terms, the market is priced at more than twice the size of the economy that supports it.

For most of modern market history, that ratio lived closer to 80 to 120 percent.
Those levels tended to coincide with long-term returns that were fair and grounded in what the economy could reasonably deliver.
Once the ratio moves above one hundred sixty percent, the pattern across decades is consistent. Forward returns weaken, not because anything breaks, but because investors begin too far above what the underlying economy can justify.

This metric does not tell us when conditions will shift.
It tells us what future returns are likely to look like when the entry point is this high.
A market that trades at more than twice the size of its economic output is one that has already consumed a meaningful portion of tomorrow’s returns today.

Why Strong Current Data Does Not Cancel This

Two talking points are often used to justify today’s valuations.

Credit metrics are improving.
Corporate guidance is rising.

Both are true, but neither relates to valuation.

Credit is a backward-looking marker. It tells you the recent past was calm. It does not tell you what is priced into the future. A quiet fire station does not mean the town is safe.

Guidance is similar. Companies sound most confident when conditions are strong and margins are wide. Their confidence reflects the moment, not the valuation.

Good conditions now do not guarantee good returns from this price.

A Practical Example of Valuation Leaving Reality

Every cycle produces a company that investors decide does not need to follow basic arithmetic.
This cycle has Palantir.

Some argue that one hundred times sales is reasonable because the business is unique.

One hundred times sales means something simple.
If Palantir handed over every dollar of revenue for the next one hundred years, without any expenses or reinvestment, investors would only break even in 2125.

A more tangible example makes it clear.

Imagine a town with a bakery known for exceptional bread. The baker works hard and people queue every morning. The bakery dominates locally. Now imagine the town decides that buying the bakery for one hundred years of revenue is fair. Then imagine the assistants behind the counter, not even the baker, being awarded equity packages worth more than the bakery will reasonably produce for years.

In any normal setting, you would call that mispriced.

That is the situation with Palantir.
Mid-level employees now sit on RSU packages exceeding twenty-five million dollars. The business is not producing that value. The valuation environment is.

The lesson is not about Palantir.
It is about how easily narratives override numbers when the surface feels calm.

A Real-Time Reminder: Figma and Adobe

Figma’s IPO offered a useful example of how valuation can overpower fundamentals, even when the underlying product is genuinely strong.

In my view, Figma is a far superior product to Adobe’s comparable tools.
It is modern, intuitive and the preferred choice for many designers. The collaboration layer creates genuine network effects that Adobe has never truly managed to replicate. If you gave both companies a blank sheet of paper and asked them to build a design platform from scratch, most people would likely bet on Figma.

Yet the market behaved exactly as it always does when a great company meets the wrong valuation.

Figma fell nearly sixty-eight percent within months of listing.
Adobe fell roughly five percent.

The outcome had nothing to do with product quality or competitive strength.
It had everything to do with valuation.

Even a company with a better product and real network effects cannot outrun the gravity of an overstretched multiple. When a price assumes too much of the future, any normal adjustment in expectations can overwhelm the fundamentals.

Figma remained a strong product.
Its valuation did not.

A good business can still be a poor investment at the wrong price.

What History Repeats When Starting Points Are High

Across cycles, the pattern is consistent.

When valuations begin at elevated levels:

• future returns shrink

• downside asymmetry increases

• sensitivity to rates rises

• small disappointments create outsized reactions

• multiple compression happens even without recession

Nothing dramatic needs to occur.
The problem is the entry point.

Valuation behaves like gravity.
Slow, consistent and eventually decisive.

How I Position When Markets Are Calm but Prices Are Not

When valuations stretch and nothing feels urgent, I simplify the process.

1. Would I buy this today at this size
If the answer is not a confident yes, I trim or exit.

2. Anything above ten times sales is treated as a trade, not a core
Either the evidence is exceptional or the position stays small.

3. Cash equals optionality
If I would not feel ready to deploy capital after a sharp decline, liquidity increases.

4. Familiarity is not a thesis
If I cannot explain in one sentence why a position deserves to remain, it leaves the portfolio.

5. The regret test
If the market fell tomorrow, which positions would I wish I had reduced. Those are reduced now.

This keeps the portfolio aligned with reality rather than mood.

Closing Thoughts

The Buffett Indicator is not predicting a crisis.
It is signalling that the future has already been priced more expensively than usual.

Strong credit, firm guidance and quiet volatility do not override valuation gravity.
They simply make it easier to overlook.

When you pay too much for the world as it stands, the world does not need to deteriorate for returns to weaken.
It only needs to continue.

One personal note.
Recently the market has felt uneventful to me. Boring, even. Earlier in my career I might have taken that as a reason to search for something new. I see it differently now.
Boredom usually means I am not forcing trades, not chasing stories and not inventing activity for its own sake.
It is often the first indication that discipline is doing its job.

My focus remains the same.
Stay selective.
Stay clear.
Stay disciplined.

Sincerely,
Neel

Some artworks reveal themselves with certainty. Others arrive slowly, like something half-remembered returning from the edges of thought. This triptych by Sabine Moritz belongs to the second kind. It does not present an image. It does not offer a clear narrative. Instead it creates an atmosphere shaped by the inner world rather than the outer one.

Moritz paints the places where emotion collects. She allows sensations to drift across the canvas in slow movements. Strokes appear and soften. Forms emerge just long enough to dissolve. The work feels alive with hesitation and intention, as if the painting is thinking its way into being.

Nothing is fixed. Everything is in motion.

