In five trading days, from 12 to 16 October, the AI trade had its worst week in a long time. The 20 most valuable AI and semiconductor companies in the world lost more than $1 trillion in market value, according to CNBC.
But the number is not the most interesting part of this story. The most interesting part is who was selling and who was buying.
What happened, in numbers
- More than $1 trillion erased from the 20 largest AI and chip stocks in a single week (CNBC).
- The Philadelphia Semiconductor Index fell eight sessions in a row, its longest losing streak since 2008, and ended about 24% below its June peak.
- The Nasdaq Composite lost 4.5% for the week. The Nasdaq 100 had its steepest weekly drop since April 2025.
- The tech sector as a whole fell 7.7%. Applied Materials lost 19%, Dell 23% after a downgrade.
- In Asia the damage was worse: SK Hynix fell 11% in a single day and about 23% from Monday to Thursday, Samsung lost close to 10% in a day, and South Korea's market finished the week down 15%.
- Fear came with it. The VIX volatility index jumped 35%, and the Fear & Greed Index dropped from 45 to 18, "extreme fear".
Meanwhile the broad market barely flinched: the S&P 500 lost 2%, and its equal-weighted version actually rose 0.5%. This was not a crash of everything. It was a crash of the AI trade.
Why: three reasons at once
1. Expensive money. US inflation came in hot at 3.7% for September. The market's odds of a Fed rate hike at the end-of-month meeting jumped to almost a third. AI runs on borrowed money: data centres cost tens of billions, and much of it is financed with debt. When borrowing gets more expensive, every new data centre gets more expensive too.
2. Debt is starting to show. CoreWeave, the cloud company that rents out Nvidia chips, became the symbol of the week. Its 2032 bonds fell to about 78 cents on the dollar, its stock lost a quarter of its value, and its credit default swaps were pricing a high chance of default within five years. Jensen Huang publicly said CoreWeave's cost of capital is "actually decreasing". The market did not listen that week.
3. China is catching up. Chinese memory maker CXMT said it has made progress on cutting-edge memory chips. For SK Hynix and Samsung, which have been minting money on AI memory, that is a direct threat to margins.
Who sold and who bought
This is where it gets interesting.
According to Goldman Sachs prime brokerage data, hedge funds net sold global AI stocks for five sessions in a row, wiping out all the net buying they had done since the start of 2025. And they were mostly selling what they owned, not betting against it: long sales outpaced new short bets by 3.5 to 1. That is professionals taking money off the table.
Retail investors did the exact opposite. SOXL, an ETF that moves three times as hard as the chip index, drew a record of about $3.9 billion from Monday to Thursday, in the same week it lost about a fifth of its value.
So in one week the market split in two: professionals leaving, private investors walking in with triple leverage. One of the two groups will turn out to be right. Which one, nobody knows yet, and anyone who tells you otherwise is guessing.
Bubble or a pause?
Opinions split.
- UBS calls it "a recalibration of expectations" rather than a break in the AI investment story.
- TSMC, the company that actually makes the chips, reported its sixth record quarter in a row in the same week, with net profit up almost 69%. Demand for chips has not disappeared.
- The sceptics point at the debt. When companies build AI on credit and credit gets more expensive, the weakest links break first, and CoreWeave's bonds are a warning sign the market has started to price.
The next big test is the Fed meeting on 27-28 October. If rates go up, the pressure on AI stocks is likely to continue. If the Fed holds, the "buy the dip" crowd will feel vindicated, at least for a while.
What this means if you trade with AI
At NeuPortal we work with AI on real markets every day, and weeks like this one are a reminder of two things.
- Leverage cuts both ways. A 3x ETF that loses a fifth of its value in a week shows how fast a bet on a trend can go wrong. Whatever you trade, know how much you can lose before you start.
- A strong story is not a signal. "AI is the future" can be true and AI stocks can still fall 20% in a week. That is exactly why we publish our AI's forecasts before the event and score them in public against the market, misses included: neuportal.ai/experiment
The bottom line
A trillion dollars left the AI trade in five days. Hedge funds sold, retail bought with triple leverage, and the Fed decides the next chapter on 27-28 October. The AI story is not over, but this week showed that it is not a one-way street.
Discuss it with us in the Telegram community: t.me/neuportal_community
Originally published at neuportal.ai. Not financial advice. Sources: CNBC, CNN, Goldman Sachs prime brokerage data via Atrani Capital, Trending Topics (data for the week of 12-16 October 2026).
Top comments (0)