Invest1 distinct publisher3 min readPublished
Leopold Aschenbrenner's fund went from $5.52bn at the end of 2025 to $13.67bn in ex-miners and power by March, against roughly $7.46bn of puts on semiconductors, which prices interconnect queues above silicon.
The Investor · Invest desk
Compiled by The InvestorSomething wrong?How this is made
Fifty-five cents of disclosed short semiconductor exposure for every long dollar [14] is the structure worth staring at, because it functions as a position on a spread rather than a market hedge: contracted megawatts reprice up while chip multiples reprice down [5][8]. Both legs only pay if those two things separate. If AI capex disappoints across the board, the miners fall alongside the chipmakers and the puts refund a fraction of the damage on a book that added $8.15bn of equity exposure in one quarter [13]. Sitting underneath the whole calculation is a disclosure problem: Crypto Briefing's account does not say whether the $7.46bn is premium paid or notional [16], and that decides whether the short leg is a rounding error against the long book or half of it.
What the miners are selling, on this reading, is elapsed time. Power purchase agreements negotiated over years, industrial sites chosen for cheap electricity, cooling and electrical plant already standing [6], with the rig-for-GPU swap treated as the easy part [6]. Core Scientific is the worked example, out of a late-2022 bankruptcy and into multi-year AI hosting contracts inside four years [7]. The corollary defines the limits of that $13.67bn position [1]: it excludes the silicon it is short [8] and the model layer above it.
Then there is Bloom Energy in the same book [4], making the on-site fuel cells that let a data centre skip the utility queue entirely [9]. Own the scarcity and own the cure to it, or rather, the more interesting version, own the cure because you know the scarcity has a clock on it. Every greenfield site with a gas connection is a competitor to an interconnect position, which caps what a permitted brownfield site can charge at roughly the cost of the alternative.
The drawdown says something about the shape of the trade rather than the merits of it. Sharp mid-2026 losses amid AI-linked volatility, large enough to force a major liquidation of public equities through a transaction with Citadel [11], and the account does not break out which leg lost [11]. A long-miners, short-chips book losing money in an AI selloff implies the miners fell at least as hard as the semiconductors did [8][11], which is the specific failure mode here: in a drawdown, the real assets amplify the underlying story rather than diversifying away from it. And you do not sell a concentrated block to one balance sheet unless the screen cannot absorb it, which is a cost paid for immediacy [11].
This is probably wrong, but the thesis looks more like a duration bet than a scarcity bet. Chip lead times can shorten with fab output; a permitted 200-megawatt site with signed power cannot be conjured on the same schedule [5]. The falsification test is the one the source itself names, contracted megawatts for AI workloads and the margin between mining a coin and renting a rack [12]: if that margin compresses toward what a new build with on-site generation costs [9], the miners are worth replacement cost and the option value goes to zero. Track the megawatts; the hash rate is a distraction.
Ranked by verification strength, evidence, and original report placement.
Situational Awareness LP, the hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, held $13.67 billion in equity exposure as of March 31, 2026.
The fund's equity exposure was $5.52 billion at the end of 2025.
Aschenbrenner is 24 years old and a former OpenAI researcher.
Top holdings prominently feature IREN, Core Scientific, Riot Platforms and CleanSpark, alongside energy firm Bloom Energy and compute company CoreWeave.
The fund's thesis is that chips are no longer the scarce resource and that the real constraints on scaling AI are power and land: securing grid connections, building substations, and getting site permits.
Core Scientific has begun the transition, announcing AI hosting arrangements including multi-year contracts, having gone from bankruptcy in late 2022 to one of the most talked-about AI infrastructure plays in under four years.
Distinct publishers with included, body-backed reporting in this cluster.
cryptobriefing.com
1 article · August 30, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Citadel cleared $4bn of a rescued AI book in three weeks. The discount was the trade.1 distinct publisher
invest
Half a book in two memory names: what Situational Awareness's 67% month actually shows1 distinct publisher
product
Two auctions, one week: the disclosure gap buried in Situational Awareness's $30bn July1 distinct publisher
invest
Public miners are unplugging hash faster than the network, and rebilling it as colocation1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single account, no document behind the numbers
Every dollar figure in this story — $13.67bn, $5.52bn, $7.46bn — comes from Crypto Briefing and from nowhere else, with no filing quoted, no as-of document named and no fund comment. The arithmetic is internally consistent, which is not the same as checked. Two of the more consequential statements, the miners' claimed conversion head start and the Citadel liquidation, arrive with no site, contract or loss number attached at all.
Conversion is happening; nobody has sized it
There are real instances behind the thesis: Core Scientific signing multi-year AI hosting deals after a 2022 bankruptcy, CoreWeave's move from mining to GPU cloud. What is missing is scale. Crypto Briefing itself nominates megawatts under contract as the metric that now matters, then reports not one megawatt figure, contract value or energisation date — and the only other hard adoption signal, a quarter-end position disclosure, measures a fund's conviction rather than the industry's build-out.
Overstated, mostly by ordering and by unit
Two things inflate this. First, 'one of the most provocative bets in recent market history' sits in the opening paragraph while the losses sit under the last subhead, unquantified — the drawdown is the outcome of the trade being described, not an afterthought to it. Second, adding a put figure of unknown basis to a long book produces a $21.13bn headline that may be comparing premium with market value. Strip both and what remains is a large, plausible, entirely unverified position report.
A crypto desk narrating crypto's second act
The people who benefit from this thesis being believed are the ones supplying it: a fund whose disclosed longs are repriced upward if power, not silicon, is accepted as the bottleneck, reported by a publication whose beat is the same miners now recast as AI infrastructure. To its credit, Crypto Briefing does not suppress the drawdown — it just leaves it unmeasured. No conflict disclosure or fund comment appears either way.
Coherent story, thin footing
Our confidence tracks the sourcing, not the plausibility. The power-and-land argument hangs together and the named companies are real, but one crypto outlet, no primary document, an ambiguous unit on the short leg and a numberless account of the losses leave little room to be firm about anything except that the thesis was expressed with real money.