NVIDIA is reallocating its manufacturing capacity at TSMC from the H200 chip to the next-generation Vera Rubin architecture, a strategic pivot reflecting the company's latest assessment of the current regulatory environment. On Thursday, citing two informed sources, the Financial Times reported that NVIDIA has halted H200 chip production for specific markets and reassigned the related TSMC capacity originally intended for H200 to Vera Rubin.
NVIDIA's Chief Financial Officer Colette Kress stated during an earnings call last week that although the U.S. government has approved limited exports of H200 products, "we have not recognized any revenue to date and do not know if any exports will be approved." This adjustment indicates that NVIDIA does not anticipate large-scale sales of the H200 in the near term. Insiders suggested this move could, to some extent, accelerate the delivery and launch timeline for Vera Rubin, which is currently in high demand from tech giants like OpenAI and Google.
The H200, an earlier generation AI processor from NVIDIA, was previously positioned as a compliant product meeting U.S. export control requirements. Vera Rubin, unveiled earlier this year, is NVIDIA's latest chip architecture designed for more complex AI systems and is experiencing robust demand in high-end computing markets. According to an internal NVIDIA source speaking to the Financial Times, the company has produced approximately 250,000 H200 chips. If only limited orders are ultimately approved, existing inventory would be sufficient to meet corresponding demand.
"Rather than waiting in uncertainty, NVIDIA must pivot to targets with more certainty, especially when advanced products are in short supply," one insider commented. NVIDIA had previously aggressively pursued sales permits for the H200. Following signals from former President Trump in December suggesting potential sales approval, NVIDIA accelerated its production pace. A prior Financial Times report indicated the company had anticipated customer orders exceeding one million units, with suppliers working around the clock to prepare stock, originally planning deliveries to begin as early as March this year.
NVIDIA CEO Jensen Huang stated in early January that demand was "very strong, we have started the supply chain, H200 is on the production line." However, the approval process subsequently stalled. Despite the shift in production capacity, NVIDIA has not completely ruled out the possibility of restarting H200 production. Insiders noted that if policy conditions change, reallocating or increasing H200 supply chain capacity would take up to three months, during which existing inventory could cover demand and delivery requirements.
Comments