This is not abstraction used to avoid meaning. It is abstraction used to reveal the quiet complexity of how memory, intuition, and perception actually behave. The painting feels like a landscape of thought rather than an object.

The Way a Thought Learns to Breathe

“Let everything happen to you. Beauty and terror. Just keep going. No feeling is final.”
Rilke

Moritz works with a kind of patience rarely seen. Her layers never arrive cleanly. They arrive as reconsiderations. Each mark feels like a possibility rather than a verdict. You can sense her listening to the painting as much as she is creating it.

Colours hold emotional temperatures instead of definitions.
Greys that invite reflection.
Blues that feel without depth.
Warm undertones that rise like memories returning after years of silence.

The surface reads like a slow exhale.

Nothing forces itself forward. Nothing insists. The painting does not ask for interpretation. It asks for attention. You begin to understand that it is not trying to show you something. It is inviting you into something.

It moves the way interior life moves: gradually, honestly, without haste.

How Three Moments Become One Truth

“No man ever steps in the same river twice.”
Heraclitus

A triptych offers a different way of seeing. One panel is a moment. Three panels are a movement.

Each section of this work feels like an emotional variation. A shift in tone. A slightly altered light. A change in the internal register. The gaps between them feel like pauses where the mind adjusts and listens again.

You move across the panels slowly. You feel the subtle transitions. You recognise how meaning forms in evolution rather than in singularity. The painting does not guide you. It lets you find your own rhythm with it.

It becomes a practice rather than a picture.
A meditation rather than a presentation.

Moritz allows the viewer to discover that truth accumulates, not in a moment of clarity, but in a series of gentle recognitions.

The Red Earth That Never Left Me

Before I ever understood art in this way, I learned something similar through a childhood moment that has stayed with me.

As a child in India, our front yard was a patch of uneven red earth. It never sat flat. Each morning we would water the ground to settle the dust, and the instant water touched soil, that deep, grounding scent of wet red earth drifted upward. A smell that carried both calm and anticipation. To this day I remember it with complete precision.

There was always a pause after the watering. A quiet moment before heat and noise arrived. The dust settled. The air felt suspended. The world seemed to wait for its own arrival.

I stood there barefoot, watching the earth darken in color, sensing the beginning of the day. That moment held a feeling I did not have words for then. It was the feeling of possibility before any narrative had formed. A quiet curiosity about how the day might unfold.

Afternoons were spent playing cricket on that same ground. Cricket on uneven earth taught me everything before I understood anything. It taught me how to read the bounce. How to judge the wind. How to adjust. How to find advantage in unpredictability. It was strategy disguised as play.

I did not recognise it then, but that yard taught me how to see the world.
It taught me that meaning lives inside texture.
That understanding happens in layers.
That anticipation is a form of awareness.

The red earth never left me.
It remains a small origin point of how I sense things now.
And when I stand before the Moritz triptych, that same quiet feeling returns.

The Rhythm My Mind Has Always Known

The connection I feel to this painting is not about subject matter. It is about rhythm. Moritz paints in a way that mirrors how I think.

Understanding has never arrived for me in straight lines. I need to circle an idea. I need to look at it in more than one light. I need the silence between perspectives to let things settle. This triptych moves with the same internal cadence.

It holds composure and tension at the same time. It changes depending on the state of mind you bring to it. Some days it feels calm. Other days it feels unresolved. It reflects internal weather, not external fact.

Moritz revises her clarity. She interrupts herself. She trusts time. These instincts feel familiar. I recognise that way of approaching meaning. I recognise the patience and the willingness to allow truth to gather slowly rather than forcing it.

The painting becomes a quiet companion to the part of me that values depth over haste and reflection over performance.

The Quiet Mirror Inside All of Us

“The only interesting answers are those that destroy the questions.”
Susan Sontag

This work speaks to something shared. Something universal. Every person carries layers that do not align neatly. Everyone has memories that shift tone depending on the day. Everyone has emotions that arrive before words. Everyone contains contradictions that feel coherent inside and confusing outside.

Most people assume this complexity is a flaw to hide.
Moritz reveals that it is the signature of an inner life.

Her painting does not ask you to interpret it. It asks you to recognise yourself in its atmosphere. It reframes ambiguity as intelligence. It reframes patience as a way of thinking. It reframes uncertainty as the space where understanding begins.

In a culture that demands instant clarity, this work stands still and reminds you that clarity is something that arrives, not something you grab.

It gives you permission to let your inner world speak in layers.

A Gentle Invitation to Listen Inward

If this painting has taught me anything, it is that the inner world has its own seasons. Its own gravity. Its own timing. It does not reward urgency. It rewards attention. It rewards the willingness to sit with what is unresolved without demanding that it resolve immediately.

The triptych reminds me that meaning often appears in the quiet intervals, in the redirects, in the subtle shifts. It reminds me that depth comes from presence, not pressure. It reminds me to trust the slow revelation of truth, the kind that does not announce itself but grows.

Moritz does not instruct.
She invites.
And in accepting that invitation, you begin to hear something inside yourself more clearly.

The more time you spend with the painting, the more it becomes a companion to the part of you that is still forming silently beneath everything else.

Dear Partners,

The last few weeks have been truly magical for me. I have stepped back from the daily noise to spend my time reading, absorbing, meditating, and perhaps most importantly golfing.

To the outside observer, this might look like leisure. But in the pursuit of being a great investor, I view it as essential maintenance. Blaise Pascal once wrote that all of humanity’s problems stem from man’s inability to sit quietly in a room alone. I believe this is especially true. Investing is not an activity of constant movement. It is an activity of constant judgment. And judgment requires a quiet mind.

