TSMC at AGM: ‘Will Be a Long Time’ Before AI Supply Catches Demand

At Taiwan Semiconductor Manufacturing Company’s annual shareholder meeting in Hsinchu on June 4, 2026, chairman and CEO C.C. Wei delivered a message that read less like a routine governance event and more like a state-of-the-AI-industry address: demand for advanced chips is still outrunning what the world’s largest foundry can produce, and TSMC is finally talking openly about leaning into its pricing power.

“We continue to see increasing adoption of AI models across consumer, enterprise and sovereign AI applications,” Wei said, according to a summary of the meeting. On capacity, he was blunter: TSMC “can only produce so much.”

The combination — structural shortage in the most advanced nodes, paired with an explicit willingness to negotiate higher prices with customers including Nvidia and Apple — is exactly the setup the bulls have been arguing for since the second half of 2025. It also helps explain why TSMC’s Taiwan-listed stock has more than doubled in twelve months.

What Wei actually said about pricing

The most quoted line of the meeting was Wei’s response on whether TSMC would raise prices. “I’d like to do that,” he said, before adding the qualifier: “We still need to make money.” That sounds anodyne; in context, it is not. Foundries historically anchor pricing to long-term wafer agreements that customers treat as semi-fixed. Wei publicly opening the door to higher prices for the most demand-constrained nodes is a signal to the buy side that gross margin can rise from here.

He was equally explicit about how TSMC plans to behave — and crucially, how it does not. “We don’t want to suddenly raise prices like memory companies do,” Wei said. “That’s not sustainable.” Memory pricing notoriously whipsaws through boom-and-bust cycles. TSMC’s brand promise is the opposite: a steady, premium logic foundry where customers can plan ten-year roadmaps.

Why this matters for TSM shareholders

TSMC reported 2025 revenue of $122.42 billion and net income of $55.13 billion on output of roughly 15 million twelve-inch equivalent wafers. The foundry’s roughly 70% global share of pure-play foundry revenue means any pricing increase on advanced nodes flows directly to operating profit — there is no second supplier with comparable yields at 3nm or 2nm to undercut.

The pricing signal arrived alongside a capacity update that, on its own, would have moved the stock. Wei said the company’s two land parcels in Arizona are sufficient for the next ten years. Stacked on the planned “gigafab” complex of six facilities and an incremental $100 billion Arizona commitment announced in mid-2025, TSMC’s U.S. footprint is now budgeted at roughly $165 billion — a number that, twelve months ago, would have been an unimaginable U.S. semiconductor capital plan.

Metric Value Note
2025 revenue $122.42B Full-year, reported
2025 net income $55.13B 45.0% net margin
2025 wafer output 15.02M 12-inch equivalent wafers
Foundry market share ~70% Global pure-play foundry
Arizona total commitment ~$165B Across six planned fabs
Profit sharing change, 2023→2024 +30% Per Wei’s AGM remarks
Profit sharing change, 2024→2025 +30% Same source
Profit sharing change, 2025→2026E +30% Wei’s on-stage guidance
Sources: TSMC 2025 reported financials via Wikipedia infobox citing TSMC filings; AGM remarks via Investing.com summary, June 4, 2026.

The 2nm transition is the unlock

The capacity narrative is inseparable from the node narrative. TSMC transitioned to its 2nm process (N2) in late 2025, swapping FinFET transistors for gate-all-around nanosheet structures. Versus an enhanced 3nm process, N2 delivers “a 10 to 15% performance boost or a 25 to 30% reduction in power consumption” at the same density. For AI accelerator designers, that translates almost directly into either more tokens per watt or smaller racks for the same throughput — the variable hyperscalers care about most.

Wei also flagged the boundary of TSMC’s tooling ambitions. The next-generation High-NA EUV lithography machines, which can cost up to $400 million each, are not currently needed for production at the company’s pace of node shrinks. That is a quiet way of saying TSMC believes it can keep extending performance on its existing tool set for at least one more node generation, rather than absorbing the capex shock that High-NA would entail.

What a 12-month chart looks like

TSMC Taiwan share price, June 2025 to June 2026 A bar chart showing TSMC’s Taiwan-listed shares climbing from approximately T$950 in early June 2025 to T$2,425 on June 4, 2026, an approximate 155% gain over twelve months. TSMC Taiwan shares: ~T$950 (Jun 2025) → T$2,425 (Jun 4, 2026) T$2,500 T$1,500 T$0 Jun 2025 T$950 Jun 4, 2026 T$2,425
Source: Share price data cited by CEO C.C. Wei at AGM, per Investing.com, June 4, 2026.

The customer concentration question

Wei named Nvidia and Apple by name during the meeting, and TSMC’s supplier list also includes Broadcom and Qualcomm as its largest customers. The market reaction tends to treat that concentration as an asset rather than a risk: while a slowdown in any one of those names would matter, the same names collectively define the most demand-resilient parts of the chip cycle. AI accelerators, modems, smartphone application processors, and merchant networking silicon are all sold into customers with either monopoly-like end markets or sovereign-AI buyers underwriting demand.

Geopolitics, of course, remains the wildcard. The accelerated Arizona buildout — six planned fabs, two parcels of land sufficient for ten years, and a budget approaching $165 billion — is the most expensive insurance policy in industrial history against Taiwan Strait risk. Wei’s framing at the AGM treated it as table stakes rather than a strategic pivot.

The takeaway for the tape

For investors watching the AI tape, the meeting reinforced three things. First, the bottleneck on AI compute remains physical, not financial — Nvidia’s order book is gated by TSMC’s wafer starts, not by hyperscaler capex appetite. Second, the foundry that controls the bottleneck has both the willingness and the customer goodwill to widen margins from here. Third, the U.S. manufacturing pivot is fully under way and is being funded by the host company, not by the U.S. government’s $6.6 billion preliminary CHIPS Act award.

None of that is a recommendation. But Wei’s calm, on-the-record signal that TSMC intends to raise prices “without doing it like memory companies” is the kind of management language that gross-margin upgrades are built on — and it landed at the same shareholder meeting where the chairman thanked employees for a tripled share price.

Sources

Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.

Leave a Comment