Leopold Aschenbrenner's Situational Awareness LP filed its Q1 2026 13F on May 18, and the disclosed equity exposure jumped from $5.5 billion at year-end 2025 to $13.67 billion as of March 31. The single largest concentration in that book is not Nvidia, not the hyperscalers, and not a pure AI software name. It is Bitcoin miners and AI data center operators, with confirmed long positions in IREN, CleanSpark, Riot Platforms, Applied Digital, Core Scientific, and Hive Digital. The same filing shows roughly $7.5 billion in notional put exposure against semiconductor stocks including Nvidia, AMD, ASML, and the VanEck Semiconductor ETF.
The ex-OpenAI researcher who wrote the most influential AGI scaling essay of 2024 is now telling the SEC that the cheapest way to long AI compute in 2026 is to own the companies that already control cheap power and cooled real estate. Here is what the filing actually shows, why miners are the trade, which names are in the basket, what it means for BTC hash rate, and the bear case that has to be taken seriously.
The $5.5 Billion to $13.67 Billion Build and What the Filing Shows
Situational Awareness LP's prior 13F, covering year-end 2025, disclosed $5.5 billion in equity exposure. The Q1 2026 filing, accepted by the SEC on May 18 and visible on the 13F manager dashboard, shows that number sitting at $13.67 billion as of March 31. That is roughly a 2.5x expansion in one quarter, and it happened during a period when broad AI equities were under pressure from the early-2026 rotation.
The headline split inside that book is unusual. Long positions concentrate on what most equity investors would call second-derivative AI plays: Bitcoin miners with disclosed AI hosting pivots, data center operators, and energy infrastructure (Bloom Energy is actually the single largest long, with 6.5 million shares worth roughly $879 million). The short side concentrates on the chipmakers themselves, with $7.5 billion in notional put exposure spread across Nvidia, Broadcom, AMD, ASML, Micron, Intel, and Taiwan Semiconductor.
That construction matters because Aschenbrenner is not bearish on AI itself, only on the consensus way to express it through chips. The "Situational Awareness" essay he published in 2024 argued that AGI-class systems would require trillions of dollars of compute infrastructure by the late 2020s. The portfolio is the trade that flows from that thesis. The full essay is still publicly available at situational-awareness.ai, and the May 18 reporting via CoinDesk's coverage of the filing is the clearest summary of how the disclosure mapped to specific tickers.
Why Miners and Not Nvidia: The Picks-and-Shovels Thesis Explained
The standard equity playbook for AI exposure is to own Nvidia and the hyperscalers (Microsoft, Meta, Alphabet, Amazon). That trade has worked for three years, and consensus is positioned for it to keep working. Aschenbrenner's filing is a direct bet that the next leg of the AI cycle moves down the value chain.
The argument runs like this. Training and running frontier models is a power problem more than a chip problem. By 2027 the bottleneck on US AI buildouts is projected to be megawatts of available capacity at competitive prices, not the H100 or B200 supply. Bitcoin miners already own the scarce resource: roughly 30 gigawatts of permitted, energized, cooled, and grid-connected industrial sites in low-cost US power markets (West Texas, Wyoming, Georgia, upstate New York). Building equivalent capacity from scratch takes three to seven years for permitting alone.
Three structural facts make miners the cleanest way to express this view. Miner sites are interchangeable, since an ASIC hall converts to a GPU hall with retrofit capex rather than greenfield construction. Miner equity trades at a fraction of the multiple of pure-play AI infrastructure. And the AI hosting revenue is contracted, recurring, and dollar-denominated, which rerates the equity from "BTC beta proxy" to "data center REIT with optional crypto upside."
The November 2025 IREN-Microsoft $9.7 billion GPU cloud agreement deploys Nvidia GB300 GPUs at IREN's 750MW Childress, Texas campus over five years. The contract is expected to deliver around $1.9 billion in annual recurring revenue at roughly 85% EBITDA margins. No pure-play AI infrastructure name is signing contracts at those terms, because no one else has the power.
The Specific Miners in the Basket
Reporting on the 13F has confirmed the following miner and AI-adjacent infrastructure longs. The list below reflects names that are either explicitly disclosed or that increased from the prior quarter:
Ticker | Name | AI hosting status | Why it is in the basket
IREN | IREN Limited | Live $9.7B Microsoft contract, GB300 deployment underway | Most-advanced miner-to-AI pivot in the public market
CLSK | CleanSpark | Pure-play BTC miner, exploring AI hosting | Cheapest BTC-per-share, lowest debt among large miners
RIOT | Riot Platforms | Rockdale TX site under evaluation for AI tenants | Largest disclosed power capacity in the US
CORZ | Core Scientific | CoreWeave hosting contract live | Already monetizing AI tenants on bankruptcy-emerged balance sheet
APLD | Applied Digital | Pure AI hosting, no BTC mining | Cleanest AI-data-center pure play
HIVE | Hive Digital | Toronto AI data center buildout disclosed | First public miner to declare an AI-first pivot
BITF | Bitfarms | Power capacity across Argentina, Quebec, Washington State | Geographic diversification on cheap hydro
BTDR | Bitdeer | Singapore-listed, mixed mining and AI hosting | Asia exposure inside the thesis
The basket is not random. Every name on it either already has an AI hosting contract or owns the kind of power capacity that makes one possible. Several pure-play miners that lack the power footprint to pivot (smaller US-listed names, some international miners) are conspicuously absent.
