The Nasdaq fell as semiconductor stocks extended a global selloff. Micron dropped about 5%, Applied Materials lost more than 4%, and Intel, Super Micro Computer and Western Digital also retreated. Nvidia declined more modestly, but the symbolism was hard to miss after its market value slipped below Apple's. The Philadelphia Semiconductor Index is now more than 20% below its June record, the usual definition of a bear market.

This is the fifth serious wobble in the AI trade in three years. It is also the most important because investors are asking who will make money from it, how much financing will be required and whether the industry is building too much capacity too quickly.

The semiconductor industry has always moved in cycles. Demand rises, factories operate near full capacity, margins expand and manufacturers rush to build more production lines. Eventually, that new capacity arrives. If demand has slowed even slightly, shortages turn into surpluses and prices fall. The market has spent the past few years assuming that AI would escape this familiar pattern. The argument is understandable. Demand for computing power is soaring, order books remain full and the largest technology companies are spending hundreds of billions of dollars on chips, servers and data centers. But a technological revolution does not repeal the laws of supply and demand. An oversupply can emerge whether demand is modest or enormous. Supply simply has to grow faster.

South Korea offers the clearest warning. Its KOSPI index, one of the markets most closely tied to the AI hardware boom, has fallen 25% since July 1. On Tuesday alone, it lost about 10%, with SK Hynix and Samsung Electronics leading the retreat. Taiwan fell 4.5%, while Japan dropped 4.3%.

Chinese competition is adding to the anxiety. According to The Information, Chinese manufacturers are rapidly improving the equipment used to produce semiconductors. That threatens an industry long dominated by a small group of American, Dutch and Japanese suppliers.

China is also challenging the economics of AI software. Moonshot AI's new Kimi K3 model appears powerful, relatively efficient and, crucially, open source. The more capable models become widely available at low cost, the harder it will be for leading developers to protect their prices and profit margins.

This does not mean the AI boom is ending. Demand remains strong, and no one knows precisely when the cycle will turn. But investors are beginning to recognize that rapid adoption and attractive returns are not the same thing.

The second concern is financing. For much of the AI boom, the largest technology companies could fund expansion from their enormous cash flows. Alphabet, Amazon, Meta and Microsoft generated so much money that data centers seemed almost affordable. Almost.

That assumption is now being tested. Alphabet unsettled investors last week by raising its capital-spending forecast while showing signs of pressure on free cash flow. Tesla has also raised questions about whether ambitious investment plans are consuming cash faster than expected.

Microsoft and Meta report Wednesday, followed by Amazon on Thursday. Apple also reports this week. Investors will examine revenue and profit, but the most important figures may be capital expenditure and free cash flow. If spending continues to rise while cash generation weakens, these companies may need to rely more heavily on debt or equity markets.

Lenders are already becoming less enthusiastic. Meta is turning to Wall Street to help finance enormous data-center projects, but investors recently demanded higher yields on debt linked to a facility in El Paso than they did for a comparable project last year. Bond prices for Microsoft and Amazon have also come under pressure. S&P has cut Oracle's credit rating to only one step above junk.

The industry's interdependence makes the problem more serious. Nvidia has reportedly agreed to guarantee financing for a major OpenAI data center. That means the company selling the chips may also be helping customers obtain the money needed to buy them. It is an ingenious arrangement, provided nobody asks too loudly where the demand ends and the financing begins.

Credit risk is therefore spreading across the AI ecosystem. Chipmakers depend on cloud companies. Cloud companies depend on model developers. Model developers depend on continued access to vast amounts of computing power. A weakness in one part of the chain can quickly affect the rest.

Higher interest rates would make the structure more fragile. The Federal Reserve is expected to keep rates unchanged Wednesday, although markets assign roughly a 36% to 38% probability to an increase. Even without an immediate move, investors expect borrowing costs to rise by at least a quarter point before year-end.

The retreat from technology does not mean investors are abandoning the stock market. They are rotating. The Dow rose about 0.5% Tuesday as money moved toward healthcare, consumer companies and financial stocks. Coca-Cola gained after raising its annual sales and profit forecasts. Johnson & Johnson advanced after announcing a roughly $5.5 billion agreement intended to resolve tens of thousands of talc lawsuits. Visa, PayPal, Boeing, Ford, Mondelez and UPS were among the companies reporting results.

Today's economic highlights:

On today's agenda: consumer confidence and unemployment benefit claims in France; the unemployment rate in Spain; in the United States, retail inventories excluding autos, the goods trade balance, wholesale inventories, the S&P/Case-Shiller home price index, the Conference Board consumer confidence, and the API crude oil stock change. See the full calendar here.

  • Dollar index: 101.531
  • Gold: 4,033
  • Crude Oil (BRENT): 86.07 (WTI) 80.98
  • United States 10 years: 4.63%
  • BITCOIN: 63,4061

In corporate news:

Analyst Recommendations:

  • Albertsons Companies, Inc.: UBS downgrades to neutral from buy with a price target reduced from USD 20 to USD 12.
  • International Paper Company: JP Morgan upgrades to overweight from neutral with a price target raised from GBP 32 to GBP 46.
  • Intuit Inc.: TD Cowen downgrades to hold from buy with a price target reduced from USD 504 to USD 304.
  • Intuitive Surgical, Inc.: UBS upgrades to buy from neutral with a price target reduced from USD 550 to USD 500.
  • Jbs N.v.: JP Morgan upgrades to overweight from neutral with a target price of USD 18.
  • Amd (Advanced Micro Devices): President Capital Management Corp maintains its buy recommendation with a price target raised from 440 to USD 574.
  • Baxter International Inc.: Stifel maintains its hold recommendation with a price target raised from 19 to USD 24.
  • Bellring Brands, Inc.: JP Morgan maintains its overweight recommendation with a price target raised from 13 to USD 17.
  • Fortinet, Inc.: Evercore ISI maintains its market perform recommendation with a price target raised from 100 to USD 150.
  • Globe Life Inc.: BMO Capital Markets maintains its market perform recommendation with a price target raised from 150 to USD 192.
  • Lamb Weston Holdings, Inc.: Bernstein maintains its market perform recommendation with a price target raised from 42 to USD 57.
  • Medpace Holdings, Inc.: Rothschild & Co Redburn maintains its neutral recommendation with a price target raised from 481 to USD 615.
  • Tenet Healthcare Corporation: Leerink Partners maintains its outperform recommendation with a price target raised from 218 to USD 309.