Sources indicate that NVIDIA (NVDA.US) is striving to meet the robust demand from Chinese technology companies for its H200 artificial intelligence chips and has approached its foundry partner, Taiwan Semiconductor Manufacturing Company (TSM.US), to increase production volumes. Two of these sources revealed that Chinese tech firms have placed orders exceeding 2 million H200 chips for delivery in 2026, while NVIDIA's current inventory stands at only 700,000 units. The specific quantity NVIDIA intends to order additionally from TSMC remains unclear, according to the sources. A third source mentioned that NVIDIA has requested TSMC to commence production of the extra chips, with manufacturing expected to begin in the second quarter of 2026. Sources also stated that NVIDIA has decided on which H200 models to supply to Chinese clients, with pricing set at approximately $27,000 per chip. This potential order would signify a substantial expansion of H200 production, coming at a time when NVIDIA has been primarily focused on ramping up output of its newer Blackwell and upcoming Rubin chip series. Previous reports suggested NVIDIA plans to utilize existing inventory to fulfill initial orders, with the first batch of H200 chips anticipated for delivery before the mid-February Chinese Lunar New Year. These developments have sparked concerns about a potential further tightening of global AI chip supply, as NVIDIA must now strike a balance between satisfying strong demand from the Chinese market and addressing supply constraints in other regions. Earlier this month, former U.S. President Trump announced a policy allowing NVIDIA to sell its H200 chips to China, with the U.S. government claiming a 25% share of the revenue from these chip exports. However, this situation could simultaneously heighten risks for NVIDIA, as the company has not yet received formal authorization to sell its H200 chips within China. In response to requests for comment, NVIDIA stated that it continuously manages its supply chain. An NVIDIA spokesperson said, "Sales of licensed H200s to authorized customers in China do not impact our ability to supply U.S. customers," adding, "China is a competitive market where domestic chip suppliers are growing rapidly. A full export ban harms our national and economic security and only benefits foreign competitors." The demand is being driven by Chinese tech giants, with the two sources indicating that the over 2 million chip orders for 2026 predominantly originate from major Chinese internet companies, which view the H200 as a significant upgrade over currently available chips. One source mentioned that of NVIDIA's current 700,000-unit inventory, approximately 100,000 are GH200 Grace Hopper Superchips—which combine NVIDIA's Grace CPU with the Hopper GPU architecture—with the remainder being standalone H200 chips. Both models will be made available to Chinese customers, according to the source. While NVIDIA has provided Chinese clients with a general pricing range, the final price will vary based on purchase volume and specific customer arrangements, the two sources noted. They indicated that an eight-chip module is expected to cost around 1.5 million yuan, slightly higher than the now-unavailable H20 module, which previously sold for about 1.2 million yuan. However, sources said that given the H200's performance is approximately six times that of the H20—a downgraded chip NVIDIA specifically designed for the Chinese market—Chinese internet companies find the pricing attractive. A report on Wednesday suggested that ByteDance plans to spend roughly 100 billion yuan on NVIDIA chips in 2026, up from about 85 billion yuan in 2025, contingent upon NVIDIA receiving approval to sell its H200 chips in China.
Comments