Winbond Approves NT$73B Kaohsiung Fab Expansion, Targeting 10 Strategic Customers
nashnova research
Winbond's board approved NT$73.1 billion (~US$2.2 billion) to expand its Kaohsiung fab and plans to sign long-term deals with 10 core customers securing capacity from 2029 to 2032 — the company's largest single expansion, betting on a structural memory shortage.
Where does the money go?
The bulk funds Module B construction plus prepayments on long-lead equipment — lithography, metrology, and specialty etch tools — some with lead times of 18 months to over two years.
Module B breaks ground in late December 2026; cleanroom completion and equipment installation are targeted for 2028–2029.
Designed capacity: 50,000–60,000 wafers per month, more than double Module A at full output. This means → one new module roughly doubles Winbond's entire Kaohsiung capacity.
Where does Module A stand now?
Module A currently runs at 16,000 wafers/month and is expected to reach full capacity of 24,000–25,000 wafers by end of 2026, with cumulative investment near NT$150 billion.
Pilot production on the 16 nm node — one generation more advanced than the current 20 nm process — has started; monthly wafer starts are set to rise from 2,000 to 16,000, lifting output by roughly 50%.
This means → the 16 nm node will become Winbond's workhorse process by H2 2027, while DDR3 stays on 20 nm for long-term contract customers.
The company is also mapping a path to 14 nm and the introduction of EUV lithography — a chipmaking technique that uses extreme-ultraviolet light to pattern circuits, currently supplied only by ASML.
How is the expansion funded?
Winbond's Q3 free cash flow came in near NT$30 billion; management says existing cash flow is sufficient to launch the build-out.
The company has also issued domestic convertible bonds and may raise additional financing as needed.
The NT$73.1 billion is expected to be paid out through end-2028. In plain terms = cash on hand covers the start; the rest is spread over several years of operating cash flow and bond proceeds.
How are customers locked in — and will prices keep rising?
President Chen Pei-Ming expects contract prices to rise another 10%–20% in Q4 2026, with the memory supply shortage intensifying through 2027–2028.
Before Module B comes online in 2029, Winbond plans to sign "Partner Long-Term Agreements" (PLTAs) with 10 core customers: clients must prepay or put down deposits to secure dedicated capacity, and Winbond can use those funds for equipment purchases.
More than 10 customers have expressed interest, but Winbond will screen for those whose product needs match its long-term capacity plan. This means → Winbond is not running a first-come-first-served model — it is selecting customers, accepting only long-term orders that fill the line.
What do the latest earnings numbers show?
Custom Memory Solutions (CMS) revenue grew 50% QoQ and 620% YoY in Q3; average selling prices rose nearly 60% QoQ and over 500% YoY.
Flash revenue climbed 34% QoQ and 255% YoY, driven primarily by price increases.
This reflects a supply-demand gap that is already showing up in current-quarter prices and revenue — not just a forward expectation.
What does NOR Flash have to do with AI servers?
Chen estimates that Nvidia's latest server racks can carry over 600 NOR Flash chips each, spanning 512 Mb, 1 Gb, and 2 Gb high-density products.
Winbond's share of this segment could exceed 60%, potentially reaching 70%. In plain terms = every new AI server rack built means several hundred chip sales for Winbond — turning AI compute expansion directly into its own shipment volume.
Whether Module B comes online on schedule in 2029 and successfully locks in strategic customers will be the most critical milestone of this expansion cycle.
市场有风险，内容仅供研究参考，不构成投资建议。