TSMC revives Longtan fab plan
- Taiwan’s Hsinchu Science Park Bureau said on May 4 it will seek review this month for Longtan phase three, reopening a TSMC fab project. - The revived site could host angstrom-class production, with local reports pegging investment at NT$500 billion to NT$600 billion after 2023 opposition froze plans. - It matters because TSMC already shifted leading-edge growth south, so Longtan’s return would ease land pressure for future AI-era capacity.
Semiconductor manufacturing is hitting a very physical limit — not transistor physics, but land, permits, and neighborhoods. That is why TSMC’s Longtan move matters. On May 4, the Hsinchu Science Park Bureau said it plans to send the third-phase expansion of Longtan Science Park to Taiwan’s National Science and Technology Council later this month, and that package includes a proposed TSMC advanced wafer fab. The twist is that this is the same Longtan project TSMC backed away from in 2023 after local opposition. ### What actually changed? The new thing is not that a fab broke ground. It is that the government review track is back on. The bureau says public hearings were already held — one in late 2025 and another earlier this year — and now the expansion plan is moving up for formal review. That turns a dead project into a live one again. TSMC had earlier planned a fab there for technology more advanced than 2 nm, but the project was shelved in 2023 because nearby residents pushed back. The land issue was the core problem. Local reporting says the site is heavily privately owned, which makes expansion slower and politically messier than just drawing a box on a map. ## Why bring it back now? Because AI demand is eating everything. TSMC just posted record first-quarter 2026 results and said AI and high-performance computing demand pushed it to raise its full-year growth outlook to above 30%. If every advanced node and packaging line is getting booked hard, a canceled fab site starts to look valuable again. ### What would Longtan actually make? The reporting points to “angstrom-class” manufacturing — basically the generation beyond today’s 2 nm branding. That does not mean literal 0.1 nm transistors. It means the next family of leading-edge processes, the kind aimed at AI accelerators and other top-tier chips. Some reports put the potential investment around NT$500 billion to NT$600 billion, or roughly $16 billion to $19 billion. ### Why not just keep building in southern Taiwan? TSMC already did that after Longtan stalled. It pushed more advanced development into Tainan and Kaohsiung, and separate reporting says it is building a huge cluster of 2 nm fabs in Kaohsiung. But concentration has a downside — one region can become a bottleneck for labor, utilities, transport but it fits the company’s expansion pattern. ### Does Longtan matter only for wafers? No — and that is easy to miss. TSMC already has an advanced backend fab in Longtan. TrendForce has also tied Longtan to advanced packaging work for high-end processors. So this is not just about making more wafers. It could tighten the link between front-end chipmaking and the packaging technologies now needed to turn AI chips into full systems. ### What is the real constraint here? Cleanrooms are hard. But turns out local consent can be just as hard. The semiconductor story often gets told as a race of process nodes and equipment. Longtan is a reminder that the gating factor can be more like airport expansion — the technology may be ready, but the site still has to survive hearings, land assembly, and politics. ### Bottom line? Longtan’s revival does not guarantee a fab. But it does show that TSMC is still hunting for every credible place in Taiwan to add future leading-edge capacity — because in