Golf is a ruthless teacher of this truth. On the course, you learn very quickly that you cannot force the ball into the hole. If you swing with tension, you fail. If you let your mind drift to the mistake you made three holes ago, or worry about the score you hope to have at the end, you lose your rhythm. You must be entirely present. You must be centered.

This centeredness is the foundation of my investment process. It allows me to detach from the frantic energy of “Mr. Market” and see things clearly. It reminds me that success does not come from doing more. It comes from doing the right things with absolute discipline.

This brings me to the fifth chapter of Poor Charlie’s Almanack, where Charlie Munger outlines a framework for thinking that goes far beyond arithmetic. He argues that to succeed in a complex world, one must master multiple disciplines. To explain this, he points to the profession of piloting. A great pilot does not rely on instinct alone. He relies on an understanding of weather, physics, and engineering. But more importantly, he relies on the mandatory practice of the checklist. Munger notes that pilots learn to master their own psychology by constantly visualizing errors. They invert the problem. They do not ask how to have a great flight. They ask what would cause a crash, and then they systematically avoid those things.

This philosophy resonated deeply with me because long before I managed capital, I learned this lesson in the cockpit of a single-engine aircraft, crossing the Bass Strait to Tasmania in the dead of night.

When you are suspended over that stretch of water in the dark, the margin for error evaporates. Below you is nothing but cold, pitch-black ocean. Above you is a void. There is no visible horizon to orient yourself. In that silence, the dangers are not just mechanical. They are sensory. Your inner ear can betray you. It can convince you that you are banking left when the instruments insist you are level. A single fuel pump failure turns the plane into a glider. Unexpected ice can alter the aerodynamics in seconds.

You do not survive that journey by being creative or relying on how you feel. You survive by adhering to a strict, unemotional process. You trust the instruments over your own senses. You scan the oil pressure. You calculate the fuel burn. You assume the engine might quit and you plan the glide path before the silence comes. You respect the physics of the situation more than your own confidence.

I have brought this same mindset to managing money, and I take the role of being a good steward of your capital seriously. The market is just as unforgiving as the Bass Strait. It punishes arrogance and rewards preparation. Therefore, we do not simply buy stocks. We run every potential investment through a pre-flight checklist. We look for structural integrity. We ignore the noise of the headlines to focus on the physics of the business.

This is why we obsess over The Physical Foundation. In the disorientation of a market panic, stories and projections vanish like fog. What remains is what is real. We look for the land, the bricks, and the machines—a lesson I learned firsthand when I built my aviation pilot training business. We want to know that if the engine stops and the business closes tomorrow, the pile of things left behind is worth more than the price we paid. It is the only way to ensure we are flying with a parachute.

It is also why we discard complex accounting and search for The Take-Home Cash. The instrument panel of a plane tells you the truth about your fuel regardless of how fast you want to go. Similarly, we look for the money a business generates only after it has paid to keep its doors open and its roof repaired. We ignore the profit that exists only on paper. We demand the cash that can actually be put in your pocket. A pilot does not fly on the promise of fuel. He flies on what is in the tank ;)

This discipline extends to our view on The Financial Weight. To a pilot, altitude is safety and airspeed is life. To an investor, cash is oxygen and debt is a heavy backpack. We prefer to fly light. We avoid companies that require perfect conditions to survive because we know that perfect conditions never last. We search for businesses that can glide through a storm without needing a lifeline from a bank.

We find comfort in The Quiet Corners. The crowd is often like a flock of birds. They turn sharply and unpredictably based on what the bird next to them is doing. They chase the excitement. We prefer the empty airspace. William James taught that the art of being wise is the art of knowing what to overlook. We overlook the noise. We look for the companies that have been forgotten in the quiet cove where the water is still. It is in this silence that we often find the greatest value.

And perhaps most importantly, we look for Shared Fate. There is an ancient Roman tradition regarding bridge building that guides us here. The engineer who designed the arch was required to stand underneath it when the scaffolding was removed. If the bridge failed, the engineer fell with it. We apply this same test to management. We only invest with leaders who are standing under their own arches with their own net worth tied to the success of the enterprise.

These principles act as our Attitude Indicator (the gyroscope). In the cockpit, this is the most important gauge in the cluster; the primary scanning technique of a pilot is to constantly return to it. These principles do the same for us. They keep us level when your inner ear tells you the world is tilting. They allow us to sit quietly while others rush. We do not need to be smarter than the crowd. We simply need to be more disciplined. We are content to check our instruments, collect our dividends, and wait for the signal to eventually drown out the noise.

Thank you for trusting us with your capital. We treat it with the same care we treat our own, planting it in good soil and watching it grow.

Regards,

Neel

When a struggling $7 million railcar inspection company pivots to edge data centres and power generation—and lands a $42 million contract in the process—you pay attention. Duos Technologies (NASDAQ: DUOT) has executed one of the most dramatic business transformations I’ve tracked in micro-caps, growing revenue 280% year-over-year in Q2 2025, and positioning itself at the intersection of three massive trends: edge computing, AI infrastructure, and data centre power. The question isn’t whether the opportunity is real—it’s whether this tiny company can execute at scale.

Here’s what makes DUOT fascinating: they’re not starting from scratch. For over a decade, they’ve been building mini data centres along railroad tracks to power AI-enabled inspection systems that scan 10 million railcars annually. Now they’re taking that edge computing expertise and deploying it where the real money is—powering the AI revolution’s insatiable appetite for distributed computing and electricity.

