Nvidia is moving urgently to ramp up production of its H200 artificial intelligence chips after Chinese technology companies placed orders exceeding 2 million units for delivery in 2026, far outstripping the U.S. chipmaker’s current inventory, according to people familiar with the matter.
The sources told Reuters Nvidia has approached Taiwan Semiconductor Manufacturing Co (TSMC) to expand production capacity, with manufacturing of additional H200 chips expected to begin in the second quarter of 2026. Nvidia currently holds about 700,000 H200 units in stock, leaving a significant shortfall relative to Chinese demand.
The scramble underscores both the intensity of China’s appetite for advanced AI hardware and the delicate balancing act Nvidia faces as it tries to serve Chinese customers while managing tight global supply chains and regulatory uncertainty.
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).
The discussions between Nvidia and TSMC, as well as the scale of Chinese demand, had not previously been reported. The exact number of additional chips Nvidia intends to order remains unclear, but one source said the company has already asked TSMC to begin preparations for increased output.
The potential expansion comes at a sensitive moment for Nvidia, which is simultaneously ramping up production of its newer Blackwell architecture and preparing for its next-generation Rubin chips. The H200, based on Nvidia’s Hopper architecture and manufactured using TSMC’s 4-nanometre process, is a previous-generation product, but one that remains highly attractive to Chinese buyers due to performance gaps in the domestic ecosystem.
Chinese technology companies, particularly large internet firms, account for the bulk of the more than 2 million chips ordered for 2026, according to two sources. These companies see the H200 as a substantial upgrade over hardware currently available to them, especially after restrictions eliminated access to Nvidia’s H20, a downgraded chip previously designed for the Chinese market.
Of Nvidia’s existing 700,000-unit inventory, around 100,000 are GH200 Grace Hopper superchips, which combine Nvidia’s Grace CPU with its Hopper GPU, while the remainder are standalone H200 chips. Both variants are expected to be offered to Chinese customers.
Pricing discussions indicate Nvidia plans to sell the H200 to Chinese clients at around $27,000 per chip, though final prices may vary depending on order size and customer arrangements. An eight-chip H200 module is expected to cost about 1.5 million yuan, compared with roughly 1.2 million yuan for the now-unavailable H20 module.
Despite the higher price, Chinese firms reportedly view the H200 as good value, given that it delivers roughly six times the performance of the H20. The pricing is also estimated to be about 15% cheaper than grey-market alternatives, which currently sell for more than 1.75 million yuan per module.
ByteDance alone is expected to spend around 100 billion yuan on Nvidia chips in 2026, up from about 85 billion yuan in 2025, if Chinese authorities approve H200 imports, according to a report by the South China Morning Post.
However, regulatory uncertainty continues to hang over the entire plan. While U.S. President Donald Trump’s administration recently allowed exports of the H200 to China, subject to a 25% fee, Beijing has yet to formally approve shipments.
Chinese officials are still weighing whether allowing large-scale imports of advanced foreign AI chips could slow the development of the country’s domestic semiconductor industry. One proposal under consideration would require Chinese buyers to bundle each H200 purchase with a specified ratio of locally produced chips, a mechanism aimed at supporting domestic suppliers.
Nvidia acknowledged managing its supply chain actively but downplayed concerns that China-focused sales could disrupt availability elsewhere.
“Licensed sales of the H200 to authorised customers in China will have no impact on our ability to supply customers in the United States,” a company spokesperson said.
The spokesperson added that China remains a highly competitive market with rapidly advancing local chipmakers, arguing that blocking U.S. exports entirely would undermine American economic and national security interests while benefiting foreign competitors.
TSMC declined to comment, and China’s Ministry of Industry and Information Technology did not respond to requests for comment. The sources cited by Reuters requested anonymity because the talks are private.
If Nvidia proceeds with a significant production ramp-up, it could further tighten global AI chip supplies, especially as demand continues to surge across data centers, cloud providers, and AI developers worldwide. At the same time, the company’s deepening exposure to China leaves it vulnerable to policy shifts on both sides of the Pacific, reinforcing the geopolitical risks now embedded in the global semiconductor industry.