TSMC’s $1 billion stakes in OpenAI and Anthropic add another loop to AI’s circular financing

TSMC’s board approved plans to invest up to $500 million in OpenAI preferred stock and another $500 million in Anthropic common shares at its expected IPO price, according to Reuters as cited by Ars Technica. The company that physically manufactures most of Nvidia’s AI data center chips, and most of Apple’s custom silicon, will hold direct stakes in two of the largest buyers of that hardware. Its fabs are mainly in Taiwan, with operations in Japan and China, an expanding Arizona cluster and a fab under construction in Germany.

The Wall Street Journal framed the move as a possible shot of confidence while the AI trade shows strain, a strain sharpened by Nvidia’s leadership crisis (coverage). Analysts quoted by the Journal also warned that circular financing among investors, suppliers and customers makes real demand harder to judge.

TSMC has reported six straight quarters of record profit, including more than $23 billion in the July through September quarter, per Reuters. A billion dollars split between the two leading labs is a modest bet for a company that already wins whenever either buys more compute. OpenAI’s reported quarterly revenue crossing $10 billion (coverage) is the kind of growth TSMC is buying into.

When a supplier invests in a customer that buys from the supplier, part of the money tends to come back as orders. Nvidia has done the same at far larger scale, with $30 billion in OpenAI and $10 billion in Anthropic so far. Each deal can make sense alone. Together they make it harder to tell how much AI demand is organic and how much is financed by those selling the picks and shovels. TechCrunch has described the same web of circular deals around Nvidia (five ways).

The Anthropic piece stands out. Committing to buy at the expected IPO price reads as an anchor for the offering, which helps Anthropic and may steady public investors. It also adds one more large player with a stake in the IPO going well. Anthropic’s own filings have flagged huge compute plans beside large losses (prospectus).

If this scales, chipmakers, cloud providers and labs all hold pieces of one another. That can speed useful build-outs. It can also mean that if demand disappoints, the losses travel in a circle too. Labs without supplier backing may find it harder to compete for scarce leading-edge capacity. Clearer disclosures should accompany these ties. More in our OpenAI and Anthropic coverage and finance coverage.