The biggest tech de-risking since 2016 was a leverage flush, not a verdict on AI

When a fund is forced to sell everything at once, the print tells you about the seller, not the assets. That is the frame for last week's AI shakeout.

The scale was historic. Over July 24, 27 and 28, hedge funds sold global information technology stocks harder than in any three-day stretch since Goldman Sachs' prime brokerage desk started tracking in 2016. The exclamation point was Situational Awareness, an AI-focused fund built by former OpenAI researcher Leopold Aschenbrenner, selling its entire public equity portfolio to Citadel after its leveraged AI positions dropped. Its largest position was the semiconductor ETF SMH, and six single names went out the door: NBIS, BE, SNDK, CRWV, SHAZ and IREN.

The fundamentals did not crack alongside the price. Goldman's read is that this is leverage reduction and a crowded-trade unwind, not deterioration. The scoreboard backs that: 64% of reporting S&P 500 firms beat EPS by at least a standard deviation, core earnings grew 26% year over year, and the median company grew 12% against 9% expected.

Leadership stayed where the cash is. With the sole exception of Meta, the other three major hyperscalers outperformed both the S&P 500 and SMH over the last five days and the last month. The names closest to monetized AI revenue held up while the leveraged, second-derivative bets were the ones liquidated.

Why it matters (allocator lens): forced selling and fundamental breakdown look identical on a one-week chart and are opposite things underneath. A record prime-desk unwind into a strong earnings tape is the profile of a positioning washout, which historically resolves up, not the profile of a growth scare.

So-what for ABC: the dispersion is the instruction. Beta through the index or through the hyperscaler-heavy complex captures the part of AI that is already converting spend to revenue, while the SMH and single-name high-beta wreckage is where the leverage lived. This is where Cash as dry powder earns its keep, because more forced selling can still overshoot before it clears.

Takeaway: the biggest tech de-risk since 2016 was about who was positioned, not whether AI works. Watch prime flow, not the one-week candle.

Analytics & education, not advice. DYOR.