VSMC's Singapore fab opens already sold out

The AI buildout is not just eating leading-edge chips. It is eating the boring ones. VSMC opened its first 300mm fab in Tampines, Singapore this morning, and Nikkei Asia reports the first-phase capacity is already sold out — before volume production has started. The plant makes 40nm to 130nm specialty chips, which is to say the semiconductor supply chain squeeze has reached the nodes that power your card reader, not just the ones in a data center.

What actually happened

Per VSMC's announcement, the fab is a joint venture between Vanguard International Semiconductor and NXP Semiconductors. Groundbreaking was December 2024; 22 months later it is open, which is fast for a 300mm plant. It is in risk production now, with volume production scheduled for Q1 2027 and full capacity — roughly 44,000 wafers per month — not arriving until 2029. About 1,600 jobs at full run rate.

The process range is 130nm down to 40nm, supporting mixed-signal, power management, analog, and interposer applications across HPC, mobile, automotive, industrial, and consumer markets. None of that is a GPU. All of it is what sits around a GPU, or inside anything with a power supply and a sensor.

Nikkei's contribution is the demand picture: capacity sold out on the AI infrastructure boom, and VIS already eyeing a second facility. Note the gap that creates. A fab announced sold out in 2026 does not add meaningful supply until 2027, and does not reach its full 44,000 wafers until 2029. VSMC's own release does not mention pre-sold capacity; that reporting is Nikkei's.

Why tight legacy-node supply matters for your business

Specialty nodes are where your hardware actually lives. Power management, analog, mixed-signal — that is the silicon in POS terminals, barcode scanners, label printers, thermostats, cameras, and every controller board in a warehouse. When AI infrastructure absorbs that capacity, the shortage does not show up as a headline. It shows up as a twelve-week lead time on a device you reorder without thinking.

Two to three years of lag is the number to plan against. Sold out today, volume production Q1 2027, full capacity 2029. Anyone telling you hardware pricing normalizes next quarter is guessing. Budget device refresh cycles assuming replacements cost more and arrive later than they did last year.

Single-SKU dependency is the actual risk. We have seen a client's whole floor operation stall because their scanner model went end-of-life and the replacement needed different drivers. If one discontinued part number can stop you taking orders, that is a software problem you can solve now: keep your integrations device-agnostic and test a second model before you need it.

Buy the spare while it is boring. If a piece of hardware is load-bearing for revenue, one on the shelf is cheaper than an expedited order during a shortage. This is unglamorous procurement advice and it has saved more weekends than any architecture decision.

Watch interposers as the tell. Interposer capacity sitting in the same fab as your power-management chips is exactly how AI packaging demand crowds out ordinary industrial parts. It is the clearest signal that these two markets now compete for the same wafers.

Sources: VSMC via GlobeNewswire, Nikkei Asia.