One week after TSMC and Samsung posted record quarters on the AI chip boom, the market decided the AI trade has a financing problem. Roughly $1 trillion of market value was wiped from the world’s 20 largest AI and semiconductor companies this week, the iShares Semiconductor ETF (SOXX) fell for an eighth consecutive session — its longest losing streak in 11 years — and South Korea’s Kospi plunged 7.6% in a single day, tripping a circuit breaker that briefly halted trading.

The trigger was not earnings. It was inflation data, a sobering growth disclosure from Anthropic, and the first real distress signal from the debt market funding the AI buildout. For a trade built on the assumption that capital is unlimited, that combination landed like a stress test — and the market failed it, at least for now.

The week’s scorecard: where the trillion went

The damage was global and remarkably synchronized. The table below compiles the reported moves across the AI hardware complex (all figures from market coverage dated Oct 15–19):

Two numbers in that table deserve emphasis. First, SOXX’s −12.7% over eight sessions works out to roughly −1.6% per day, every day, for nearly two trading weeks — the worst such run for the chip ETF in 11 years, according to Dow Jones Market Data. Second, the selloff hit hardest in Asia: SK Hynix’s single-session collapse was about seven times the SOXX’s average daily loss, and the Kospi’s 7.6% plunge forced a circuit breaker. This was not a US tech wobble. It was a global repricing of the AI supply chain.

Trigger one: hot inflation reprices the cost of money

The fuse was lit Wednesday, when September US CPI came in hot at 3.7%, reviving expectations that the Federal Reserve could raise rates again at its October 27–28 meeting. The Nasdaq slid 1.88% on the news alone. Thursday brought the second blow: September PPI rose 6.3% year over year, above the 5.5% consensus and the fastest increase since December 2022.

For most of the year, the AI trade treated interest rates as background noise. Last week the noise became the signal. AI infrastructure is arguably the most rate-sensitive trade in the market: hyperscalers and neocloud operators are funding data-center expansion with borrowed money, and every extra basis point of borrowing cost makes the next gigawatt harder to finance. When September’s inflation prints forced markets to price in tighter-for-longer policy, the financing math behind the buildout cracked first — and chip stocks, the purest expression of AI capex expectations, cracked with it.

(For the full breakdown of the CPI shock, see our Hot September CPI (3.7%) Puts October Fed Hike Back in Play, and for the PPI aftermath, Fed Officials Cool October Hike Bets After Hot 6.3% PPI.)

Trigger two: Anthropic’s growth scare

On October 14, Anthropic told prospective IPO investors that third-quarter revenue reached $16.8 billion — up 45% from the second quarter, but a sharp deceleration from the 145–147% sequential growth it posted in Q2 (revenue: $4.7 billion in Q1, $11.6 billion in Q2, $16.8 billion in Q3). The figures, reported by the Financial Times and Reuters and widely covered, landed just as the company is building the case for a November IPO at a valuation that bankers have discussed at around $2 trillion — more than double its $965 billion private valuation from May.

The absolute numbers remain extraordinary. But markets trade on the second derivative, and the second derivative just bent down — hard. If the fastest-growing AI lab’s revenue trajectory is normalizing this quickly, investors asked the obvious question: what does that imply for the hardware orders, data-center leases, and chip demand that were priced for perpetual acceleration? The disclosure acted as a warning flare across an already jittery AI trade, and Thursday’s chip rout was the market’s answer.

Trigger three: China memory and the neocloud stress test

Two more developments widened the cracks. First, competition: reports that China’s ChangXin Memory Technologies (CXMT) made an architectural breakthrough in memory chips reignited fears that the most profitable corner of the chip cycle — high-bandwidth memory — could face Chinese supply by 2028, a timeline Elon Musk publicly endorsed on X. SK Hynix and Samsung, the HBM duopoly, bore the brunt: both hit their lowest levels since August.

Second, financing distress went from theoretical to visible. Bonds of neocloud provider CoreWeave sank to levels signaling distress — market data cited by CNBC put its 9.75% bond due 2031 at a yield near 15.8%, with five-year credit-default-swap spreads topping 1,100 basis points. Its shares fell 12% on Friday for a 25% weekly loss, the worst week since November 2025. As GAM Investment Management’s Jian Shi Cortesi told the Taipei Times: “Financing costs are like gravity for the AI boom. Rising borrowing costs could make infrastructure projects harder to finance and delay spending.”

The contrast with a week earlier could not be starker. Recall our TSMC and Samsung Post Record Quarters on AI Chip Boom coverage: record revenue, record profit forecasts — and the stocks fell anyway. When great earnings can’t save a sector, the problem isn’t demand. It’s the discount rate.

Bull vs. bear: what each side needs to be right about

The bull case is that this is a rates-and-positioning washout, not a fundamentals break. TSMC and ASML both delivered strong results this month; AI compute demand is still growing; and eight-day losing streaks of this magnitude are rare precisely because they tend to mark capitulation. If inflation cools and the Fed holds on October 28 — now the market’s base case, with CME FedWatch pricing an 83.9% chance of no change — the financing panic could unwind as fast as it arrived.

The bear case is that the financing question is structural, not cyclical. The AI buildout is being funded with leverage at a moment when inflation is re-accelerating, the Fed chair is refusing to rule out further hikes, and memory competition from China threatens the profit pool funding the capex. CoreWeave’s distressed bonds are the canary: if the most aggressive AI borrowers can’t fund themselves cheaply, the entire capex chain reprices.

Both sides agree on one thing: the market has stopped pricing AI infrastructure as a risk-free compounding machine.

What to watch next

China’s data dump, Monday. Q3 GDP plus September industrial production and retail sales are scheduled for release Monday. Consensus expects growth around 4.4%, just below Beijing’s 4.5% annual target — the number will set the tone for Asian risk assets to start the week.

The FOMC blackout. With the October 27–28 meeting now inside the quiet period, no official can walk back market pricing. Chair Kevin Warsh’s Friday appearance at the IMF offered no preview — he said he does not “prejudge” rate decisions. Prediction markets currently put October hike odds near 13–15%, with December priced as the likelier hike (around three-in-four). Any surprise in the weeks ahead reprices the entire AI financing chain.

Big Tech earnings. The rest of the hyperscalers report in the coming days. Watch capex guidance, not revenue: the market now wants proof that AI spending generates returns, not just that it continues.

The SOXX bounce test. After an 11-year-record losing streak, the chip ETF is deeply oversold. A stabilization would suggest last week was panic; a fresh leg down would suggest the financing story has further to run. Either way, CoreWeave’s bonds remain the real-time gauge of whether AI credit stress is contained or spreading.

The AI boom was never going to be a straight line. Last week was the market’s reminder that the line bends wherever the cost of money goes — and right now, the cost of money is going up.

Macrometer is an independent macroeconomics publication. This article is for informational purposes only and does not constitute financial advice.