TSMC's Senior Vice President Hou Yung-ching stood at the company's North America Technology Symposium in late April and said something that should have made every competing chipmaker and every CHIPS Act administrator sit up straight: the company is building five advanced-node fabrication plants simultaneously in a single coastal city, and every wafer they will produce through the end of 2026 is already sold. He called it 'expansion at twice the speed.' What he meant was that TSMC has stopped waiting for the market to prove demand exists. The company is now building capacity based on a certainty that demand exceeds what they can physically manufacture.
This is not a product announcement or a marginal capacity upgrade. TSMC is constructing five 2nm fabs in Kaohsiung, Taiwan, all designated for its most advanced process node family, with phases already ranging from volume production to early construction. Fab 22 Phase 1 (P1) began volume production in the second half of 2025; Phase 2 completed equipment installation and entered trial production; Phase 3's building is nearly finished; and Phases 4 and 5 are already under construction, with full operational status targeted for Q4 2027. The Kaohsiung City Government has officially confirmed the plan. TSMC's SEC filing backs it up with $45 billion to $50 billion in committed capex by 2026 alone. Each fab costs approximately $9.3 billion. Five simultaneous ramps of the same process node in TSMC's own history has never happened before.
The numbers that matter most are the ones that measure what TSMC can make versus what customers need. With five 2nm fabs in operation, TSMC expects to increase output capacity by up to 45% compared to what the same period delivered for 3nm fabs. The company is guiding for 70% compound annual growth in 2nm capacity from 2026 through 2028. But here is the critical detail that reframes everything: analysts and supply-chain sources report that customer demand for 2nm wafers has already consumed TSMC's projected output through the end of 2026. TSMC executives have said on recent earnings calls that order visibility for the N2 node is 'very strong.' This does not mean demand is outpacing supply. It means TSMC is confident enough in future demand to bet $28 billion on three new Kaohsiung-area sites before customers have formally ordered all the capacity. The company is also planning to upgrade or install nine new factories annually, effectively doubling its historical expansion rate.
What created the conditions for this move is straightforward: demand for AI accelerators is exploding, and TSMC is the only company on Earth that can manufacture them at leading-edge nodes at scale. Wafer shipments for AI accelerators have increased 11x. Demand for larger chips featuring advanced packaging technologies has increased 6x. These are not modest growth rates. They are the kind of numbers that force a manufacturer to decide between rationing supply or building enough fabs to meet it. TSMC chose to build. Intel is closing fabs. Samsung's foundry business is burning cash. SMIC cannot access the most advanced equipment. TSMC is the only operator in the world moving decisively to increase advanced-node capacity at this speed and scale.
The winner is obvious and it is singular. TSMC's customers, Nvidia, Apple, AMD, Google, Meta, every major AI and HPC chipmaker, get guaranteed access to the world's most advanced manufacturing at the moment when advanced manufacturing is the constraint. TSMC's competitors lose. The company's ability to reach yield on 2nm is superior to what Samsung and Intel achieved on 3nm at similar stages of production. The performance leap from N3E to N2 is 10 to 15 percent at the same power, or 25 to 30 percent lower power at the same performance. When you combine manufacturing lead, yield advantage, and installed customer base, the mathematics of competition become asymmetrical. Samsung has announced no equivalent 2nm expansion. Intel is years away from manufacturing anything at N2 equivalence.
Here is what this actually means: the CHIPS Act and the EU Chips Act were designed to reduce geographic and vendor concentration in advanced semiconductor manufacturing. Instead, TSMC is pulling further ahead and consolidating control of the bottleneck that matters most. Taiwan already produces the vast majority of the world's leading-edge logic chips. Kaohsiung now becomes the single largest concentration of next-generation capacity anywhere on Earth. Geographic diversification has become a slower process than TSMC's ability to increase capacity. Speed of execution trumps policy intent. The company is not being forced to share the market with competitors gaining ground; it is rationing access to customers willing to pay premium prices for guaranteed advanced-node volume. Every dollar of the CHIPS Act incentive for Intel and Samsung makes TSMC's move feel less like an act of confidence and more like an act of dominance.
Watch for three things over the next eighteen months. First, track Intel's manufacturing cadence, if the company moves beyond roadmap talks and actually commits capex to bring a U.S. 2nm equivalent online by 2028 or 2029, the competitive picture changes; if Intel keeps deferring, you are watching the slow consolidation of a critical bottleneck. Second, monitor whether TSMC's order book remains fully booked through 2027; demand signals that stay 'very strong' through Q4 2026 would confirm that advanced-node supply is inelastic in the AI era. Third, watch Taiwan's political environment and cross-strait relations, this level of concentrated capacity in a single island location is structurally fragile in ways that five fabs in Arizona or Germany would not be, and any geopolitical friction that disrupts Kaohsiung's output becomes an economy-wide constraint for every AI company on Earth.