From railroad safety to the AI gold rush

DUOT’s legacy business sounds niche because it is: their Railcar Inspection Portal (RIP) uses machine vision and AI to inspect freight trains travelling at 125 MPH, capturing 224 megapixels per railcar and running 40+ detection models on NVIDIA GPU farms.

They’ve scanned 44% of North America’s freight-car population—700,000 unique railcars—and serve four Class 1 railroads, as well as Amtrak.

It’s legitimate technology with 10 patents and real customers.

But here’s the problem: railroads move slowly. Despite achieving over 95% accuracy and processing 1.3 terabytes daily, DUOT’s rail business generated just $7.28 million in 2024, basically flat year-over-year.

Customer deployment delays plagued them throughout 2023-2024. CFO Adrian Goldfarb put it bluntly in Q2 2025: “The speed at which the rail industry would adopt our solutions and, as a small company, our ability to influence the industry, seeing as if the time it might take and the financial resources needed might not be compatible with providing the returns our shareholders are expecting.”

So in June 2024, CEO Chuck Ferry announced a radical pivot. Actually, not a pivot—a diversification into two entirely new businesses that would dwarf the rail operations within months.

The transformation playbook: leveraging what you already know

Ferry didn’t pick these new businesses randomly. His background tells you everything you need to know about the strategy. Before joining DUOT in September 2020, he was CEO of APR Energy from 2018 to 2020, running a global power generation company with $325 million in revenue, 800 employees, and 17 power plants worldwide. From 2016 to 2018, as President/COO, he managed a 2.2 gigawatt fleet generating $1.3 billion in revenue. Before that, he scaled ARMA Global Corporation from $20 million to $200 million in revenue before its acquisition by General Dynamics. Oh, and he’s a decorated combat veteran with a Silver Star and a Bronze Star for Valour, including participation in the Battle of Mogadishu.

This is a crucial context because the transformation leverages Ferry’s exact domain expertise.

The first new business: Duos Edge AI

On June 18, 2024, DUOT announced the formation of Duos Edge AI and hired Doug Recker as President. Recker brings serious credibility—he founded EdgePresence, sold it to Ubiquity in 2023, and founded Colo5 Data Centers before that, which Cologix acquired in 2014. Two successful exits in edge data centers. He’s a 30-year telecommunications veteran who won Jacksonville Business Journal’s Ultimate CEO Award.

The insight was elegant: DUOT had already been building mini edge data centres for its rail inspection portals—structures housing 10 servers with 48 NVIDIA T4 GPUs each, processing 1.3TB daily in remote, harsh environments. They had the engineering expertise, the deployment speed (90 days from order to operational), and relationships with power providers. Why not build edge data centres for the exploding AI/edge computing market?

By September 4, 2024—less than three months later—their first Edge Data Centre went operational in Amarillo, Texas, at the Region 16 Education Service Centre, serving 60 school districts, 226 campuses, and 83,000 students across 26,000 square miles.

The unit economics are compelling: each EDC costs $1.2-1.4 million to build and generates $300-500K in annual recurring revenue at full utilization, with target gross margins in the mid-70% range and EBITDA margins just above 50%. At maturity, each EDC generates roughly $300K in free cash flow annually starting in year two. With 5-year contracts (plus 5-year extensions) and 10-year carrier deals, you’re looking at highly predictable, recurring infrastructure revenue.

The deployment schedule is aggressive: 15 EDCs by the end of 2025, 65 by the end of 2026, and 150+ by mid-2028. As of Q2 2025, they had nine sites identified for commercial development, with signed letters of intent or contracts, including deployments in Victoria, Corpus Christi, Waco, and Dumas, Texas. They’ve partnered with FiberLite, a national fibre provider, to accelerate the pipeline.

Do the math: 150 EDCs generating $400K annually each equals $60 million in recurring revenue from this business line alone at maturity.

The second new business: Duos Energy Corporation

On November 20, 2024—just five months after announcing the edge strategy—DUOT dropped another bombshell: a $42 million, two-year Asset Management Agreement with Fortress Investment Group to manage 850 megawatts of mobile gas turbines acquired from APR Energy.

Here’s where Ferry’s background becomes critical. This isn’t some random diversification—Ferry’s old team from APR Energy literally installed over one gigawatt of power from 2016-2020. COO Chris King also came from APR Energy, where he led all power plant operations across 16 sites with 500+ employees. The management team has over 100 years of combined experience in power projects.

The deal structure is brilliant: DUOT doesn’t buy the assets (Fortress bought them). Instead, DUOT provides asset management, installation supervision, and operations management services while taking a 5% equity stake in the parent company that owns the assets (valued at $7.2+ million as of Q2 2025). The $42 million represents estimated fees over two years, plus they received a $5 million advance for working capital.

Why does this matter? Because AI data centres have a massive power problem. As TD Cowen analyst Michael Elias noted: “AI is driving a re-rating of the data centre demand as the data centre leasing seen in the last 12 months is the equivalent of adding an entire New York City to the US electricity grid.”

Traditional utility power takes 24+ months to provision. DUOT’s mobile turbines can be deployed in 30 days. They proved it by installing a 150MW gas turbine plant in Mexico in just 35 days.

In the first three months of operation, Duos Energy secured 390MW worth of contracts. Ferry noted they’re tracking 35+ opportunities, with 30 being data centre developers. If they capture just 5-10 of those opportunities, the entire 850MW fleet sells out.

