TSMC Board Approves $40.9B 2027 Capex and Anthropic, OpenAI Stakes

TSMC's board approved roughly US$40.9 billion in capital appropriations for 2027 on November 10, 2026 — and, in the same meeting, cleared up to US$500 million to buy Anthropic PBC common shares and up to US$500 million to buy OpenAI Group PBC preferred stock, plus a separate US$500 million vehicle for broader semiconductor-ecosystem investments. The resolutions, disclosed in TSMC's own investor filings, came the same week the company reported October 2026 revenue up 59.7% year-on-year, underscoring how directly the world's largest contract chipmaker is now tying its balance sheet to the two best-funded AI labs it supplies.

TSMC did not publish a stated rationale for the Anthropic and OpenAI stakes in its press release or linked board resolutions — the filings describe only the mechanics (amounts, share classes, and pricing) and leave the "why" unstated. This article reports what was approved, in TSMC's own words, and lays out the context needed to understand why it matters for the AI chip supply chain, without putting words in TSMC's mouth.

What TSMC's board actually approved on November 10

At its board meeting on November 10, 2026, TSMC (TWSE: 2330, NYSE: TSM) passed several distinct resolutions, bundled together in the same "Board of Directors Meeting Resolutions" release. Besides routine items — a Q3 2026 earnings confirmation and a quarterly cash dividend — the board approved four capital-allocation decisions worth tracking separately:

• Approximately US$40,926.70 million (~US$40.9 billion) in capital appropriations for 2027.
• Up to US$500 million to establish a new investment company for "strategic investments in the semiconductor industry and related ecosystem."
• Up to US$500 million to purchase Anthropic PBC common shares, priced at Anthropic's own per-share initial public offering price.
• Up to US$500 million to purchase OpenAI Group PBC preferred stock.

That is as far as TSMC's own language goes. The release does not say what stake size those dollar caps would buy, does not attach a strategic-rationale quote from any TSMC executive, and does not disclose a timeline for completing either purchase. Both AI-lab investments are described as caps — "not to exceed" the stated amount — not as a committed, final purchase price. (See TSMC's investor relations site for the company's broader financial disclosures.)

The ~US$40.9 billion in 2027 capital appropriations

TSMC's board resolution sets the 2027 capital-appropriations figure at approximately US$40,926.70 million. According to the company's own breakdown, that money is earmarked across four buckets:

1. Installing and upgrading advanced-technology (leading-edge logic) capacity.
2. Installing and upgrading advanced-packaging, mature-node, and specialty-technology capacity.
3. Fab construction and installation of fab facility systems.
4. 2027 R&D capital investments, sustaining capital expenditures, and capitalized leased assets.

TSMC's filing does not further subdivide the US$40.9 billion by geography or by specific fab project, so this article does not attribute dollar amounts to any individual site. The figure covers appropriations approved for calendar-year 2027, on top of whatever TSMC separately confirms as its full-year 2026 capital-expenditure total in its next quarterly earnings call.

Two US$500 million stakes: Anthropic and OpenAI

The headline-grabbing items in the same resolution are the two AI-lab investments. TSMC's board approved, separately:

Anthropic PBC: up to US$500 million to purchase Anthropic's common shares, priced at Anthropic's own per-share IPO price. Because the purchase is pegged to an IPO price, the dollar cap functions as a ceiling on what TSMC will spend, not a guarantee that the full amount will be deployed or a disclosure of how many shares that buys.

OpenAI Group PBC: up to US$500 million to purchase OpenAI Group PBC preferred stock. TSMC's release gives no further pricing detail, no share count, and no indication of which funding round or share class this preferred stock belongs to.

Both investments sit alongside — but are legally distinct from — the up to US$500 million TSMC's board separately approved to set up a new investment company aimed at "strategic investments in the semiconductor industry and related ecosystem." That vehicle is broader in scope than the two named AI labs and appears designed to let TSMC make further ecosystem investments beyond this single board meeting.

Both Anthropic and OpenAI's operating entities are structured as public benefit corporations — Anthropic PBC and OpenAI Group PBC, respectively — a corporate form that lets a company pursue a stated public-benefit mission alongside ordinary shareholder returns, while still issuing common and preferred stock like a conventional corporation. That structure is why TSMC's resolution can reference "common shares" for Anthropic and "preferred stock" for OpenAI: both labs have cap tables with multiple share classes, same as any venture-backed company raising outside capital, even though each also carries a public-benefit charter.

TSMC is already the primary foundry behind the advanced logic chips that power both companies' frontier AI models, largely through its relationships with chip designers such as Nvidia, AMD, and the custom-silicon programs each AI lab runs with cloud partners. Taking direct equity positions in Anthropic and OpenAI — on top of being their chips' ultimate manufacturer — ties TSMC's capital structure to their financial fortunes in a way that a pure supplier relationship does not. TSMC's own materials do not explain that linkage; it is a straightforward reading of what a foundry buying equity in two of its largest end customers' labs implies, not a claim TSMC itself has made.

For deeper background on the leading-edge process node most relevant to the chips these AI labs depend on, see Pandromeda's explainer on TSMC's 2nm (N2) process.

The new semiconductor-ecosystem investment vehicle

Separate from the Anthropic and OpenAI line items, the board authorized up to US$500 million to establish a new investment company. TSMC's own description of its purpose is narrow: "strategic investments in the semiconductor industry and related ecosystem." No portfolio, mandate document, or named leadership for the vehicle was included in the release TSMC published alongside the resolutions.

