US Mulls Strict 75,000 Nvidia H100 GPU Cap for Chinese Firms Amid Escalating Trade Tensions

US authorities are contemplating a significant clampdown on AI chip exports to China, proposing to restrict individual Chinese firms to just 75,000 Nvidia H100 GPUs annually. This limit falls short of the higher volumes requested by Chinese buyers, reflecting ongoing efforts to curb Beijing's access to cutting-edge American technology. The proposal emerges against a backdrop of fluctuating policies: while the Trump administration recently lifted bans on certain Nvidia chips like the H20 and H200 to resume shipments and recover billions in revenue, new measures aim to prevent over-reliance and potential military applications.

The restrictions matter deeply for Nvidia and the broader tech sector, as China represents a massive market driving AI hardware demand. Capping H100 exports—Nvidia's high-performance chips critical for AI training—could slash potential sales, even as 'green-zone' alternatives like the H20 keep some revenue flowing. This balances national security concerns, such as limiting China's AI advancements for surveillance or military use, with economic pragmatism, ensuring US firms stay ahead while Chinese companies pivot to domestic alternatives like Huawei's Ascend chips. Tensions are heightened by reports of workarounds, including cloud rentals from Southeast Asia, prompting bipartisan legislation to close such loopholes.

Looking ahead, the cap could reshape global AI competition. If implemented, it might accelerate China's push for semiconductor self-sufficiency by 2027, boosting local innovation while challenging Nvidia's dominance—Chinese data centers still rely heavily on Nvidia's CUDA platform. Commerce Department negotiations, possibly tied to rare-earth mineral supplies, will determine final rules, with case-by-case licenses and volume limits like 50% of US shipments under scrutiny. Nvidia's recovery of $4.5 billion in inventory hangs in the balance, as does the delicate US-China tech truce.