TSMC becomes a shareholder in OpenAI and Anthropic while manufacturing both companies' chips

Taiwan Semiconductor Manufacturing Company (TSMC), the world's largest contract chipmaker, has stopped being merely a supplier to those building artificial intelligence and become a partner of theirs. On Tuesday (10), the company's board approved up to $500 million to buy Anthropic common shares at its initial public offering (IPO) price, plus another up to $500 million to buy OpenAI preferred shares. A third investment, also of up to $500 million, will go to a strategic investment arm dedicated to the semiconductor ecosystem, according to SCMP.

From neutrality to direct investment

The decision marks an explicit shift. In July 2026, asked during an earnings call whether TSMC would follow Nvidia's path and start financing end customers of its silicon, president and CEO C.C. Wei was direct:

So far, no.

C.C. Wei, president and CEO of TSMC

At the time, Wei added that the existing commercial agreements with clients were working well, with no need for equity stakes. Four months later, the same company approved direct investments in the two AI labs that consume the most advanced manufacturing capacity on the planet.

Not the first time, but new territory

TSMC already had a track record of backing key ecosystem partners before this. In 2012, it bought 5% of the Dutch company ASML, the exclusive supplier of the extreme lithography machines used in the most advanced processes. In 2023, it allocated $100 million to the IPO of chip designer Arm Holdings. But investing in AI model developers, software companies that don't design silicon, is a different frontier: TSMC is buying stakes in those who buy its clients' products, not in those who manufacture inputs for itself.

Why now: tight capacity and a five-year forecast

The strategy emerges at a time when the foundry is looking beyond its immediate fabless clients, such as Nvidia and AMD, to gauge real end-user demand amid an unprecedented investment cycle in the sector. Wei said last month that TSMC sent teams to the United States to personally inspect data center construction and power grid capacity, factoring real energy constraints and hyperscalers' financial balance sheets into fab expansion planning.

Citi analyst Laura Chen noted on October 30, after meetings with TSMC's management, that the company's five-year forecasting model already factors in demand for AI models and gives special weight to the financial resilience of established cloud providers:

This should help mitigate overcapacity risks during potential cyclical corrections.

Laura Chen, Citi analyst

Holding equity stakes in these labs gives TSMC a closer view of the real computing capacity needs of the companies that buy its products, which helps plan expansion without repeating cycles of oversupply or shortage.

The bottleneck that already delayed deliveries

The urgency for this visibility has a concrete reason. In its own updated IPO filing disclosed on Tuesday, Anthropic stated that TSMC's capacity constraints had, on some occasions, already delayed scheduled chip deliveries. The prospectus also revealed that TSMC Partners, the foundry's subsidiary, will buy non-voting preferred shares (Class N) in a $26 billion private placement by Anthropic, alongside investors such as Google and Nvidia, contingent on the offering's completion.

Billions in capex in parallel

At the same board meeting, TSMC approved roughly $40.9 billion for projects including advanced-node manufacturing, packaging capacity, and construction of new fabs. Wei said last month that nearly 20 facilities are under simultaneous development around the world. Goldman Sachs projected in October that the company's annual capex should rise to $85 billion next year and reach $98 billion by 2028.

The line between chip designer and model developer is disappearing

The filings from both AI companies themselves show another structural shift: they're not just buying capacity, they're influencing the design of the chips that run their models. Anthropic said its partnerships with Amazon and Nvidia already allow it to influence the architecture of future processors. OpenAI stated it is collaborating with Broadcom to deploy 10 gigawatts of custom accelerators for its own use.

In Anthropic's private placement, AMD's $2.5 billion share purchase counts toward the processor maker's commitment to acquire up to $5 billion in Anthropic shares under a hardware supply agreement signed in July 2026. Chip supplier and chip buyer, increasingly, sit at the same investor table.

What hasn't been said yet

TSMC's board statement did not reveal whether the equity commitments come alongside chip supply or joint-design agreements with OpenAI or Anthropic. In other words: it's not known whether the equity stake also unlocks priority allocation of manufacturing capacity for these two companies, at the expense of the foundry's other clients.

What this changes for developers in Brazil

For those building products running on rented GPUs or language model APIs, this move isn't distant financial market news: it's supply chain news. The manufacturing capacity shortage that Anthropic itself admits having suffered helps explain GPU instance waitlists, tight cloud provider quotas, and why certain models become more expensive or slower to scale at certain times of the year.

There's a second point, less immediate but more structural: with TSMC becoming a partner of the two labs that compete most for cutting-edge manufacturing capacity, the risk of power concentration in the AI chain increases.

If the foundry doesn't disclose whether the investment comes with wafer allocation priority, engineering teams that depend on third-party infrastructure (whether via Azure, AWS, or Google Cloud, which host these models) become more subject to decisions made outside their control, in a negotiation that combines equity stakes, supply agreements, and influence over chip architecture all at once. It's worth watching whether Anthropic's IPO, still underway, will spell out this detail.