A 24-year-old ex-OpenAI researcher just turned $225M into over $13.67B in under 2 years. And his portfolio just revealed something even more extreme than his returns. Leopold Aschenbrenner was fired from OpenAI in April 2024. After that, he wrote a 165-page thesis predicting AGI by ~2027. Then he launched a fund and did something unusual: He fully positioned around that thesis. He initially avoided the obvious AI winners: Zero $NVDA Zero $MSFT Zero $GOOGL Zero $AMZN Instead, he targeted what AI physically runs on. His early “AI infrastructure” longs included: • Bloom Energy $BE • Lumentum $LITE • SanDisk $SNDK • CoreWeave $CRWV • Iris Energy $IREN The thesis was simple: AI isn’t just software. It’s constrained by: • power • bandwidth • storage • compute infrastructure And those bottlenecks were massively mispriced. The results were explosive: • Bloom Energy: +1,422% • Lumentum: +1,331% • SanDisk: +3,130% • IREN: +583% • CoreWeave: +166% This is what turned his initial $225M into ~ $5.5B by end of Q4 2025. Fast forward to his latest SEC filing (Q1 2026): His disclosed exposure has surged to $13.67B equivalent across 42 positions. A near 3x jump in a single quarter. But the structure of the portfolio changed dramatically. He didn’t just stay long AI infrastructure. He built a two-sided portfolio; Massive bearish positioning on semiconductors (puts totaling ~$7.46B): • $SMH ETF PUT: $2.04B • $NVDA PUT: $1.57B • $AVGO PUT: $1.01B • $AMD PUT: $969M • $MU PUT: $583M • $TSM PUT: $535M • $ASML PUT: $494M • $ORCL PUT: $1.07B • $INTC PUT: $159M At the same time, he STILL holds long exposure to the AI infrastructure stack: • $BE : $878M • $SNDK: $724M • $CRWV: $556M • $IREN: $401M • $CORZ: $389M • $APLD: $320M • $RIOT: $142M • $CLSK: $104M • $SEI: $62M • $TE: $43M • $KEEL: $38M • $BTDR: $29M • $PSIX: $26M • $WYFI: $20M • $BW: $19M • $SHAZ: $18M • $PUMP: $13M • $HIVE.NE: $6M He also added CALL OPTIONS on select names: • $MU CALL: $422M • $SNDK CALL: $388M • $TSM CALL: $354M • $CRWV CALL: $140M • $BE CALL: $55M So the positioning is not a simple “AI is over” trade. It’s more specific: He still believes AI infrastructure expands aggressively… …but thinks semiconductor leaders may have pulled forward too much optimism. In other words: He is long the physical buildout of AI and short the market’s most crowded AI expectations at the same time. From $225M → $5.5B → $13.67B… The real signal isn’t just performance. It’s that his view of AI has evolved from: “AI wins” to “the winners of AI may not be who the market thinks.” Are you going to ignore him again?
