← Voice corpusEdition 2026-10-10, preview.
XQuote

Source item

Open original source ↗
My understanding is that TSMC always takes a highly cautious approach to capacity expansion. Unless there's a dramatic industry reversal in the short term, it's not easy to see significant order reductions (variations in the low to mid-single digits are considered reasonable, as no one can predict future demand with pinpoint accuracy.) Last year, at the height of AI and Nvidia market optimism, Nvidia pushed TSMC to expand its monthly CoWoS capacity to 100,000 wafers—a request that TSMC rejected. Incidentally, using this as an example, if outsiders had obtained Nvidia's optimistic request—a common practice when customers push for expansion during bullish markets—some market participants might overestimate TSMC's capacity or Nvidia's orders based on the leaked information, generating market chatter.

Quoted context

These are the quoted author’s words.

I conducted an industry survey on TSMC and Nvidia with the following takeaways: 1. TSMC's CoWoS expansion plan remains unchanged (gradually increasing quarter by quarter, targeting approximately 70,000 wafers per month in December 2025). 2. Nvidia's CoWoS orders at TSMC for 2025 are approximately 370,000 wafers (variations within low single digits are considered normal range), showing no significant changes since Q4 2024. Market rumors suggest that Nvidia's 2025 CoWoS wafer allocation plan at TSMC has been reduced from over 400,000 to 380,000 wafers, which differs from my takeaways. Possible explanations include: 1. Some market participants had overly optimistic expectations for Nvidia's 2025 wafer allocations. 2. Some market participants were too optimistic about TSMC's overall 2025 CoWoS capacity, consequently overestimating Nvidia's orders. 3. Nvidia's significantly accelerated transition timeline between B200 and B300 series products led to production plan adjustments, which were misinterpreted as Nvidia reduced CoWoS wafer orders.

mingchikuo ↗
Library as of 2026-10-22