TSMC is building five fabrication plants in Kaohsiung, Taiwan, all dedicated to its 2nm process node, creating the largest concentration of next-generation chip manufacturing capacity anywhere on Earth. The Kaohsiung City Government has officially confirmed the five-fab plan, and TSMC’s own annual filing with the SEC backs it up with tens of billions of dollars in committed capital spending on advanced nodes. Meanwhile, analysts and supply-chain sources report that customer demand for 2nm wafers has already consumed TSMC’s projected output through the end of 2026, a claim consistent with remarks TSMC executives have made on recent earnings calls about “very strong” order visibility for the node.
If those demand signals hold, the consequences ripple across the entire tech industry: from the smartphones and laptops shipping in 2027 to the AI accelerators powering the next wave of data centers.
Why Kaohsiung, and why five fabs
TSMC already operates packaging and testing facilities in Kaohsiung, so the city is not starting from zero. But the scale of the 2nm commitment is unprecedented for a single metropolitan area. Mayor Chen Chi-mai has framed the project in competitive terms, telling reporters that “Opportunity Favors the Prepared” as the city works to convert former industrial zones into high-tech semiconductor parks.
The municipal government has been coordinating land use, water infrastructure, and workforce development specifically to support the new fabs. Each advanced fabrication plant requires thousands of highly trained engineers and technicians, and Kaohsiung has responded by expanding vocational programs and forging university partnerships aimed at building a local talent pipeline. That human-capital effort matters as much as the concrete and steel: a 2nm fab is only as productive as the people who operate and continuously refine its process tools.
Physical resilience is another factor. Kaohsiung sits on Taiwan’s southwestern coast, exposed to typhoons and periodic water stress. The city has invested in reservoir capacity, desalination plants, and flood-control systems to reduce the risk of disruptions. For a 2nm production line, even a brief interruption in ultra-pure water or stable power can destroy wafers worth millions of dollars, so that infrastructure spending is not optional.
What TSMC’s SEC filing reveals about spending
TSMC’s Form 20-F for fiscal year 2025, filed with the U.S. Securities and Exchange Commission and archived on EDGAR, provides the financial scaffolding behind the Kaohsiung expansion. The filing details multi-year capital expenditure plans covering advanced nodes, including 2nm, and confirms that TSMC is spending at historically elevated levels to stay ahead of Samsung Foundry and Intel in the process-technology race.
The 20-F does not break out spending by individual fab site, so it cannot confirm construction timelines for each of the five Kaohsiung plants. What it does confirm is that TSMC’s board has approved the capital required to pursue aggressive capacity growth at the leading edge, and that the company views sustained demand for its most advanced nodes as the central justification for that spending.
How solid is the “sold out” claim?
The assertion that every 2nm wafer through 2026 is already spoken for deserves careful handling. TSMC has not published a customer-by-customer booking ledger, and no regulatory filing spells out allocation percentages by quarter. The “sold out” framing originates primarily from industry analysts and supply-chain reporting, bolstered by comments TSMC executives have made during earnings calls about strong demand and high order visibility for the N2 node.
That said, the underlying dynamics make the claim plausible. Apple is widely expected to be the first major customer on TSMC’s 2nm process, following the same pattern it established at 5nm and 3nm. Nvidia, AMD, Qualcomm, and MediaTek are all designing chips that target 2nm-class manufacturing. And the explosion in demand for AI training and inference hardware has added a new layer of urgency: hyperscale cloud operators are competing fiercely for access to the most power-efficient transistors available, and 2nm’s gate-all-around nanosheet architecture promises a significant jump in performance per watt over the current 3nm generation.
Still, readers should treat “sold out through 2026” as informed industry consensus rather than a verified TSMC disclosure. The distinction matters because the degree of constraint depends on how many of the five Kaohsiung fabs reach volume production within that window. Advanced fabs typically take two to three years from groundbreaking to high-volume output. If only one or two of the five are producing wafers by late 2026, the bottleneck is real but narrower than the headline suggests. If three or more are ramping, the supply picture looks different.
What this means for chip buyers
For large chip designers, the Kaohsiung cluster reinforces a pattern that has defined the semiconductor industry since the pandemic-era shortages: secure capacity early or risk being shut out. Long-term wafer supply agreements, prepayments, and co-investment deals with TSMC have become standard practice for companies like Apple and Nvidia. Smaller design firms without the financial muscle to lock in multi-year commitments may find themselves pushed to older process nodes or forced to partner with larger customers who can bundle orders.
TSMC also tends to segment capacity by customer and application, reserving specific production lines for certain chip families. A smartphone processor and a data-center GPU may both be manufactured on 2nm, but they could draw from different fabs with different bottlenecks and lead times. That segmentation is rarely discussed publicly, yet it can be decisive for buyers trying to plan product launches around wafer delivery schedules.
The geopolitical dimension Kaohsiung cannot escape
Concentrating five of the world’s most advanced chip factories in a single coastal city carries geopolitical weight that no amount of local infrastructure investment can fully offset. Taiwan already produces the vast majority of the world’s leading-edge semiconductors, and the Kaohsiung expansion deepens that concentration rather than diversifying it.
TSMC has been building capacity in Arizona, Japan, and Germany partly in response to pressure from governments that want domestic access to advanced chips. But those overseas fabs are focused on older nodes (4nm and 3nm variants in most cases), not 2nm. For the foreseeable future, anyone who needs a 2nm wafer will need it from Taiwan, and increasingly, from Kaohsiung specifically.
Policymakers in the United States, Europe, Japan, and South Korea are watching this dynamic closely. Subsidies like those in the U.S. CHIPS Act and the EU Chips Act were designed to reduce dependence on a single geography for critical semiconductors. The Kaohsiung five-fab plan is a reminder that TSMC’s technological lead, combined with the sheer speed at which it can build and fill new capacity, continues to outpace those diversification efforts.
A bold bet with real unknowns
The confirmed facts are striking on their own. TSMC is building five 2nm fabs in one city, backed by tens of billions in capital and coordinated support from local government. The company’s SEC filings confirm the financial commitment, and the demand signals from major customers and the AI hardware boom make the investment look well-timed.
What remains genuinely uncertain is the production timeline for each fab, the precise allocation of output among customers, and whether the “sold out through 2026” characterization holds up as construction schedules and demand forecasts evolve. Until TSMC discloses more granular data, the market will be navigating a landscape where the physical foundations are visible in concrete and steel, but the allocation of that capacity is still shaped as much by negotiation and leverage as by public disclosure.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