This creates perfect synergy: DUOT can offer edge data centers with integrated behind-the-meter power solutions, solving both the computing and electricity bottlenecks simultaneously.

The IREN parallel: bitcoin miners pivoting to AI infrastructure

The closest comparable to DUOT’s transformation is Iris Energy (NASDAQ: IREN), which successfully transitioned from pure bitcoin mining to AI/HPC infrastructure, achieving spectacular results.

IREN’s journey provides a roadmap. Founded in 2018 as a bitcoin miner focused on renewable energy sites, they IPO’d in November 2021 at $28/share. Through the 2022 crypto winter, the stock crashed to $1.02 in December 2022. Unlike competitors that went bankrupt (Core Scientific filed Chapter 11), IREN used the downturn to diversify into AI cloud services starting in early 2024.

The financial results tell the story. FY24 (ended June 2024): $187.2 million revenue, $28.9 million net loss, $54.7 million adjusted EBITDA. FY25 (ended June 2025): $501 million revenue (168% growth), $86.9 million net profit (turnaround from loss), $270 million adjusted EBITDA (395% growth). By Q4 FY25, they generated $176.9 million in net profit in a single quarter.

The stock went vertical: +500% in 2023, another +44% in 2024, and +434% year-to-date in 2025,hitting an all-time high of $74.15 in October 2025 from the $1.02 low.Current market cap: $13.72 billion. Analysts have average price targets of $59-71 with a “Buy” consensus.

IREN’s model differs from DUOT—they own and operate 10,900 NVIDIA GPUs (targeting 23,000 by September 2025) with 810MW of operating data centres and 2,910MW of contracted grid connections. They’re vertically integrated with 100% renewable energy. But the strategic parallel is identical: leverage existing data centre infrastructure and power expertise to capture AI infrastructure demand that pays 2-3x more per megawatt than bitcoin mining.

Key differences: IREN is 70x larger, has a clean balance sheet, and holds NVIDIA Preferred Partner status, and has avoided bankruptcy. Core Scientific (another miner-to-AI transition story) emerged from bankruptcy in January 2024 and focuses on hosting (not owning) AI equipment for hyperscalers like CoreWeave, with 570MW committed to AI/HPC. Their stock rose 312% in 2024.

The pattern is clear: companies with power infrastructure, data centre expertise, and speed-to-market are capturing outsized value in the AI buildout. DUOT is following the same playbook, but at a much smaller scale and at an earlier stage.

The edge data center opportunity: massive and growing

Multiple research firms peg the total addressable market for edge data centres at $10-16 billion today, growing to $50-110 billion by 2030-2033 at compound annual growth rates of 18-33%. The range reflects different definitions of “edge,” but the direction is unambiguous.

Grand View Research projects $12.36 billion in 2024, growing to $109.91 billion by 2033 (28.9% CAGR). Grand View ResearchGrand View Research estimates $10.4 billion in 2023 reaching $29.6 billion by 2028 (23.2% CAGR). MarketsandMarketsMarketsandMarkets Precedence Research shows $15.46 billion globally in 2024 hitting$84.41 billion by 2034 (18.5% CAGR). Precedence ResearchPrecedence Research

The drivers are fundamental: 5G network rollout (1.76 billion global connections by 2025), IoT device proliferation (27-42 billion connected devices by 2025), Verified Market Research latency-sensitive AI applications (autonomous vehicles, industrial automation, telemedicine), and data sovereignty requirements (GDPR, CCPA). Grand View Research +2 Smart city initiatives alone represent billions in government spending on distributed infrastructure.

North America dominates with 34-42% market share, driven by hyperscaler concentration, advanced 5G deployment, and enterprise adoption. MarketsandMarkets +4 Asia Pacific is growing fastest at 25-31% CAGR due to China and Indiadigitalization. Grand View ResearchGrand View Research

The edge AI infrastructure segment is explicitly even hotter. Grand View Research pegs it at $20.78 billion in 2024, growing to $66.47 billion by 2030 (21.7% CAGR). Grand View Research STL Partners forecasts $54 billion in 2024 reaching $157 billion by 2030 (19% CAGR), with total edge computing TAM hitting $424 billion by 2030.

The economics favour AI over traditional workloads dramatically. VanEck estimates AI/HPC generates $1.30 per kWh versus $0.43-0.65 for bitcoin mining—a 2-3x revenue premium. If bitcoin miners convert just 20% of their capacity to AI by 2027, it could create $37.6 billion in net present value across the sector.

DUOT is targeting the small-to-medium edge facility segment in Tier 2/3 cities—the fastest-growing subsegment at over 22% CAGR—in markets underserved by Digital Realty, Equinix, and EdgeConneX. It’s a classic David-versus-Goliath strategy: avoid direct competition with $40-75 billion market-cap players by serving rural Texas school districts, small municipalities, and regional enterprises that need localized compute within 12 miles of end users.

Following the money: Q3 2024 and recent financial results

The numbers prove the transformation is fundamental, not just press releases.

Q3 2024 results (reported November 2024):

• Total revenue: $3.24 million, up 112% year-over-year

• Technology systems: $1.69 million, up 139%

• Services and consulting: $1.55 million, up 88%

• Gross margin: $919K (28.4%), up 306% from $226K (14.8%)

• Net loss: $1.40 million, improved 53% from $2.95 million

• Operating loss: $1.92 million, improved 35% from $2.97 million

The gross margin expansion was key, even though the company was still unprofitable. Operating expenses fell 11% to $2.84 million as management cut 5% of staff and reallocated personnel costs to power consulting projects.

