NVIDIA wins H200 approvals, no shipments
- Nvidia said on May 14 the United States had approved H200 sales to about 10 Chinese customers, but Reuters reported no chips had shipped. (bis.gov) - The clearest detail is the gap itself: roughly 10 approved Chinese buyers, yet zero delivered H200 units so far, according to Reuters. (bis.gov) - Next, deliveries depend on export licenses being executed and Chinese buyers deciding whether to wait, according to BIS and company reporting. (bis.gov)
Nvidia said on May 14 that U.S. approvals for H200 sales to about 10 Chinese customers had opened a path back into part of the China market, but Reuters reported that no approved H200 units had actually been delivered yet. The gap matters because the H200 is one of Nvidia’s higher-end Hopper chips, with 141 gigabytes of HBM3e memory and 4.8 terabytes per second of memory bandwidth, according to Nvidia. (bis.gov) Chinese technology groups, meanwhile, have spent months building around domestic alternatives as U.S. restrictions kept Nvidia largely out of the market. That has left the story split between a regulatory opening on paper and a supply decision that still has to happen in practice. ### If Washington approved sales, why are there still no shipments? The Bureau of Industry and Security said on January 13 it had revised its licensing policy so applications for Nvidia H200, AMD MI325X and similar chips to China would be reviewed case by case if security conditions were met. That change created a route for approvals, but it did not amount to blanket permission for unrestricted shipments. Reuters reported on May 14 that about 10 Chinese firms had been cleared to buy H200 chips, yet none of the approved units had been delivered. The distinction is operational: export approval, customer allocation, logistics and installation are separate steps, and Reuters said the process had not produced actual deliveries. (bis.gov) ### What exactly is the H200 that Chinese buyers want? Nvidia says the H200 is the first GPU in its lineup to use 141 GB of HBM3e memory and deliver 4.8 TB/s of memory bandwidth. The company markets it as a Hopper-based part for generative AI, large language models and high-performance computing. Nvidia said when it introduced the H200 that the product offered more memory capacity and bandwidth than the H100. (bis.gov) Those specifications matter for training and inference workloads that are constrained by memory size and movement of data rather than only raw compute. ### What changed for Chinese customers while Nvidia was waiting? Chinese technology companies spent the period of restricted access expanding use of domestic chips, CNBC reported on May 14. (cnbc.com) The report said Nvidia’s absence had helped fuel a homegrown chip push and that Chinese groups had reduced their dependence on imported Nvidia hardware. CNBC reported in April that Chinese semiconductor companies including SMIC and Hua Hong posted record 2025 revenue as AI demand and U.S. curbs strengthened domestic suppliers. (nvidia.com) That backdrop helps explain why an approval to buy Nvidia hardware is not the same thing as an urgent purchase order. ### Why does the delay matter even if approvals now exist? (nvidia.com) Chinese buyers choosing AI infrastructure are weighing more than benchmark performance. Reuters said the absence of delivered H200 units after approvals had pushed discussions toward practical availability, ecosystem continuity and political risk. CNBC reported that Chinese companies have been deploying local alternatives while Nvidia remained shut out. (cnbc.com) Once software stacks, engineering teams and procurement plans begin adapting to domestic chips, switching back can become a business decision that includes timing and supply certainty, not only chip capability. That conclusion is an inference from the reported shift to local deployment and the lack of deliveries. (cnbc.com) ### Does this mean Nvidia is back in China? Nvidia has a regulatory opening, not a restored market position. The January 13 BIS policy says H200-related applications are reviewed case by case, and Reuters said approved sales had still not translated into delivered product by May 14. (cnbc.com) China remains important because it is one of the world’s largest AI and data-center markets, but the current arrangement leaves Nvidia operating through approvals tied to named customers and conditions rather than broad access. That means each shipment still depends on licensing, supply and customer willingness to proceed. (cnbc.com) ### What should readers watch next? The next concrete marker is delivery. Reuters said approvals covered roughly 10 Chinese firms, and the unresolved question is when any of those customers receive H200 units rather than just export clearance. The next policy marker is BIS licensing practice. (bis.gov) The January 13 rule remains the governing framework for H200 exports to China, and any further public approvals, denials or revisions from the Commerce Department will show whether the opening becomes a shipping channel or remains a narrow exception. (cnbc.com)