Intel and Broadcom CEOs cite capital access as potential AI constraint

Two of the most important men in American semiconductors have just described economic gravity. Washington will be tempted to treat it as a malfunction.

Broadcom's Hock Tan warned that financing may constrain artificial intelligence more than shortages of power, chips, or memory, with demand for computing capacity running above supply through 2029 and possibly into 2030. Intel's Lip-Bu Tan said the cost of capital is rising and that investors will increasingly want clear returns. Both then added the line that should have ended the conversation: longer-term growth depends on whether businesses adopt the technology and whether the spending produces economic benefits.

That is not a national emergency. It is a market clearing its throat.

"Access to capital" is what executives say when money stops pretending to be free. You cannot run structural deficits on this scale and then act shocked that a data center or a fab must clear a higher hurdle. Crowding out is not a seminar word. A treasurer in Santa Clara and a treasurer at the Treasury are bidding for the same patient money, and the one who never faces an earnings call usually wins. If Republicans mean what they say about growth, the capital constraint on AI begins in the appropriations process.

Intel's version of the warning deserves less sympathy, not more. This is the company Washington already anointed a national champion — CHIPS Act money, political oxygen, the full pageant — while the operational turnaround kept turning. Lip-Bu Tan is right that investors will demand returns. That demand is not a public problem to be solved. When a subsidized incumbent discovers that private capital is getting picky, the honest reply is: good. Pickiness is what the subsidy was supposed to make unnecessary, and did not. Another check will not install execution where execution has been missing. It will socialize the next miss and call it strategy.

The objection that has to be faced, not waved away, is China. Beijing allocates capital by political command, conceals waste, and will not ease off its semiconductor drive because the American cost of capital went up. A country that intends to remain a great power cannot be casual about whether leading-edge manufacturing exists on its own soil. But a narrow security interest in capacity the Pentagon and the economy cannot do without is not a license to underwrite the hyperscale spending boom. Most of that boom is commercial model training and cloud capacity, not a weapon. Matching Chinese waste because financing looks tight is how you import their overcapacity without their talent for hiding the losses.

Nor do I accept the ranking. Capital is not mysteriously missing. It is declining to fund projects this country has made grotesque to finish — power plants, transmission lines, data centers — after years of litigation, delay, and political whim. Investors are pricing a governing failure and calling it a credit problem. Republicans ought to be furious about the permits and the grid, not volunteering the taxpayer as the investor of last resort.

Believe the executives on the part they might have preferred you skim. Adoption. Returns. Economic benefit. If the technology pays, capital will find it in a country that stops consuming the savings, stops blocking the concrete, and stops confusing a CEO's cost of funds with the national interest. If it does not pay, the correction will be ugly and useful. America's edge was never that our chip executives enjoyed easier money than their rivals. It was that our money eventually had to come home. Do not surrender that because the cost of capital has remembered how to count.