The Taiwan Semiconductor Manufacturing Company (TSMC) just made the kind of move that venture capital firms usually reserve for their biggest bets, and the message it sends to the semiconductor industry is as important as the amount involved. The world's largest chip manufacturer approved investments of up to $500 million each in the two most valuable artificial intelligence companies on the planet — Anthropic and OpenAI — in a gesture that marks a turning point in how the hardware and software supply chain reconnects.
The decision, announced in a board statement on November 10, has precise parameters. TSMC may purchase up to $500 million in Anthropic common shares at the company's initial public offering (IPO) price, whose range was set between $70 and $78 per share — and whose target is to raise up to $64.35 billion from the sale of 825 million shares. For OpenAI, the investment will be up to $500 million in preferred stock. Both investments are subject to the terms of the respective offerings and do not mean the purchases have been completed. The approved amounts are ceilings, not executed commitments.
But it is precisely that detail — the difference between what was approved and what will actually be purchased — that makes the move more interesting. By approving these caps, TSMC's board is not just diversifying its investment portfolio. It is signaling a strategic posture change that few analysts expected within the next five years, and much less in a sector where the company has built its reputation.
Since it was founded by Morris Chang in 1987, TSMC operated on a model that became almost a dogma of the industry: the pure-play foundry. The company manufactures chips for other companies — Nvidia, Apple, AMD — but does not compete with its clients. This separation between design and manufacturing is what allows Nvidia to design an AI chip without having to worry about fabs, and Apple to have a supplier that is not a direct competitor.
The approved investments partially break that paradigm. By becoming a shareholder of OpenAI and Anthropic, TSMC is creating a bilateral financial relationship that has never existed between a foundry and its semiconductor clients. The company will continue manufacturing the processors powering OpenAI and Anthropic's data centers — the companies' chips are manufactured by TSMC for designers including Nvidia and other customers — but now it will also have a direct financial interest in the success of these companies as AI developers.
This verticalization is a reflection of a reality already evident in the sector: the AI value chain is becoming more integrated. It is not just about buying and selling chips. Investors in semiconductor infrastructure are taking active positions in the businesses that create demand for these chips — and in doing so, they are exposing themselves to the risk that this demand may not materialize.
The immediate question is: why would a manufacturing company buy shares of software companies? The financial answer is straightforward, but the strategic answer is more complex. Dan Ives of Yorkville Ives argued that Anthropic would benefit from disciplined IPO pricing amid volatile AI shares. But the real reason lies in the relationship between the foundry and its clients.
TSMC is already deeply exposed to the AI infrastructure investment cycle — the AI accelerator sector has become an important growth driver for the company. But this exposure occurs through chip orders, not through equity ownership. By purchasing shares in OpenAI and Anthropic, TSMC is assuming a different type of risk: the risk that the AI companies themselves may not become profitable, or that their valuations may collapse independently of chip demand.
This is a move that reflects a silent uncertainty on Wall Street: how much can AI infrastructure demand sustain itself? Kevin Net of Financière de l'Échiquier argued in a Bloomberg interview that strong earnings alone are insufficient to justify enthusiasm for TSMC without convincing investors that margins, investment plans and durable AI demand are truly consolidated.
The timing of the investments is not accidental. TSMC reported consolidated revenue of $45.50 billion (NT$1,494.24 billion) and net income of $23.23 billion (NT$763.06 billion) in the third quarter of 2026 — robust numbers that reflect unprecedented demand for processing capacity. Diluted earnings per share stood at $0.90. In October, revenue surged 59.7% year-over-year to a record NT$586.89 billion, driving the company's shares up more than 4% in Taipei.
This financial performance is what allows TSMC to consider investments outside its core business. But it is also what makes the potential risk more relevant: when a company earns this much, any capital allocation outside of manufacturing becomes a signal that management's view about the sector's future may be more cautious than market optimism suggests.
The board also approved capital appropriations of approximately $40.93 billion to support long-term capacity requirements — a number that, curiously, is about 40 times larger than the total value of investments in OpenAI and Anthropic. TSMC is, in absolute terms, betting much more on manufacturing than on equity ownership. But what matters is not the amount, but the signal it sends.
The outlook for TSMC in the coming months is complex. Anthropic is days away from its IPO, with a roadshow seeking to raise up to $64.35 billion. The opening price will be a moment of truth both for Anthropic and for TSMC: if the shares rise significantly after the IPO, the foundry's investment in OpenAI's preferred stock may seem too conservative; if they fall, TSMC will have taken advantage of an IPO price that was already considered optimistic.
The relationship with OpenAI is particularly interesting because OpenAI has no scheduled IPO — the investment in preferred stock is a way for TSMC to gain exposure to a private company without waiting for a public offering. This puts TSMC in the position of someone betting on OpenAI's longevity as an independent company, not just as a chip customer.
The relationship with Anthropic has a more predictable mechanism: the IPO price is already defined, and TSMC can buy at the same price as public investors. This means there is no information asymmetry — the foundry is not buying before others, it is buying alongside them.
The question worth more than the numbers is: what happens if the AI investment cycle goes wrong? TSMC is not the only foundry to consider this type of verticalization — Samsung already collaborated with TSMC on 3nm to produce base dies for Nvidia's HBM chips, and GlobalFoundries entered a manufacturing agreement with TSMC to supply silicon interposers. The semiconductor sector is integrating in ways never seen before.
But when a foundry buys shares of a software company, it is saying something about its confidence in the entire chain. If AI demand holds, TSMC will have intelligently diversified its revenue. If the AI infrastructure bubble bursts, the company will have bought shares in companies that may no longer have customers for the chips it manufactured.
The market reacted positively — TSMC shares rose more than 4% in Taipei — suggesting that investors see the move as a signal of confidence, not desperation. But economic history is full of companies that bet on their own supply chains and were surprised when they did not hold on their own. TSMC likely knows this better than anyone: Morris Chang built the company precisely because he realized that chip designers did not want to assume the risk of building fabs.
What happens when the foundry decides it also wants to share the risk of software? The market is already answering, but the definitive response is still to come.
Sources: Investment Monitor, Technode, The Hindu BusinessLine, Invezz
✓ Independent sources cross-checked and verified before publishing
