Lazy Trader AI13F season is the quarterly ritual where institutions disclose their U.S. equity positions. The Q1 2026 filings, which hit the public record by mid-May, were unusually coherent. In most quarters the filings sprawl across themes. This one didn't.
The dominant pattern was concentration: nearly every major filer added to their highest-conviction positions while cutting names they viewed as non-core, too cyclical, or too far from the AI story. And the conviction overwhelmingly pointed in one direction — AI infrastructure: NVIDIA, CoreWeave, TSMC, and the semiconductor and optics layers around them.
When dozens of independent funds all concentrate into the same handful of names, that's information — but it's also a crowding risk. The signal and the warning are the same data point.
Altimeter's infrastructure bet
Altimeter Capital's Q1 filing was a clean example of the concentration theme. The fund:
- Added roughly 15% more NVIDIA.
- Raised its Uber position by about 43%.
- Added to CoreWeave and Taiwan Semiconductor.
- Opened new positions in Arm Holdings and Axon Enterprise.
Read the list and the thesis writes itself: compute (NVIDIA, Arm), the neo-cloud that rents that compute (CoreWeave), and the foundry that fabricates it (TSMC). This isn't a diversified book. It's a concentrated wager on the picks-and-shovels layer of AI.
NVIDIA's own portfolio
One of the more striking filings came from NVIDIA itself, which also files a 13F for its strategic equity stakes. NVIDIA grew its position in neo-cloud provider CoreWeave by 95% — from 24.28 million to 47.21 million shares. At quarter-end that stake was worth roughly $3.66 billion, making CoreWeave NVIDIA's second-largest disclosed holding. It also added Coherent for exposure to the optics layer.
When the company selling the shovels is also buying the miners, the vertical integration tells you something about how durable they think the demand is.
There's a circularity worth noting here — NVIDIA sells GPUs to CoreWeave and holds equity in CoreWeave — and that circularity is exactly what some critics point to when they worry about the AI capex loop. Read the filing both ways.
The doubling-down trade
The clearest “conviction” data point came from Leopold Aschenbrenner's Situational Awareness LP, which reported $13.68 billion in U.S. equity and options exposure for Q1 2026 — more than double the $5.52 billion disclosed at year-end 2025. The fund's reported posture: doubling down on AI and energy infrastructure while hedging the semiconductor layer.
That last clause is the sophisticated part. “Long the theme, hedge the most crowded leg of it” is a very different position than “long everything AI.” A 13F shows the long equity; it does not always show the hedges, which is a reminder that the filing is a partial picture.
How to read a crowded theme
When a theme shows up across this many filers, you have to hold two ideas at once:
- Confirmation. Independent, well-resourced funds reaching the same conclusion is genuine corroboration of the thesis.
- Crowding risk. A trade that everyone owns is a trade that unwinds violently if sentiment turns, because everyone reaches for the same exit.
Crowded trades aren't bad — until they are. The same concentration that drives the theme higher is what makes a drawdown sharper. The Q1 13Fs tell you the smart money is committed; they do not tell you the trade is safe.
Remember the 45-day lag
Everything above describes positions as of March 31, 2026 — disclosed up to 45 days later. By the time you read a 13F, the quarter is two months gone. Some of these positions have been added to; some have been trimmed; the hedges referenced in passing may have grown.
The Q1 13Fs are not a buy list. They're a map of where conviction was concentrated a quarter ago. (We wrote a whole piece on why that lag matters and how to use it anyway — linked below.)
What it means now
- The AI-infrastructure thesis is the consensus institutional position. That's both its strength and its vulnerability.
- Watch the hedging layer. The most sophisticated filers are long the theme and hedging the crowded leg. If you're only long, you're running a different risk profile than they are.
- Track changes, not snapshots. The interesting question for Q2 is who keeps adding and who starts trimming.
- Cross-reference with faster signals. 13F data is the slowest input you have. Pair it with options flow and insider activity for timing.
Q1 2026 13Fs = concentration into AI infrastructure · Altimeter, NVIDIA, Situational Awareness all leaned in · NVIDIA grew CoreWeave 95% · Smart money is hedging the semi layer · Crowded confirmation cuts both ways · Data is 45 days stale — track changes, not snapshots.
This article is for educational purposes and is not financial advice. Past positioning does not predict future results. Lazy Trader AI is a market-monitoring tool, not a brokerage.