Taken together with the AI-lab stakes, the three US$500 million approvals suggest TSMC's board is treating equity investment — not just contract manufacturing — as a tool for deepening ties across the AI compute stack. That reading is an inference from the resolutions as published, not a stated TSMC objective.

Q3 2026 results and the dividend behind the same resolutions

The same board meeting formally confirmed TSMC's third-quarter 2026 results: consolidated revenue of NT$1,494.24 billion, net income of NT$763.06 billion, and diluted earnings per share of NT$29.43. Using an approximate conversion of NT$31 to US$1 (actual exchange rates move daily, so treat this as a rough guide rather than an exact figure), that works out to roughly:

• Q3 2026 revenue: ~US$48.2 billion
• Q3 2026 net income: ~US$24.6 billion

The board also approved a cash dividend of NT$9.0 per share (roughly US$0.29 per share at the same approximate rate) for the third quarter of 2026. For common shareholders, the record date is March 17, 2027, with the shareholder register closed from March 13 through March 17, 2027, and payment set for April 8, 2027; the ex-dividend date for both ADS holders and common shares is March 11, 2027.

October 2026 revenue: up 59.7% year-on-year

In a separate release issued the same day, TSMC reported October 2026 consolidated revenue of approximately NT$586.89 billion — up 14.7% from September 2026 and up 59.7% from October 2025. At the same approximate NT$31-to-US$1 conversion, that is roughly US$18.9 billion for the month.

Cumulative January–October 2026 revenue reached NT$4,485.62 billion, up 43.3% year-on-year, or roughly US$144.7 billion at the same approximate conversion. TSMC's revenue reports are issued monthly and are denominated in New Taiwan dollars; the company's quarterly and annual results, reported separately, typically also carry a U.S. dollar reconciliation in its earnings materials.

Why this matters for the AI chip supply chain

Three things changed in one board meeting that are each individually significant, and more so together:

TSMC is funding capacity ahead of demand it expects to keep growing. A ~US$40.9 billion capital-appropriations approval for a single calendar year, layered on top of 2026's spending, signals TSMC's board expects advanced-node and advanced-packaging demand — driven heavily by AI accelerators — to keep justifying that level of investment into 2027.

TSMC is now a direct shareholder in two of its largest AI customers. Up to US$1 billion combined in Anthropic and OpenAI securities makes TSMC a capital stakeholder, not just a manufacturing partner, in the two companies most often named as the frontier of large-scale AI model development. That is a structural change in the relationship, whatever TSMC's internal reasoning — which it has not disclosed.

A new, open-ended semiconductor-investment vehicle gives TSMC more room to do this again. The up to US$500 million investment company is not limited to Anthropic or OpenAI by name, which means further equity stakes elsewhere in the AI/semiconductor ecosystem are plausible, not merely theoretical.

None of this is TSMC publicly explaining its own motives — it is three disclosed dollar figures and the structural read of what they mean. Readers should treat any broader claim about "why" TSMC did this — competitive hedging, customer-relationship insurance, long-term demand visibility, or anything else — as inference, not as something TSMC's board resolutions themselves say.

It is also worth being precise about scale. A combined US$1 billion cap across the Anthropic and OpenAI purchases is small relative to TSMC's own ~US$40.9 billion 2027 capital-appropriations approval, and smaller still relative to either AI lab's total disclosed funding to date. The significance is less about the dollar amount and more about the fact that the world's dominant advanced-logic foundry — the company that already manufactures the chips underpinning both labs' AI accelerators, directly or through partners such as Nvidia and AMD — is now also a prospective shareholder in both of them at once, inside a single board resolution.

For related coverage of the AI compute buildout from the demand side, see Pandromeda's report on CoreWeave's Q3 2026 earnings, and for recent changes at the GPU side of the AI chip supply chain, see the piece on Jensen Huang's departure as Nvidia CEO.

What happens next

Several concrete dates and disclosures to watch follow directly from this board meeting:

• Q3 2026 dividend timeline: ex-dividend March 11, 2027 (ADS and common shares); record date March 17, 2027; shareholder register closed March 13–17, 2027; payment April 8, 2027.
• Anthropic share purchase: pegged to Anthropic's IPO per-share price — meaning TSMC's actual spend and timing depend on when, and at what price, Anthropic goes public. Neither has been disclosed by TSMC.
• OpenAI preferred-stock purchase: no disclosed timeline or pricing; TSMC has not said when the up to US$500 million would be deployed.
• New investment company: no disclosed launch date, mandate document, or leadership; future ecosystem investments from this vehicle would likely surface in subsequent TSMC board-resolution releases.
• Full 2027 capex guidance: TSMC typically narrows and confirms a full-year capital-expenditure number on its quarterly earnings calls; the ~US$40.9 billion appropriation is a board-approved ceiling for appropriations, not necessarily the final reported capex figure.

Pandromeda will follow up as TSMC discloses further detail on any of these items, including whether the Anthropic and OpenAI purchases are completed and at what effective price.

Written byMara Lindqvist

Mara Lindqvist edits the hardware desk. She covers graphics cards, processors, memory and storage, the foundries and chip designers behind them, and what the numbers on a spec sheet mean for people choosing a PC. Specifications in her stories come from manufacturer spec pages and datasheets.

More from Mara LindqvistAll Hardware coverage