On the short side, the put book is the opposing trade. If miners and energy infrastructure outperform, the implicit view is that chipmakers compress because more of the AI dollar moves to the power-and-real-estate layer of the stack. That compression is the open question the trade depends on.
What This Means for BTC Hash Rate and Miner Stock Prices
For crypto traders, the second-order effects are more interesting than the equity trade itself, and three things start to matter beyond the share-price action.
First, hash rate. If a non-trivial share of miners with AI optionality (IREN, HIVE, CORZ, parts of RIOT) divert ASIC capacity toward GPU hosting, network hash rate growth slows or even contracts. That has happened twice already on smaller scales since the 2024 Bitcoin halving, and Phemex's hash rate explainer covers why a flatter hash curve historically supports BTC price during cycles where new supply is constrained.
Second, the miner-equity-to-BTC correlation. Through most of the 2020-2024 cycle, the large miners traded as a leveraged BTC beta, with a roughly 1.5x to 2.0x daily beta to spot. Once AI hosting revenue becomes a material share of the income statement, that correlation decouples. IREN already trades partly on Microsoft contract milestones, not BTC price. CORZ trades partly on CoreWeave volumes. The pure-play miners (CLSK, BITF) still trade on BTC. Traders running miner-stock baskets need to start splitting the universe by AI hosting share of revenue rather than treating it as one bucket.
Third, the Bitcoin mining cost curve flattens as AI hosting revenue subsidizes the marginal BTC produced. A miner with 30% of revenue locked into a Microsoft-style contract can keep mining at hash prices that would shut down a pure-play competitor. That extends the marginal cost floor down, which historically has been one of the strongest reflexive supports for BTC price during drawdowns.
For traders, the practical read is straightforward. Watch the miner basket as a leading indicator for the AI-power narrative, watch hash rate for confirmation that the pivot is real, and treat the BTC-miner equity correlation as a thesis that will keep weakening over the next twelve months.
The Bear Case to the Thesis
The trade is not obvious and the bear case is real, with four pressure points that deserve attention before sizing the thesis.
The AI compute boom may not stay supply-constrained. If chip lead times normalize and US permitting reforms accelerate greenfield data center buildouts, the scarcity premium on existing miner sites compresses. The chip puts assume demand weakens too, and if demand stays strong while supply catches up, the shorts lose money faster than the miner longs make it.
Miner balance sheets are still messy. Most public miners ran cash-burn operations for years and only recently turned free-cash-flow positive. AI hosting contracts require significant capex upfront for GPU procurement, retrofit, and liquid cooling. The financing path tends toward more equity raises and convertible debt, both of which dilute existing holders.
The BTC price tail risk has not disappeared. Even the most AI-pivoted miner still has BTC exposure on the balance sheet and in the legacy ASIC fleet. A 40% BTC drawdown still craters reported earnings across the entire group, regardless of how much Microsoft revenue is contracted.
Concentration risk inside the basket. A handful of names (IREN, CORZ, APLD) drive most of the disclosed AI hosting economics today. If any one of those contracts gets repriced, delayed, or canceled, the whole narrative gets challenged in a single news cycle.
The honest read is that this is a high-conviction, high-concentration trade with asymmetric payoff if the power-bottleneck thesis is right. Aschenbrenner is putting his book on the line on a specific prediction about which layer of the AI stack captures the next wave of margin, and that prediction may be early, late, or wrong.
Frequently Asked Questions
Is Aschenbrenner actually long Bitcoin itself or just miners?
The 13F discloses only US-listed equity and option positions, not direct crypto holdings. The disclosed exposure is to mining companies and AI infrastructure, not to spot BTC. If Situational Awareness holds BTC directly through other vehicles, that exposure is not public information.
Which miner has the largest AI hosting contract today?
IREN's five-year, $9.7 billion deal with Microsoft, signed in November 2025, is the largest publicly disclosed contract in the sector. Core Scientific's CoreWeave hosting arrangement and Applied Digital's pure-AI buildout are the next most material in revenue terms.
Why is Bloom Energy in the long book if the thesis is about miners?
Bloom makes on-site fuel-cell power systems that data centers and miners increasingly use to bypass utility-grid interconnect queues. The position is a direct expression of the "power is the bottleneck" view, sitting one layer upstream of the miners themselves.
Could BTC hash rate actually drop if miners pivot to AI?
It could. Roughly 30% of current network hash rate sits with public miners that have disclosed AI hosting optionality. A sustained pivot of even 10% of that capacity would slow hash rate growth materially, and would historically have been supportive for BTC price by reducing marginal sell pressure from miners.
Bottom Line
Aschenbrenner's filing reframes the AI trade away from chips and toward power. The disclosed long basket bets that Bitcoin miners are the cheapest publicly-listed way to own AI compute capacity, because they already control the scarce resource. The short basket bets that chipmakers compress when more of the AI dollar moves downstream. The two-sided construction is what makes the trade interesting: it does not require AI to fail, it requires the margin to relocate. Watch IREN's next contract milestone, watch hash rate for confirmation, and watch the put book on Nvidia for the first real test of the thesis. If chip earnings hold up through Q3 and miner AI revenue underwhelms, the trade is wrong. If the opposite happens, this 13F will be cited as the first major institutional disclosure of a structural rotation.
This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency trading involves substantial risk. Always conduct your own research before making trading decisions.
