Leopold Aschenbrenner may be the most talked-about young investor in the world right now. A former OpenAI Superalignment researcher, he graduated from Columbia at 19 as valedictorian and left OpenAI in 2024 after publishing a 165-page essay called Situational Awareness: The Decade Ahead. That essay argued that AI systems could match top AI researchers by 2027, triggering a compressed path to artificial general intelligence and a historic buildout of power, chips, and data centers.
Since publishing that piece, he has built a hedge fund around it. Situational Awareness LP launched in September 2024 with roughly $225 million in seed capital from Patrick and John Collison, Nat Friedman, and Daniel Gross. Less than two years later, the fund’s disclosed U.S. equity exposure sits at more than $7 billion, and reporting from Fortune and the Wall Street Journal has pegged total AUM (including options and shorts) north of $20 billion.
What makes the portfolio so unusual is what’s not in it. There is no meaningful position in NVIDIA, Microsoft, Google, Amazon, or Meta. Aschenbrenner’s thesis is that the AI boom’s real bottleneck is not model quality. It’s electricity, storage, and neocloud capacity. His book is a concentrated bet on the picks-and-shovels layer that must exist before the software layer can scale.
Below is Situational Awareness LP’s full public equity portfolio, based on its most recent 13F and other regulatory disclosures:
| Symbol | Issuer Name | Shares | Value ($MM) | % of Portfolio |
|---|---|---|---|---|
| NBIS | Nebius Group | 12,410,060 | 2,867.8 | 40.04% |
| BE | Bloom Energy | 6,485,408 | 878.7 | 12.27% |
| SNDK | Sandisk | 1,140,119 | 724.4 | 10.11% |
| CRWV | CoreWeave | 7,177,919 | 556.1 | 7.76% |
| SHAZ | SharonAI Holdings | 5,396,127 | 456.8 | 6.38% |
| IREN | IREN Ltd | 11,698,835 | 401.0 | 5.60% |
| CORZ | Core Scientific | 26,008,473 | 389.1 | 5.43% |
| APLD | Applied Digital | 13,478,438 | 320.0 | 4.47% |
| RIOT | Riot Platforms | 11,502,137 | 142.2 | 1.98% |
| CLSK | CleanSpark | 12,276,139 | 104.5 | 1.46% |
| SEI | Solaris Energy Infrastructure | 1,105,551 | 62.5 | 0.87% |
| TE | T1 Energy | 10,000,000 | 43.9 | 0.61% |
| KEEL | Keel Infrastructure | 19,875,840 | 39.0 | 0.54% |
| BTDR | Bitdeer Technologies Group | 3,439,450 | 29.8 | 0.42% |
| PSIX | Power Solutions International | 432,300 | 26.3 | 0.37% |
| WYFI | WhiteFiber | 1,757,600 | 20.9 | 0.29% |
| AMD | Advanced Micro Devices | 99,138 | 20.2 | 0.28% |
| BW | Babcock & Wilcox Enterprises | 1,353,900 | 19.9 | 0.28% |
| PUMP | ProPetro Holding | 910,300 | 13.1 | 0.18% |
| SMH | VanEck Semiconductor ETF | 26,950 | 10.3 | 0.14% |
| INTC | Intel | 202,344 | 8.9 | 0.12% |
| TSM | Taiwan Semiconductor | 22,423 | 7.6 | 0.11% |
| HIVE | HIVE Digital Technologies | 3,391,547 | 6.3 | 0.09% |
| ASML | ASML Holding | 4,633 | 6.1 | 0.09% |
| MU | Micron Technology | 17,362 | 5.9 | 0.08% |
| GLW | Corning | 5,283 | 0.7 | 0.01% |
| NVDA | NVIDIA | 2,855 | 0.5 | 0.01% |
If you’re curious to see the full holdings of over 10,000 hedge funds, mutual funds, family offices, and institutional investors, you can sign up for free on TIKR. It’s the easiest way to track investor movements, analyze portfolios, and spot ideas early.
