UMC Raises Capex to $2 Billion, Sets First AI Revenue Target
N.R. Finch
UMC (TWSE: 2303) raised its 2026 capex from $1.5 billion to $2 billion and set its first AI revenue target — roughly $300 million this year, aiming to cross $1 billion within three years. The mature-node foundry is betting its next decade on AI infrastructure.
Where does the extra $500 million go?
Capex rises from $1.5 billion to $2 billion, driven by stronger-than-expected AI demand and faster customer ramp schedules.
The board separately approved nearly $5 billion in new investment for phased expansion in Singapore and Taiwan, paced by customer commitments.
This means → UMC is not patching a short-term budget gap. It is writing AI infrastructure into its five-to-ten-year capital plan.
How does UMC define "AI revenue"?
UMC redrew the line: any product whose end use is an AI data center or AI infrastructure now counts as AI revenue — not just GPU or AI accelerator work.
This means → high-speed interconnect chips, PMICs — power management ICs that regulate voltage for chips — FPGAs — field-programmable gate arrays, chips whose logic can be rewritten — silicon photonics, and advanced packaging all fall inside the new scope.
In plain terms = "AI revenue" used to mean "parts that go into a GPU." Now it means "parts that end up in an AI rack." The definition is broader, but it also reflects genuine demand spreading beyond accelerators.
Where are the new fabs, and what do they make?
Singapore's P4 fab will focus on a silicon photonics platform. Taiwan's STSP 12A P7 and P8 fabs will center on advanced packaging.
Each new fab takes roughly 20 months to build; the earliest revenue contribution is 2028–2029.
This means → the money is not chasing today's orders. It is a bet that AI infrastructure demand for optical interconnects and packaging will surge two to three years out.
How full are existing fabs running?
12-inch fab utilization tops 90%; 8-inch fabs hold at roughly 85% — both above year-ago levels.
12-inch gains are driven by 22/28 nm and specialty nodes. 8-inch fabs are sustained by industrial-control, automotive, and partial consumer restocking.
In plain terms = the advanced lines are near full. The legacy lines are kept busy by traditional customers restocking — two completely different demand engines.
How is this cycle different from the last chip shortage?
UMC is explicit: this round is not a broad-based rebound. Demand is driven by AI investment and is gradually spilling into other end markets.
Smartphones, PCs, and general consumer electronics have not fully recovered; end-market demand is visibly split.
This reflects a pivotal call — whether the $1 billion AI revenue target is met within three years will be the real test of the thesis that "AI pull replaces broad recovery."
Content is for reference only, not financial advice.