The story: TSMC moves from supplier to stakeholder
TSMC has stepped more visibly into the AI capital cycle. On November 10, the chipmaker’s board approved investments of up to $500 million in Anthropic common shares and up to $500 million in OpenAI Group PBC preferred stock . In the same board package, TSMC also approved up to $500 million to establish an investment company focused on strategic investments across the semiconductor industry and related ecosystem .
The headline is not just that TSMC is backing two of the most important AI model companies. It is that the world’s largest contract chipmaker is taking equity exposure to the very firms whose growth is pulling more capacity through the semiconductor supply chain. Reuters described the move as a way for TSMC to benefit not only from supplying AI infrastructure, but also from owning stakes in influential AI developers . South China Morning Post made the same strategic point more directly: the investments give TSMC direct financial ties to frontier AI labs that are driving demand for its business .
This is a subtle but important shift. TSMC remains a foundry, not a model lab or a cloud company. But by buying into Anthropic and OpenAI, it is no longer only the toll road for AI compute; it is also taking a small position in the traffic.
What the board approved
The details matter. TSMC said the Anthropic investment would buy common shares at Anthropic’s initial public offering price, with the amount capped at $500 million . The OpenAI investment, also capped at $500 million, would buy preferred stock in OpenAI Group PBC . A separate $500 million ceiling was approved for a new investment company that will make strategic investments in the semiconductor industry and related ecosystem .
Those approvals sat beside a much larger capacity decision. TSMC’s board approved capital appropriations of about $40.93 billion for advanced technology capacity, advanced packaging, mature or specialty technology capacity, fab construction, facility systems, 2027 research and development capital investments, sustaining capital expenditure and capitalized leased assets . The board also approved a third-quarter cash dividend of NT$9 per share, with the common-share ex-dividend date set for March 11, 2027, the record date for March 17, 2027 and payment scheduled for April 8, 2027 .
The Taipei Times reported that the NT$9 dividend is a record quarterly cash dividend and up from NT$7 for the second quarter . It also reported that TSMC posted third-quarter net profit of NT$763.06 billion, up 68.7 percent from a year earlier and 8.0 percent from the previous quarter . In other words, the investment announcement came bundled with a message of financial strength: TSMC is spending heavily, returning more cash to shareholders and placing targeted bets in AI demand creation at the same time.
Why Anthropic and OpenAI matter to a foundry
The obvious answer is demand. Anthropic and OpenAI do not manufacture chips, but their models require enormous quantities of compute. That compute is purchased through clouds, hyperscalers, custom accelerator programs and GPU supply chains. Much of the most advanced silicon in that chain ultimately depends on TSMC’s process technology and packaging capacity.
Reuters noted that TSMC is a critical supplier to leading AI firms, including Nvidia, whose graphics processors underpin many current AI models . Taipei Times reported that TSMC’s board also approved a roughly $40.93 billion capital budget for expansion amid the AI boom, covering advanced manufacturing capacity, fab construction and research and development . The connective tissue is clear: model demand becomes accelerator demand; accelerator demand becomes advanced wafer and packaging demand; and those bottlenecks become TSMC’s growth problem and growth opportunity.
The Anthropic timing is also notable. Taipei Times reported that Anthropic plans to raise up to $64.3 billion in its upcoming IPO by selling 825 million Class A common shares at an estimated price range of $70 to $78 per share . The same report said the filing includes a separate private placement of $26 billion to investors including AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC . TSMC’s approved purchase of Anthropic common shares at the IPO price therefore places the foundry inside a broader financing round involving several major AI infrastructure players.
OpenAI is different because TSMC is buying preferred stock rather than IPO common shares . But the strategic logic is similar: TSMC is gaining financial exposure to a model developer whose compute needs influence the road maps of cloud providers, accelerator designers and advanced packaging customers.
The circular AI capital stack
The move deepens what has become one of the defining features of the AI boom: capital circulates between compute suppliers, infrastructure buyers and model developers. Chipmakers, cloud companies and strategic investors fund AI labs. AI labs use capital to train and deploy larger models. Larger models require more accelerators, more advanced packaging and more leading-edge wafers. Those requirements reinforce demand for the suppliers that helped fund the labs in the first place.
TSMC’s version of this loop is not the same as a chip designer offering a customer financing package. TSMC does not sell finished GPUs under its own brand. Its leverage sits deeper in the stack: manufacturing, advanced nodes and packaging capacity. That may make the investment look smaller than the supply agreements and mega-rounds elsewhere in AI, but strategically it is highly focused. TSMC is not trying to become OpenAI or Anthropic. It is buying a window into the demand centers that decide how much compute the world will need next.
South China Morning Post reported that the investments could give TSMC clearer insight into AI demand as it backs two leading model developers and prepares an investment arm . That insight may be as valuable as the equity upside. In a world where capacity planning requires multiyear commitments and tens of billions of dollars, better visibility into future AI workloads can shape how aggressively TSMC builds advanced-node capacity, packaging lines and fab infrastructure.
Shareholders get a dividend signal too
The dividend increase makes the announcement more than a pure AI strategy story. TSMC’s board approved NT$9 per share for the third quarter, while Taipei Times said the company began paying quarterly cash dividends in 2019 . The signal is that TSMC can fund capacity, pursue strategic investments and increase shareholder payouts from a position of unusually strong profitability.
That matters because the AI cycle is expensive. TSMC’s capital appropriations of about $40.93 billion are aimed at long-term capacity plans based on market demand forecasts and the company’s technology road map . Reuters reported that TSMC expects full-year 2026 revenue growth of about 44 percent and has said demand continues to outstrip manufacturing capacity . AP also reported that Taiwan’s Taiex led regional markets as TSMC rose 3.1 percent after reporting October revenue nearly six times higher than a year earlier, and noted the board’s approval of up to $1 billion for Anthropic and OpenAI .
The market read-through is straightforward: investors are rewarding evidence that AI demand is still converting into foundry revenue. But the strategic risk is just as obvious. If AI spending slows, TSMC’s exposure is no longer limited to customer orders and capacity utilization. It will also own pieces, however small, of the model developers at the center of that spending cycle.
The bigger meaning
TSMC’s Anthropic and OpenAI investments do not transform it into a venture capital firm overnight. The dollar amounts are small compared with its capital budget and tiny relative to the valuations being discussed around leading AI labs. But the symbolism is large. The foundry at the base of the AI hardware pyramid is now taking direct financial stakes in the companies at the top of the AI application and model pyramid.
That is the new AI flywheel in miniature. Model growth raises demand for silicon. Silicon capacity enables model growth. Strategic investments tighten the relationship between the two. The loop now runs through advanced nodes, advanced packaging and increasingly expensive wafers.
For TSMC, the bet is pragmatic: if Anthropic and OpenAI keep scaling, someone must manufacture the chips and package the systems that make that scaling possible. TSMC already earns from that bottleneck. Now it wants a small share of the upside on the other side of the bottleneck too.
Comments
Be the first to comment.