The top five positions account for more than 76% of the entire portfolio. Below, we break down each one and why it has risen to prominence in Aschenbrenner’s book.
Nebius Group (NBIS)
Nebius is the fund’s single largest position at a staggering 40% of the portfolio. This is a brand-new stake, disclosed via a 13G filing in May 2026, and it tells you almost everything you need to know about how Aschenbrenner is thinking about the next 24 months.
Nebius is a pure-play AI cloud provider spun out of the international assets of what used to be Yandex. The company operates what it calls an “AI Factory” model, leasing high-end NVIDIA H200, Blackwell, and Rubin GPUs to AI labs and enterprises through long-term contracts. What sets it apart from other neoclouds is the customer-financed CapEx model: multi-year commitments from customers like Microsoft (a reported $17.4 billion deal) and Meta (a roughly $27 billion deal) effectively pre-sell capacity, which lets Nebius fund the buildout without diluting shareholders or piling on debt.
Management has guided 2026 revenue to $3.0 to $3.4 billion and is targeting $7 to $9 billion in annualized run-rate revenue by year-end. NVIDIA also invested roughly $2 billion in Nebius earlier this year, which reinforces the strategic partnership. The company is also aggressively expanding its physical footprint, with a recent UK expansion that adds more NVIDIA-powered infrastructure, additional customers, and new cloud capabilities for agentic and enterprise AI workloads. For Aschenbrenner, Nebius is the most direct public-market bet on the physical AI compute layer he described in his essay.
Bloom Energy (BE)
Bloom Energy makes solid-oxide fuel cells that provide onsite, off-grid electricity. That may sound like an obscure industrial niche, but it turns out to be one of the most direct plays on the AI infrastructure bottleneck. Utility interconnection queues in most U.S. markets now stretch four to seven years, which is far too long for hyperscalers trying to bring gigawatts of compute online in the next 24 months. Bloom’s fuel cells can be deployed in months, not years, and they run on natural gas or hydrogen without waiting on the grid. For a detailed look at the business model and unit economics, this equity research deep dive on Bloom Energy lays out the full case.
Bloom’s own 2026 Data Center Power Report projects that 27% to 38% of data centers will use onsite generation by 2030, up from roughly 1% today, and AI data centers alone will account for 23% of new capacity additions. Revenue crossed $2 billion in fiscal 2025, driven largely by AI data center customers. Situational Awareness trimmed the position by roughly 36% in the quarter, likely booking profits after the stock’s massive run, but Bloom still sits at 12% of the portfolio.
Sandisk (SNDK)
Sandisk is the NAND flash memory business that was spun out of Western Digital in early 2025. On the surface it looks like a legacy storage name, but Aschenbrenner’s thesis is that AI inference workloads are creating an entirely new demand curve for high-capacity, low-latency storage. Training clusters need vast amounts of hot storage for datasets, and inference workloads increasingly rely on KV caches and retrieval systems that hammer NAND. As inference passes 50% of total AI compute this year, storage demand grows alongside GPU demand.
The NAND market has historically been brutally cyclical, but the AI-driven demand shock is tightening supply just as memory makers cut back on capex. Situational Awareness added roughly 8% to its Sandisk position in Q1, and the stock now represents just over 10% of the portfolio. It’s a classic Aschenbrenner move: find a mispriced commodity business sitting directly in the path of the AI capex wave.
CoreWeave (CRWV)
CoreWeave is the neocloud that arguably defined the category. Founded as a crypto miner, the company pivoted its GPU fleet to AI workloads years before the ChatGPT moment and now operates as one of the largest independent providers of NVIDIA-based compute to labs like OpenAI, Anthropic, and Microsoft. Aschenbrenner grew the position by nearly 18% in the quarter, bringing it to just under 8% of the portfolio.
