The AI bubble bursts on Wall Street: US chips fall more than 20% since July
US chips accumulate drops of more than 20% since July: Wall Street demands evidence of return to artificial intelligence.
Eight sales sessions and one uncomfortable question on the trading floor
The correction does not come from bad business data. It comes from a question that until three months ago no one asked out loud on the equity desks: when are the billions committed in hardware to train models going to return money. The semiconductor sector has been losing steam for four months and sales intensified during the first half of October, according to analysis published by the Washington Examiner.
The most eloquent detail is technical. Until October 15, the sector’s main exchange-traded funds—ETFs, baskets of securities that are bought and sold like a stock—chained eight consecutive sessions of losseswith accumulated declines of around 12%. There was no collapse of orders. There was profit taking and very concentrated positions that amplified the fall.
Wall Street has no patience.
The arithmetic that concerns us is textbook. In this phase of the cycle, the investment easily exceeds the return: it is the moment of maximum cash burn. The question is not if AI will generate value, but when it will return it in cash form. And in a market that discounts five-year profits, “when” is the only variable that is quoted.
Anthropic opens the books: 1,400% growth and 800,000 million more to spend
This same week the antidote arrived. Anthropic filed its S-1—the registration prospectus that a company submits to the SEC before going public—with annual revenue growth of more than 1,400% at the close of its June quarter and a positive operating margin of 19%. The company plans to debut on November 16 with a valuation of between 1.5 and 2 trillion dollars, and warns that it plans to invest another 800,000 million in its operation. Ecosystem values rebounded, with Nvidia and Micron at the head.
The market does not discuss the potential of artificial intelligence. Discuss whether the capital that supports the deployment can withstand the wait.
The rest of the complex remains in the expensive phase. Goal has just announced a capital increase of around 95 billion to finance its investment in AI, and Microsoft offers voluntary redundancies to 60,000 employees while trying to sustain the investment effort without damaging the balance sheet. Nobody disputes the potential. The consensus of large investment houses places the crossover—incremental cash flow above spending growth—at the end of 2027, and the return on capital at 12%-14% towards 2030-2031.
Washington Logic
In Washington, no one has the intention of setting the price of technology companies. The lever of administration donald trump It is another: the energy that powers data centers, grid connection permits, chip supply, and export controls. The stated priority is the speed of deployment, not the multiple of shares, and behind it is a recognizable coalition: the states competing to attract data campuses, the manufacturers that benefit from the CHIPS Act —the 2022 law that subsidized semiconductor factories on American soil— and an electorate that reads AI as industrial employment rather than as a bubble.
It’s happened before. Between 1999 and 2001, the dot-com bubble wiped out companies that had built entire fiber optic networks, and the Federal Reserve He did not come out to support valuations: he left the capital to purge itself. amazon It was listed at $106 in December 1999 and, adjusted for the 20-for-1 split in 2022, was worth 30 cents in September 2001. Whoever lasted two decades turned $1,000 into more than $90,000. The reading is not that every fall is an opportunity: it is that the cycle is measured in years.
In Spain the transmission channel is double. The first is financial. The funds indexed to the Nasdaq and the S&P 500 marketed by Spanish banks and management companies, under the supervision of the CNMVhave channeled retail savings into American Big Tech: a sustained correction of the chips reaches the Spanish participant before the income statement of any Ibex company.
The second channel is industrial and energy. The electrical demand of data centers already appears in the plans of Iberdrola and Endesaand the projects announced in Aragon, Madrid and Catalonia—several linked to American hyperscalers, the large data center operators—depend on someone continuing to pay for computing.
The path of balance remains. Santander and BBVA They have incorporated AI into their three-year plans because of the cost, not the multiple: if the hardware gets cheaper, they win; If the cycle gets stuck, the bill comes in the form of lower technological spending and clients with less investment appetite.
The next window is dated. The IPO of Anthropicscheduled for November 16, will be the first real appetite test since the October corrective. If the market receives it well, the debate on the artificial intelligence bubble is postponed for a quarter. If it is received poorly, the rise in global capital prices will also reach Spanish debt, which is financed in a market where the risk premium does not distinguish between a data center in Aragon and a Treasury bond.
Case File
- The case: The US semiconductor sector enters a correction after four months of doubts about the return of spending on AI, and the IPO of Anthropic becomes the thermometer of global investment appetite.
- Key data: Fall of more than 20% in chips since July; eight consecutive selling sessions until October 15 with a 12% decline in the sector’s ETFs; Anthropic revenue growing at 1,400% with operating margin of 19%; exit valuation of $1.5-2 billion; 8 trillion investment in AI in five years, with a return of 12%-14% not before 2030-2031.
- For Spain: The impact comes in two ways: index funds that channel retail savings towards American technology and the electrical demand of data centers that supports the plans of Iberdrola and Endesa.
