The Ex-OpenAI Researcher Betting $13.7B on AGI Infrastructure — and Against Chips

From OpenAI to Wall Street

Most hedge fund launches attract modest attention. Leopold Aschenbrenner's debut was anything but modest.

Aschenbrenner, a former OpenAI safety researcher who gained widespread attention for his "Situational Awareness" essay predicting rapid AI progress toward AGI, launched Situational Awareness LP in late 2024. What surprised markets wasn't the fund's existence — it was the scale. The fund rapidly grew to $13.7 billion in disclosed U.S. equity and options exposure by Q1 2026, more than double the $5.52 billion it held at year-end 2025.

His Q1 2026 13F filing, posted to the SEC on May 18, 2026, is one of the most unusual institutional disclosures of the year. And it tells a very specific story about where one of Silicon Valley's most prominent AI thinkers believes the money will be made — and lost — in the AGI transition.

The Core Thesis: Electrons, Not Silicon

Aschenbrenner's portfolio rests on a single, powerful conviction: the bottleneck for the AI revolution isn't models, software, or even chips. It's power.

As AI capital expenditures continue to scale exponentially, data centers are bumping into a hard constraint — the availability of reliable, large-scale electricity. Aschenbrenner has apparently decided that this infrastructure choke point is the most important investment theme of the decade.

This shows up clearly in his long book:

• CoreWeave — the AI-native cloud provider building massive GPU clusters for model training and inference
• Bloom Energy — a fuel cell power company; Aschenbrenner initiated this position in Q4 2025, accumulating 10.1 million shares worth approximately $875 million, before the stock ran 176% higher
• Crypto miners — used as proxies for large-scale power consumption and data center infrastructure

These aren't household names in most institutional portfolios. They're picks from someone who believes the physical infrastructure layer of AI is systematically undervalued.

The Other Side: $8.47 Billion in Semiconductor Puts

Here's where the filing gets genuinely surprising.

While Aschenbrenner goes long on power and compute infrastructure, he's placed an enormous hedge — or outright bet — against the semiconductor layer. His fund holds $8.47 billion in notional put exposure across a wide basket of chipmaker stocks, including:

• $2.0B in puts on the VanEck Semiconductor ETF (SMH)
• $1.6B in puts on Nvidia (NVDA)
• Additional put positions against Broadcom (AVGO), AMD, Micron (MU), ASML, Intel (INTC), Taiwan Semiconductor (TSM), Oracle (ORCL), and Corning (GLW)

This isn't a diversified hedge. It's a concentrated view that the chip layer — despite the AI boom narrative — is overvalued relative to the infrastructure layer beneath it. The logic: Nvidia and its peers have captured enormous speculative premiums, while the companies actually delivering the power and compute capacity to run AI at scale remain under-appreciated.

Whether you agree or not, it's a coherent, thesis-driven position that stands out sharply from the crowded consensus long-AI trade.

What 13F Filings Reveal About Conviction

One thing that makes Aschenbrenner's filing worth studying is the concentration. With 42 disclosed holdings but a dominant options position, this is not a diversified multi-strategy book. It's a directional macro call expressed through carefully structured positions.

This is exactly why 13F filings remain one of the most valuable free datasets available to retail investors. They show not just what institutional managers own, but how they're sizing it and which layer of the value chain they're targeting.

You can explore similar patterns in real time using InvestorLens's institutional flow tracker, which aggregates 13F data across hundreds of top funds. Or compare how different managers are positioned across AI-adjacent sectors using the portfolio overlap tool — useful for spotting whether a thesis is crowded or still contrarian.

The Macro Signal

Aschenbrenner's positioning aligns with a broader pattern visible in aggregate 13F data: top-tier institutional investors are moving down the AI stack, away from the application and chip layers, toward physical infrastructure — power generation, data center real estate, networking, and compute-as-a-service providers.

You can track how this consensus is evolving across the entire institutional landscape at InvestorLens Macro Consensus, which synthesizes sector positioning signals from leading 13F filers.

Key Takeaways

Aschenbrenner's Q1 2026 13F is a masterclass in expressing a macro thesis through asymmetric positioning. The fund is:

• Long the physical infrastructure enabling AI (power, compute, data centers)
• Short the semiconductor layer it believes is overpriced relative to that infrastructure
• Concentrated in a conviction-driven structure, not diversified for diversification's sake

Whether this pays off depends on whether the power bottleneck thesis plays out as forecast. But the filing offers a rare window into how someone deeply embedded in the AI research world is translating that knowledge into a portfolio.

That's the edge 13F data provides — not just tracking what's popular, but finding the managers with genuine informational advantages and understanding their reasoning.

Explore all disclosed institutional holdings and sector trends at InvestorLens.

---

Data sourced from public SEC 13F filings. Educational research only — not investment advice.