The bull case is straightforward. CoreWeave sits between NVIDIA (its largest supplier and equity partner) and every major AI lab that needs compute but doesn’t want to build data centers themselves. Its backlog and contracted revenue give real visibility, and its early-mover advantage in Blackwell deployments is meaningful. This is exactly the kind of “AI arms dealer to the arms dealers” position that fits Aschenbrenner’s framework. For a deeper look at CoreWeave’s business model and its role in the AI infrastructure stack, this analysis walks through the full thesis.
SharonAI Holdings (SHAZ)
SharonAI is the least well-known name in the top five, but the position size and the growth in shares held (up 578% in the period) makes it clear this is a high-conviction bet. SharonAI is another AI infrastructure and data center operator, and Situational Awareness disclosed a 17.4% stake, well above the 5% threshold that triggers activist-level scrutiny.
The playbook here mirrors the fund’s broader thesis: identify smaller, publicly listed operators with power contracts, land, and interconnection agreements that can be repurposed or expanded for AI compute. These are the assets that would take three to five years to build from scratch. Aschenbrenner is essentially arbitraging the market’s misclassification of these businesses, buying them as legacy or speculative names before the rest of the market re-rates them as AI infrastructure.
The Common Thread: Power and Compute
Zoom out and the shape of this portfolio becomes obvious. Aschenbrenner is betting that whoever wins will need to consume enormous amounts of electricity, storage, and GPU capacity, and that the companies supplying those inputs are dramatically underpriced relative to the demand curve.
That’s why the portfolio holds Nebius and CoreWeave (neoclouds), Bloom Energy and Solaris Energy Infrastructure (onsite power), Sandisk and Micron (storage and memory), plus a dense basket of former Bitcoin miners like IREN, Core Scientific, Applied Digital, Riot, CleanSpark, and Bitdeer. Those miners aren’t in the book because he likes crypto. They’re in the book because they own the exact physical assets (megawatts of contracted power, cooling, substations, land) that would otherwise take years and billions to build.
What makes this book so unusual is the concentration. More than 76% of capital sits in five names, and the top position alone is 40% of the fund. This is not a diversified portfolio. It’s a concentrated expression of a very specific view about what the next few years look like. If Aschenbrenner is right about the compute buildout, these are the businesses that capture the value. If he’s wrong, the same concentration cuts the other way.
Either way, Situational Awareness LP is one of the most closely watched portfolios in the market right now, and for good reason. Very few investors have gone from a 165-page essay to a multi-billion-dollar book in less than two years, and even fewer have positioned themselves this precisely for a thesis they helped popularize.
Quickly value any stock with TIKR’s powerful new Valuation Model (It’s free!) >>>
Want to Know What a Stock Could Return in as Little as 30 Seconds?
TIKR has a powerful Valuation Model tool that shows you how much upside (or downside) a stock could have based on real analyst forecasts.
In seconds, you’ll see:
- What a stock’s fair value is based on revenue growth, margin, and P/E multiple estimates
- Return projections across bull, base, and bear case scenarios
- Whether Wall Street expects the stock to outperform, underperform, or stay flat based on consensus estimates
This tool helps you value stocks smarter in under a minute. No Excel required. No finance background needed.
Click here to sign up for TIKR and try TIKR’s new Valuation Model, completely free.
Looking for New Opportunities?
- See what stocks billionaire investors are buying so you can follow the smart money.
- Analyze stocks in as little as 5 minutes with TIKR’s all-in-one, easy-to-use platform.
- The more rocks you overturn… the more opportunities you’ll uncover. Search 100K+ global stocks, global top investor holdings, and more with TIKR.
Disclaimer:
Please note that the articles on TIKR are not intended to serve as investment or financial advice from TIKR or our content team, nor are they recommendations to buy or sell any stocks. We create our content based on TIKR Terminal’s investment data and analysts’ estimates. Our analysis might not include recent company news or important updates. TIKR has no position in any stocks mentioned. Thank you for reading, and happy investing!