Full year 2024:

• Revenue: $7.28 million (down 3% due to customer deployment delays on Amtrak high-speed portals)

• Backlog entering 2025: $50.5 million, up dramatically from $6.4 million at end of 2023

• Of that backlog, $22.6 million expected to be recognized in 2025 plus $8-9 million in near-term renewals

Q2 2025 results (most recent):

• Revenue: $5.74 million, up 280% year-over-year

• H1 2025 revenue: $10.69 million, up 314% year-over-year— the highest six-month revenue in company history

• Gross margin: $1.52 million (26.5%), up 808% year-over-year

• Net loss: $3.52 million (still unprofitable but improving trajectory)

The revenue mix shifted dramatically. In Q2 2025, services and consulting revenue were $5.69 million (99% of total) with only $52K from technology systems. Within services, $4.76 million came from the APR Energy Asset Management Agreement. An additional $904K came from recognising a portion of the 5% equity interest in the APR Energy parent company (accounted for as 100% margin revenue under equity method accounting).

2025 guidance: Management projects $28-30 million in total revenue, representing 285-312% growth from 2024’s $7.28 million. More importantly, CFO Goldfarb stated they expect to achieve adjusted EBITDA profitability by Q4 2025. CEO Ferry was even more direct: “These revenues, along with the backlog we already have, and expected growth of our edge data centre business and rail car inspection portal business in the coming year, allow me to say we will become profitable in 2025 confidently.”

Management: deep domain expertise with execution questions

The leadership team has the right backgrounds for this transformation, but the track record raises concerns.

Chuck Ferry (CEO since September 2020) brings unquestionable operational credentials. Beyond his APR Energy pedigree running a $325 million global power company, he scaled ARMA Global from $20 million in revenue and 80 employees to $200 million in revenue and 1,200 employees before its General Dynamics acquisition. His military background (26 years in the Army, Silver Star, Bronze Star for Valour, Battle of Mogadishu participant) suggests strong execution discipline. He owns approximately 2.65% of DUOT directly, plus 522,889 unvested RSUs that vest January 1, 2028, creating multi-year alignment.

But there’s a significant red flag: Ferry simultaneously serves as Chairman and CEO of APR Energy as of 2025. Running two companies creates conflicts of attention and raises questions about where his priorities lie.

Doug Recker (President, Duos Edge AI) is the strongest hire in the transformation. Two prior exits (EdgePresence sold to Ubiquity in 2023, Colo5 sold to Cologix in 2014), 30+ years in telecommunications/data centres, multiple industry awards, including Jacksonville Business Journal’s Ultimate CEO Award. He has the exact expertise needed to scale the edge data centre business and brings operational credibility.

Valuation: overlooked micro-cap with coverage gap

At $9.48 per share (recent trading), DUOT has a market cap of roughly $192 million and an enterprise value of around $160 million.

Valuation multiples:

• Price-to-sales (TTM): 4.86-8.1x depending on calculation (trailing twelve months ~$15.39 million)

• EV/forward sales: approximately 5.5-6.6x on 2025 guidance midpoint of $29 million

The software sector peer average P/S ratio is 5.1-5.3x, so DUOT appears roughly in line on forward metrics. But here’s the critical distinction: average software companies grow 5-10% annually, while DUOT is guiding for 285-312% growth in 2025 with a line of sight to $60+ million from 150 edge data centres at maturity by 2027-2028.

The primary driver of undervaluation is the coverage gap. When micro-caps with strong growth metrics are “discovered” by institutional investors, rapid revaluations of 50-100% can occur. DUOT’s addition to the Russell Microcap Index, oversubscribed $40 million offering with institutional lead, and improving trading volume (from less than 10K shares/day to 300K+ shares/day per CFO Goldfarb) suggest the discovery process has begun but isn’t complete.

Competitive realities: small fish in big ponds

Edge data centre market: Competition is vastly more intense. Tier 1 players include Digital Realty Trust ($40 billion market cap, 300+ data centres globally), EdgeConneX (private, owned by EQT Infrastructure, 40+ data centres across 30 markets with a global footprint), and Equinix ($75 billion market cap, 250+ facilities, dominant in interconnection).

DUOT has less than 0.1% market share, with 5-10 operational or under-construction facilities, compared to thousands of edge facilities globally. They’re David fighting Goliaths.

But DUOT isn’t competing head-to-head. Their strategy targets a specific niche: small/medium edge sites in Tier 2/3 U.S. markets (rural/suburban Texas initially), integrated power solutions leveraging energy expertise others lack, a partnership-driven model with schools and municipalities versus wholesale hyperscale customers, and speed over scale with 90-day deployment timelines.

The closest comparable is Leading Edge Data Centres in Australia, which follows a similar “bridge the digital divide” mission serving regional markets with Tier 3 facilities.

DUOT’s differentiation centres on three factors:

1. Integrated “behind the meter” power solutions: The ability to deploy mobile gas turbines in 35 days (proven in Mexico) solves the number one bottleneck for data centres—power availability. Competitors like Digital Realty and EdgeConneX lack in-house power generation expertise.

2. Underserved market focus: Targeting Tier 2/3 cities where Digital Realty and Equinix don’t operate avoids direct competition. Texas school districts, rural healthcare systems, and small municipalities need edge computing but can’t access or afford hyperscale providers.

