Nvidia could slow its chip orders to TSMC, according to unconfirmed supply-chain discussions reported by Commercial Times. TechPowerUp also reported that the company could reduce its orders to TSMC, but the available data does not describe a formal order to halt production or a decision announced by Nvidia.
The possible adjustment coincides with a review of AI server orders from some Asian customers and with component supplies that remain under pressure. For now, the question is not how much Nvidia will reduce its production, but whether it will modify its orders to the foundry and what effect that change would have on deliveries and manufacturing capacity.
The signal about orders is still unconfirmed
On November 10, Zero One Investment Research reported that Commercial Times had picked up industry discussions about a possible slowdown in Nvidia’s chip orders to its foundry. Taiwanese suppliers consulted by the newspaper said that, up to that point, both orders and capacity remained unchanged. The question they raised is whether order revisions will ultimately delay deliveries.
The distinction matters: a slowdown in new orders does not mean that TSMC has halted production lines or that Nvidia has canceled orders already placed. TechPowerUp presented the matter as a possible reduction in orders to TSMC, while Zero One’s report describes unconfirmed discussions and notes that suppliers had not observed changes in activity.
If capacity became available, the potential beneficiaries would be AMD and custom chip designers such as MediaTek and Alchip. Zero One cited a custom chip supplier who said its customers would absorb any TSMC capacity that became available. It does not specify how much capacity might be freed up or whether Nvidia will reduce its orders.
Asian customers face a new review
Commercial Times also reported that some AI server orders from Asian customers would need to be reviewed again. The review covers both cases Nvidia had already approved and others that were still pending. Suppliers described possible delivery delays as the issue to be resolved; they did not report actual changes to orders or capacity.
Systems with H200 account for a small share of Taiwanese assembly shipments, according to information reported by Zero One. Those shipments have shifted toward B300, GB300, and the Vera Rubin platform, which is ramping up production. This puts the order review in the context of a server market where the configurations cited already include different generations and platforms.
The supply situation is not limited to Nvidia chips. Qisda Chairman Peter Chen said at an investor conference on November 9 that the constraint for Taiwanese AI infrastructure suppliers lies in parts, not orders. According to his remarks reported by Liberty Times, memory and other key components are in short supply and their prices are rising, so some suppliers with orders are unable to deliver.
Server demand continues to grow
The possibility that Nvidia may adjust its orders comes alongside signs of expansion in other parts of the supply chain. Qisda expects its AI server revenue in 2027 to be at least five times this year’s. The company said it would begin low-volume shipments of AI servers and 1.6T high-speed switches late this year, with a significant contribution starting in the second quarter of 2027.
Three Taiwanese AI hardware suppliers also posted record monthly revenue in October, according to data reported by Zero One. Yageo recorded NT$17.6 billion, up 55% year over year and 3.1% from September. ASUS reported NT$107.62 billion, up 62.25% year over year, and Delta Electronics reported NT$74.66 billion, up 30% year over year and 14.1% month over month.
ASUS attributed its results to server shipments and deferred orders that are now being delivered and recorded. The company works with HGX and GB300 NVL72 rack designs. Yageo, for its part, pointed to AI demand alongside growth in its specialized and standard components. These figures do not clarify Nvidia’s orders to TSMC, but they show that the expansion of AI infrastructure continues to drive orders in other areas.
The review comes amid supply pressure
Parts availability also affects how much demand can be converted into deliveries. Qisda’s Chen said that a slowdown in data center construction in the United States after the midterm elections would give the supply chain time to recover. That was his assessment; it was not presented as a confirmed forecast that spending will slow.
Separately, Zero One reported data on Chinese buyers’ pressure for Nvidia systems and cards. A B300 rack that sells for about US$600,000 on the international market reportedly reached US$2.8 million in China the previous month, according to Tom’s Hardware, which cited the Financial Times. Chinese companies also reportedly bought more than one million Nvidia gaming and workstation cards this year to address the shortage.
The same report noted that Chinese customers had ordered more than six million RTX Pro 5500 cards, at least double the three million Nvidia could manufacture in 2027, according to figures attributed to those reports. These are products and markets distinct from manufacturing orders to TSMC, so the data does not prove that Nvidia will reduce its chip orders.
The next checkpoint is November 17
Zero One noted that the next opportunity to check the situation will be Nvidia’s comment on its orders on November 17. The company had not communicated a specific reduction in the cited material, and Taiwanese suppliers maintained that orders and capacity remained stable. Meanwhile, the review of orders from Asian customers raises the possibility of delivery delays, not a change that has already been confirmed.
Any capacity TSMC might free up would be of interest to AMD and companies such as MediaTek and Alchip, but there is still no figure for affected production or announced reallocation. The immediate focus is on whether Nvidia provides details about its orders on November 17 and whether the customer reviews result in delays.
Sources: TechPowerUp, Zero One Investment Research