Public equities · November 8, 2026
Public Equity Thesis
In public markets, Tuuri Capital owns the physical layer of artificial intelligence: the power, buildings and components that turn electricity into compute, and the machines that put compute to work in the physical world. We buy it through companies that hold something scarce — secured power, permitted sites, long-term leases with the companies building AI, or manufacturing capacity that takes years to replicate — and we hold them for years.
How we see it
The fund has one job: compound capital faster than the market. Everything on this page serves that, and nothing on this page outranks it.
Compute is the new oil. The companies that control it will set the terms for everything built on top of it, and the next decade's value will go to whoever owns the scarce part of that supply chain.
Five beliefs sit under every position we hold:
- Intelligence is becoming an industrial input. Every business will buy it the way it buys electricity, and spend on AI capacity will keep rising for years, with pauses along the way.
- The bottleneck moves. It was chips; it is now power, land, cooling and the buildings that turn electricity into compute, and the slowest layer earns the margin.
- Markets price the obvious layer and miss the ones beside it. The chip designer is famous; the operator that paid for the grid interconnect is not, and that is where we look.
- We lean toward what has to be built. Software is rewritten every quarter; a substation, a turbine, a battery plant and a permitted site are not. It is a bias, not a rule: the best return in the stack gets the capital, wherever it sits.
- Ownership beats prediction. We cannot time the cycle, so we buy a durable share of a shift we are confident in, size it to survive being early, and let the shift do the work.
The fund today
The thesis is the whole physical stack; the fund holds one slice of it today. What we own is where the best price met the scarcest layer, not a ranking of conviction.
AI needs more than chips. AI is a five-layer cake: energy, chips, infrastructure, models, and applications; our reading of the layers is below. Most of the attention, and most of the valuation, sits on the chips. Today we own the layers on either side of them: the power that runs them and the buildings that house them. The infrastructure layer is the AI factory: the land, power delivery, cooling and systems that turn electricity into intelligence. The table follows the power from the plant to the chips.
The five layers
NVIDIA popularized the picture of AI as a five-layer cake. We use the same five layers, read from the bottom up, because that is the order the money has to flow.
Energy.
Generation, transmission and the grid interconnect that brings hundreds of megawatts to one site. Power is the slowest thing in the stack to build and the hardest to replicate, so a secured interconnect is worth more than a plan for one.
Chips.
The accelerators, memory and networking inside the racks. The most visible layer and the most completely priced: its suppliers earn the widest margins today and face the most competition tomorrow.
Infrastructure.
The AI factory: the land, buildings, cooling and power delivery that house the chips, and the clouds that network them and sell the output. Contracts here run ten to fifteen years and appear in filings. This is the layer we own today.
Models.
The labs that train and serve intelligence. Capital-hungry and fast-moving, and the leader changes every few quarters.
Applications.
The products people and businesses actually use. The largest market and the thinnest moat; software is rewritten every quarter.
Everything above the infrastructure layer rents from it; everything below it feeds it. We buy the layers with long contracts and physical scarcity, energy and infrastructure first, and the components beside the chips when they are sold out for years. We pass on the layers that are rewritten every quarter.
| Company | What it does | Where it sits in the AI factory | Five-layer cake |
|---|---|---|---|
| Keel Infrastructure (opens in a new tab) | Formerly Bitfarms; develops AI data centers and the power behind them in Pennsylvania, Washington and Québec | Power in, and the land; its buildings are in development | Energy and infrastructure |
| TeraWulf (opens in a new tab) | Builds and leases AI data centers in New York and Kentucky | The building: land, power delivery and cooling | Infrastructure |
| Cipher Digital (opens in a new tab) | Builds and leases AI data centers in Texas | The building: land, power delivery and cooling | Infrastructure |
| Applied Digital (opens in a new tab) | Builds and leases AI data center campuses in North Dakota, Louisiana and Alabama | The building: land, power delivery and cooling | Infrastructure |
| CoreWeave (opens in a new tab) | A cloud built for AI: rents GPU computing to AI labs and hyperscalers | The machine inside: networks the chips, runs them as one, and sells the output | Infrastructure |
What would change our mind:the largest cloud buyers (Microsoft, Alphabet, Amazon, Meta) guiding data center spending flat or down; CoreWeave's contracted backlog shrinking; the hosting operators going several quarters without signing new capacity; a tenant failing to pay or renegotiating a lease down; or a contracted project that cannot be financed at any price. We track each of these in the companies' own filings, not in the share price.
What we look for, and what we pass on
We ask five questions of every company. Profit today is not one of them; contracts are. Price comes last, and it decides when we buy, not whether.
| We ask | A fit looks like | Not a fit |
|---|---|---|
| Is it on the right side of a shift we believe in? | Sells into demand that is still steepening, measured in gigawatts and backlog | Growth that depends on a cycle turning or a trend reversing |
| Does it own something scarce? | Secured power, grid interconnects, permitted land, or manufacturing capacity and technology that take years to replicate | A position anyone with capital can replicate in a year |
| Is the demand contracted, or promised? | Signed leases, orders or backlog from creditworthy buyers: hyperscalers, AI labs, utilities, or counterparties with investment-grade credit behind them | Letters of intent, pipeline megawatts with no tenant, or revenue that exists only in a presentation |
| Can it fund what it has signed? | Cash, project financing or credit support in hand to deliver the contracted capacity | A build that depends on the next raise, with no backstop |
| Can we check it ourselves? | Contracts, capacity and cash that appear in the 10-Q | A story that lives in the share price or the press |
We do not require GAAP profits. The companies we own today are landlords building ahead of rent, and a landlord with a signed 15-year lease and an unfinished building loses money on paper until the tenant moves in. We require the tenant, the term, and the money to finish the building.
We pass on what we cannot underwrite: anything whose economics we cannot trace through a filing. We do not need a company to be famous, well-liked or in a fashionable sector, and we do not care about the name on the door. Positions are chosen on expected return, and a company that stops earning its spot does not keep it out of loyalty.
How we hold
Concentrated.
A handful of positions, each large enough to matter. We would rather know a few companies deeply than own fifty we cannot follow.
For years.
Each position is sized to be held through a full cycle, including the part where we are early and wrong on timing. A drawdown is not a thesis change. Earnings, contracts, and capacity are.
Sold on the thesis, not the tape.
We add on evidence, and we sell when the thesis breaks or the capital has a better use, not because of a bad month.
Benchmarked honestly.
The fund is measured against the total U.S. market, over years, not quarters. That is the only scorecard we keep.
Holdings are shown for information only and may change at any time. Nothing on this page is investment advice, an offer to sell, or a solicitation to buy any security. Tuuri Capital does not currently manage money for others.