3. Speed to deploy: 90-day delivery on modular edge data centres versus 12- 24+ months for traditional builds creates a time arbitrage opportunity

The risk is that major players (Digital Realty, EdgeConneX) or hyperscalers (AWS Local Zones, Google Edge) could easily enter DUOT’s markets if they’re proven profitable. Telecom companies (AT&T, Verizon) are also building edge networks that could compete directly. DUOT’s window to establish market position before larger competitors arrive is 18-36 months.

In power/energy services, DUOT benefits from management’s deep APR Energy experience, but they’re managing assets (not owning them), which limits long-term value creation compared to asset-heavy models. The 5% equity stake in the New APR Energy parent is interesting but speculative.

Unit economics on edge data centres:

• Total installed cost: $1.2-1.4 million per EDC

• Annual revenue per EDC: $300-500K (fully commercialised)

• Gross margin target: mid-70% range

• EBITDA margin target: just above 50%

• Free cash flow (year 2+): ~$300K annually per EDC

• Payback period: approximately 2 years

At 150 EDCs by mid-2028, you’re modelling $45-75 million in edge revenue alone (using $300-500K per EDC range) with EBITDA of $23-38 million. Add rail and power services, and the $60-80 million revenue target management discusses becomes plausible.

The capital requirement is substantial: $180-210 million to build 150 EDCs. DUOT has raised ~$105 million between September 2024 and July 2025, enough for 15-20 deployments. Scaling to 65 EDCs by the end of 2026 requires another $50-70 million (assuming 40-50 incremental units from the current base). Scaling to 150 requires partnerships, joint ventures, or additional capital raises, all of which will dilute shareholders.

Recent catalysts validating the transformation

The last 12 months have produced a series of validation points:

November 2024: $42 million Asset Management Agreement with Fortress/New APR Energy—largest contract in company history, 5% equity stake valued at $7.2+ million

October 2024: 5-year strategic agreement with CN Railway enabling subscription service for AI wayside detection data, building on 5+ years of CN using DUOT technology

September 2024: First edge data centre operational in Amarillo (Region 16 ESC), serving 60 school districts—proof of concept achieved on schedule

Q3 2024: $1.4 million contract modification for Amtrak high-speed portal project; Amtrak subscription renewal utilisingthree portals

June 2024: Doug Recker appointed President of Duos Edge AI; formation of subsidiary announced; strategic partnership with Class 1 railroad for subscription-based inspection system

Q1 2024: $2.7 million RIP contract for new industrial application, expanding addressable market beyond traditional rail

Early 2025: 390MW in power contracts secured in the first 3 months of Duos Energy operation; multiple edge data centre deployments (Victoria, Pampa, Dumas, Waco, Corpus Christi); strategic partnership expansion with FiberLite

May 2025: Brigadier General Craig Nixon appointed Board Chairman, bringing M&A expertise and operational scaling experience

July 2025: $40 million oversubscribed public offering at $6.00/share led by institutional investors, plus $12.5 million ATM offering—total $52.5 million capital raised

August 2025: Congressman Ronny Jackson’s staff visited the Region 16 EDC deployment in Amarillo, indicating government interest

The technology validation is also solid: 10 active U.S. patents for Railcar Inspection Portal, six additional patents pending, a new patent granted in 2024 for a high-resolution image capture device, and IP infringement letters issued to a Class 1 railroad and a technology partner, demonstrating a defensible intellectual property position.

Management quotes throughout 2024-2025 show increasing confidence. CEO Ferry in Q4 2024: “Duos Energy is capitalising on unprecedented demand for behind-the-meter power solutions, securing contracts for 390MW in just the first three months of operation, with additional deals in negotiation. The synergies between our power and edge computing businesses have exceeded expectations, opening doors to new opportunities across both sectors.”

CFO Goldfarb in Q2 2025: “For the first time in the company’s history, we are sufficiently capitalised to take advantage of the new markets we have entered. In addition to the rise in the share price, our average trading volume has increased from less than 10,000 shares a day, which The Street calls trade by appointment, to more than 300,000 shares per day.”

And most tellingly, Goldfarb on the transformation rationale: “Despite the outstanding achievements we had made in developing the technology underlying the railcar inspection portal, the speed at which the rail industry would adopt our solutions and as a small company, our ability to influence the industry, seeing as if the time it might take and the financial resources needed might not be compatible with providing the returns our shareholders are expecting.” Translation: the rail business wasn’t going to make shareholders money fast enough, so we’re pivoting to where the money actually is.

The bull case: massive opportunity with proof of concept

Now the optimistic scenario:

TAM expansion: DUOT went from a $7.5 billion rail inspection TAM, where they were stuck at $7 million in revenue, to a $50-110 billion edge data centre TAM growing 18-33% annually. Even capturing just 0.1% of the edge market by 2028 would yield $50-110 million in revenue. The power/energy services addressable market is similarly massive, with AI data centres driving unprecedented demand.

Revenue visibility: $50.5 million backlog (including near-term extensions) with $22.6 million expected in 2025, plus $8-9 million in near-term renewals, provides high confidence in the $28-30 million guidance. Q2 2025’s 280% growth proves the transformation is fundamental, not just a matter of press releases.

Unit economics work: The first edge data centre in Amarillo was deployed on schedule and is operational. At $1.2-1.4 million cost and $300-500K annual revenue, you achieve a 2-year payback with 70%+ gross margins and 50%+ EBITDA margins. This is proven, not theoretical. Scaling to 150 units generates $45-75 million in recurring revenue with $23-38 million EBITDA—a legitimate business at maturity.

