Winbond Electronics Accelerates Kaohsiung Fab Expansion, Customer Capacity Negotiations Already Booked Through 2030
Nashnova编辑部
Winbond Electronics (華邦電子) is merging phases two and three of its Kaohsiung Luzhu fab into a single Module B build — breaking ground in 2027, tool-in by early 2029 — as customers lock in capacity through 2030, driven by agentic-AI memory demand.
Why compress two build phases into one?
Winbond originally planned to expand in stages. Now phases two and three merge into Module B, with cleanroom construction set for 2027 and tool installation starting early 2029.
This means → Winbond is shifting from "expand when demand arrives" to "secure capacity years ahead." Visibility now stretches beyond three years.
The trigger: agentic AI — AI systems that autonomously execute multi-step tasks — is driving sustained growth in memory capacity and advanced packaging. Institutional sources say the trend "will persist for a considerable time."
How strong was this quarter?
Q2 consolidated revenue hit NT$59.843 billion, up 56.4% quarter-on-quarter and 184.7% year-on-year. EPS reached NT$5.40.
The engine was pricing: DRAM rose roughly 100% in a single quarter; Flash climbed about 43%. Gross margin followed, reaching 66.2%.
In plain terms = volume and price moved up together, pushing profitability to a record high.
Can the momentum hold into Q3 and Q4?
Q3: niche DRAM — smaller-volume DRAM aimed at specific applications — and SLC NAND are both in short supply; prices are expected to rise about 50% quarter-on-quarter. NOR Flash, pulled by aggressive stocking from major cloud clients, is forecast to gain roughly 30%.
Q4: price gains are expected to narrow to 2%–5%, but rising DRAM capacity and shipment growth should keep revenue and profit on a sequential uptrend.
This means → the sharpest pricing phase is Q3; in Q4, volume growth takes over from price elasticity as the primary driver.
What will Module B actually produce?
The product slate covers Standard DRAM, CUBE DRAM, Wafer-on-Wafer — stacking full wafers directly on top of each other — and silicon capacitors (Si-Cap), supporting 14 nm and future 12 nm DRAM processes, with EUV tools to follow.
In plain terms = from standard memory to advanced-packaging components, Module B's entire lineup points squarely at AI.
This reflects a broader shift: AI is pushing memory toward higher density and more advanced packaging. Winbond needs to lock in process nodes and capacity early — or risk falling behind when customer demand materializes.
Why is the silicon capacitor called a "third growth curve"?
Si-Cap — a high-density capacitor built on a silicon substrate — is positioned by institutional sources as Winbond's third growth engine, beyond logic and memory.
Winbond has invested roughly NT$4 billion in a dedicated Si-Cap line. Current production density sits at about 3,300 nF/mm²; the latest samples reach approximately 5,400 nF/mm². Target applications are AI advanced packaging and AI power management, with volume production expected by late Q1 2027 at the earliest.
This means → AI chips demand extreme power stability, and Si-Cap provides high-density decoupling — filtering electrical noise — at the package level. Whether Winbond can complete customer qualification and secure share before the production milestone will be the key test of this third curve.
Content is for reference only, not financial advice.