Nvidia has approached Taiwan Semiconductor Manufacturing Company about placing a new order for its H200 artificial intelligence chips as demand from China jumps, according to sources familiar with the matter. The move signals Nvidia’s effort to capture a fast-growing market even as U.S. export controls continue to limit the most advanced AI semiconductors that can be shipped to Chinese customers. In a parallel development, the U.S. government has approved Samsung and SK Hynix to ship chipmaking tools to China for use in 2026, easing some supply chain pressure and reshaping the competitive landscape, sources said.
Nvidia’s Outreach to TSMC
People briefed on the talks said Nvidia has sounded out Taiwan Semiconductor Manufacturing Company on a potential new order specifically for H200 chips, marking an escalation in production discussions that follows a fresh wave of demand signals from Chinese buyers. The outreach, described in detail in an exclusive account of Nvidia’s contacts with TSMC, indicates that the company is not simply rolling over existing contracts but is exploring additional capacity tailored to this particular GPU line. For Nvidia, which already relies heavily on TSMC for its most advanced accelerators, securing more H200 output is critical to maintaining its lead in data center AI hardware while competitors race to bring rival chips to market.
According to sources cited in a separate report on Nvidia’s H200-focused approach to TSMC, the inquiry differs from prior orders that bundled multiple GPU generations and instead zeroes in on the H200 variant to meet accelerated delivery timelines. That shift suggests Nvidia is aligning its manufacturing pipeline more tightly with regulatory constraints, prioritizing a chip that can be shipped into China under current rules while still offering strong performance for training and inference workloads. For cloud providers, internet platforms, and state-backed AI projects in China, the outcome of these negotiations will determine how quickly they can scale new clusters and whether they can keep pace with global advances in generative AI.
Surging Demand in China
Sources quoted in an analysis of the Chinese market said demand for Nvidia’s H200 chips in China has jumped significantly, driven by a wave of local data center expansions and large-scale AI infrastructure builds. As described in a detailed account of how Nvidia’s H200 orders are being pulled forward by Chinese buyers, orders from the country are now outpacing earlier internal forecasts, prompting Nvidia to reassess how much capacity it can allocate to the region. The surge reflects a broader push by Chinese technology groups and government-linked entities to secure as many compliant AI accelerators as possible before any further tightening of export rules.
Industry sources say this uptick follows a period in which U.S. authorities eased restrictions on certain non-advanced chips, creating space for H200 variants that fall below the strictest performance thresholds to fill a critical gap in China’s AI supply chain. According to people familiar with procurement plans cited in the same report on surging Chinese demand for H200, Chinese customers are prioritizing these GPUs for new data centers in major hubs such as Beijing, Shanghai, and Shenzhen, where hyperscale facilities are being built or upgraded to support large language models and recommendation engines. For Nvidia, the rapid acceleration of orders from China could lift near-term revenue from Asia but also heightens its exposure to policy shifts in Washington that might further narrow what can be sold into the market.
U.S. Approvals for Chip Tool Shipments
While Nvidia navigates export controls on AI chips, the U.S. government has granted annual approval for Samsung to ship chipmaking tools to China, allowing the South Korean group to continue operating and upgrading its semiconductor fabrication facilities there. According to sources cited in a detailed account of how Samsung won U.S. approval to send chipmaking tools to China for 2026, the authorization covers equipment that will be used in Chinese plants in that year, giving Samsung a clearer planning horizon for investments and maintenance. The decision underscores a nuanced U.S. approach that restricts the most advanced AI hardware while still permitting some manufacturing tools to flow, particularly when they support memory production rather than cutting-edge logic chips.
SK Hynix has also received U.S. clearance for similar tool shipments to China, with sources saying the approval is likewise designated for use in 2026 and tied to specific facilities. The same report on Samsung’s and SK Hynix’s tool approvals notes that these decisions help stabilize the supply of DRAM and NAND memory produced in China, which in turn feeds into global data center and AI server builds that rely on Nvidia GPUs. For Nvidia and its customers, continued memory output from Chinese fabs reduces the risk of bottlenecks in key components such as high-bandwidth memory modules, even as restrictions remain tight on the GPUs themselves.
Implications for Supply Chain Dynamics
People familiar with Nvidia’s planning say TSMC’s potential ramp-up for H200 production could strain global capacity, since the foundry already runs at high utilization for advanced process nodes that serve multiple customers. According to the exclusive report detailing Nvidia’s new H200 order discussions with TSMC, Nvidia is weighing how to prioritize China-bound shipments relative to orders from the United States, Europe, and other Asian markets that are also racing to build AI infrastructure. If more wafer starts are earmarked for H200 units tailored to Chinese buyers, some cloud providers elsewhere could face longer lead times or be pushed toward alternative Nvidia models and competing accelerators.
The combination of rising Chinese orders and U.S. approvals for Samsung and SK Hynix tool shipments signals what sources describe as a partial thaw in certain parts of the tech trade relationship, even as core restrictions on advanced AI chips remain in place. As outlined in the account of Nvidia’s strategy to secure TSMC capacity for H200, Nvidia’s revenue from Asia is likely to accelerate if it can lock in additional production and convert the current wave of inquiries into firm purchase commitments. For policymakers, the evolving pattern illustrates how targeted controls on specific chip types coexist with a broader willingness to keep memory and manufacturing tools flowing, a balance that will shape investment decisions across the semiconductor ecosystem in the coming years.