Management expertise perfectly aligned: Ferry’s APR Energy background (CEO of $325M company), Recker’s two edge data centre exits (EdgePresence to Ubiquity, Colo5 to Cologix), and King’s operational experience (16 power plants, 500+ employees) mean this team has done it before. They’re not learning on the job

they’re executing a playbook they’ve run successfully elsewhere.

Competitive differentiation: The integrated power + edge data centre offering is unique. No major competitor combines behind-the-meter mobile turbine deployment (35 days in Mexico) with edge colocation facilities. Speed matters enormously when hyperscalers need compute capacity immediately. Targeting underserved Tier 2/3 markets avoids Digital Realty and Equinix while serving real customer needs (rural schools, healthcare, municipalities).

Financial inflection point: After 13 years of losses, DUOT is targeting profitability in Q4 2025. If achieved, it triggers a fundamental re-rating from “perennial loss-maker” to “profitable growth company.” The path is clear: $29 million revenue at 26.5% gross margins (Q2 2025 level) yields $7.7 million gross profit. With operating expenses at ~$2.76 million quarterly ($11 million annually), breakeven requires maintaining current margins and delivering guided revenue—achievable with $4.76 million quarterly from the APR Energy agreement alone, plus edge deployments ramping.

Validation from sophisticated capital: The July 2025 $40 million offering was oversubscribed and led by a “leading long-only mutual fund” institutional investor at $6.00/share. Fortress Investment Group committed $42 million via the APR Energy agreement. These aren’t retail speculators they’re sophisticated investors who conducted diligence and wrote extensive checks.

Coverage gap opportunity: Only two analysts covering creates massive inefficiency. Research shows micro-caps with limited coverage trade at 15-30% discounts. When institutional discovery happens (triggered by Russell Microcap addition, improved liquidity, profitability achievement), revaluations of 50-100% occur.

Comparable valuations: If DUOT achieves $60-70 million in revenue by 2026-2027 with EBITDA profitability and a recurring revenue model, comparables include: EdgeConneX (private, valued at billions with 40+ facilities), Digital Realty (10x sales), and small public data centre operators (6-8x sales). Applying a 5-6x multiple to $65 million revenue yields $325-390 million valuation versus $192 million today 70-100% upside over 2-3 years.

IREN parallel: Iris Energy’s transformation from $187 million in revenue to $501 million (168% growth), driven by profitability, drove stock appreciation of 500%+ and created a $13.7 billion company from the ashes of a $1.02 stock during crypto winter. DUOT is following the same playbook (leveraging power/data centre infrastructure for AI demand) 1-2 years behind IREN, but at a tiny fraction of IREN’s valuation. If DUOT achieves even 20% of IREN’s success, shareholders will do very well.

Recession resistance: Data centre and power infrastructure are counter-cyclical. AI/edge demand is being driven by secular technology adoption (5G, IoT, AI workloads), not economic cycles. School districts, healthcare systems, and telecommunications companies will continue investing in digital infrastructure regardless of GDP growth.

My take: high-risk micro-cap growth with asymmetric upside

DUOT is precisely the kind of Peter Lynch “fast grower” story that can generate 5-10x returns or lose 50% if execution fails. It’s not for widows, orphans, or anyone who checks their portfolio daily.

Here’s what I find compelling: this isn’t a speculative story about technology that might work someday. The edge data centre model has been proven (Amarillo EDC operational since September 2024). The unit economics are validated ($1.2-1.4 million cost, $300-500K annual revenue, 2-year payback). The revenue growth is real (280% in Q2 2025). The backlog is substantial ($50.5 million). The capital is in place ($52.5 million raised July-August 2025). The management team has relevant expertise (Ferry ran a $325 million power company, Recker has two data centre exits). And the TAM is enormous (edge data centres growing from $10-16 billion to $50-110 billion by 2030).

What concerns me: Ferry’s dual CEO role, CFO instability, customer concentration on APR Energy (83% of Q2 revenue), never achieving profitability in 13 years, and the reality that Digital Realty or EdgeConneX could crush DUOT in their target markets if they wanted to. The need for additional capital raises before reaching scale will further dilute shareholders.

The valuation at $9.48/share ($192 million market cap) seems fair, not cheap. At 5.5-6.6x forward 2025 revenue, you’re paying for the growth story with little margin of safety. If they miss guidance, the stock trades down to $6-7. If they achieve Q4 profitability and deploy 15 EDCs on schedule, the stock is likely to reach the $11-13 analyst target range (15-37% upside). If they execute flawlessly through 2026 (65 EDCs deployed, $50+ million revenue, sustained profitability), you get to $15-20+ (60-110% upside).

The IREN comparison is instructive. They went from $1.02 to $74 (7,150% gain) by executing a similar pivot from bitcoin mining to AI infrastructure. DUOT doesn’t need anything close to that level of success to generate strong returns just steady execution in 65-150 edge data centres over 2-3 years while maintaining the APR Energy relationship.

The coverage gap is real and creates opportunity. With only 2 analysts and 32.5% institutional ownership versus 53.7%sector average, there’s room for 30-50% expansion if the story gets discovered. In addition to the Russell Microcap Index and the recent $40 million oversubscribed institutional offering, the discovery process has begun.

My conclusion: DUOT offers asymmetric risk/reward for aggressive growth investors willing to accept micro-cap volatility and execution risk. The transformation from $7 million rail tech company to $28-30 million diversified AI infrastructure play is legitimate, backed by concrete contracts, proven technology, and experienced management.