How hedge funds are reading the AI markets.

Aschenbrenner's book grew from $5.5B to $13.7B QoQ as he scaled into Q1 2026. For each name in his book, we tracked how every other active filer in our universe moved on it.

Where consensus caught up: BE, SNDK, CRWV. His top picks-and-shovels bets are now also crowded by 12+ active filers, with aggregate dollars up double digits this quarter.

Where he is increasingly alone: IREN (others -14%), APLD (others -25%), RIOT (others -35%). He sized up these positions while everyone else cut. The bet is that *these specific* miners and small-cap infra names are the survivors of the AI compute buildout.

New starter positions in Q1: NVDA, TSM, MU, ASML, GLW, AMD. Small (<$50M) toehold positions in mainstream semis. Either a hedge against his concentrated infra book or a soft thesis pivot toward chip incumbents.

Net flows across our filer universe for the AI book, Q4 2025 to Q1 2026. The picks-and-shovels names led: NVDA, AAPL, AMZN, GOOG.

Biggest add was NVDA (+$187.2B), held by 37 of our filers. Biggest cut was WYFI (−$14M).

Coatue's AI book in early 2023 was the obvious longs: NVDA, MSFT, META, TSLA, LRCX. Their NVDA position has been trimmed actively while the stock climbed. The names they added since are dateable.

• Q3 2023VRT$430M$246M
• Q4 2023AAPL$471M$471M
• Q1 2024AVGO$552M$1.7B
• Q1 2024CEG$379M$1.3B
• Q1 2024DELL$482M$1.0B
• Q1 2024QCOM$620M$209M
• Q1 2024VST$224M$409M
• Q4 2024SMCI$270M$304M

The thesis crystallized in Q2 2023 when Eaton and Equinix entered, six to nine months before mainstream coverage discovered the AI power bottleneck. ASML was the newest addition in Q1 2026.

For each fund and ticker, we computed the correlation between quarterly position changes and insider net flow in that same quarter. We required at least 30 paired observations per fund.

Renaissance at +0.36 is the most insider-aligned. ARK at +0.24 is the surprise, since they are a pure-quant shop.

At the bottom, Whale Rock at -0.27 systematically buys when insiders sell. The passive giants cluster near zero, as expected.

Magnetar Financial structured chunks of CoreWeave's pre-IPO debt. Since the unlock window opened, they have sold $5.6B across 1,715 transactions between 2025-08 and 2026-05.

Our 13F universe now holds $6.1B of CRWV across 27 filers, a near-mirror of what Magnetar sold. But 70% of that institutional book sits in passive index funds, who had to buy once CRWV joined the indexes.

Active conviction is much narrower: Situational Awareness $556M, Altimeter $349M, Fidelity $251M, ARK $197M. The unlock is going to index plumbing, not to discretionary capital betting on the name.

From 2023 through 2025, Berkshire's 13F shows zero AI or datacenter names. Their only prior AI holding was Taiwan Semiconductor, held briefly in Q3 and Q4 2022 and sold before any of NVDA's 5x run.

Then in Q1 2026 they opened a position in Alphabet worth roughly $1.0B. The most respected value investor on Earth sat out the entire cycle, then bought the cheapest of the megacap AI names.