Aschenbrenner bets $7.46B in puts against semiconductors while maintaining AI infrastructure longs

What's new: Leopold Aschenbrenner, the former OpenAI researcher who subsequently ran a fund that reportedly grew from $225 million to $5.5 billion over roughly 12 months, filed his Q1 2026 13F disclosing $13.67 billion in total equity exposure across 42 positions as of March 31, 2026, up from $5.5 billion the prior quarter. The headline change is $7.46 billion in newly initiated put options against the semiconductor complex, a position category that did not exist in his Q4 2025 filing. The 13F is a regulatory disclosure that captures long equity and options exposure but not short selling or derivatives structured off-exchange, so the picture is partial.

The numbers: The put book is concentrated in broad-index and large-cap semiconductor names: SMH VanEck Semiconductor ETF at $2.04 billion, Nvidia at $1.57 billion, Oracle at $1.07 billion, Broadcom at $1.01 billion, AMD at $969 million, Micron at $583 million, TSMC at $535 million, ASML at $494 million, and Intel at $159 million. Simultaneously, his long equity book remains heavily weighted toward power generation, data center operators, and storage infrastructure: Bloom Energy at $878 million, SanDisk at $724 million, CoreWeave at $556 million, and a collection of Bitcoin miners and smaller energy names. He is also running call options on Micron ($422 million), SanDisk ($388 million), TSMC ($354 million), CoreWeave ($140 million), and Bloom Energy ($55 million), making the position more nuanced than a flat bet against chips.

How it works: The structure is a classic pairs trade layered over a thematic macro view. Puts on the chip names that carry AI valuation premiums — Nvidia, Broadcom, AMD, ASML — are offset by calls on the infrastructure and memory names he believes are less fully priced. The long book in power and data center real estate suggests he still expects physical AI buildout to continue; what he is fading is the notion that fabless chip designers and the equipment vendors have further re-rating room.

Why it matters: For practitioners watching AI infrastructure spend, the trade encodes a specific thesis: the bottleneck is shifting from silicon to electrons and square footage, meaning the marginal dollar of AI capex will accrue to energy providers and colocation operators rather than to GPU vendors. This tracks commentary from hyperscalers who have flagged power availability as the binding constraint on new cluster deployments. If correct, valuations on Nvidia and its peers would need to compress even as absolute AI investment continues to grow — a sector rotation rather than an AI bear call.

Caveats: A 13F captures notional option exposure, not net delta or the cost basis of the puts, so the $7.46 billion figure overstates directional risk in a way that headline coverage consistently ignores. Put options can also be sold to generate premium rather than bought as directional bets, and the filing alone does not disambiguate. Attribution of the prior fund performance to Aschenbrenner personally also comes from a single source without independent verification of the fund's audited returns.