Jensen Huang dropped a bombshell at GTC 2026 in San Jose — and it was not the one everyone expected. Not Vera Rubin, not the next GPU generation. Instead, news with far greater geopolitical weight: NVIDIA is producing H200 chips for China again.
Ten Months of Silence
For roughly ten months, NVIDIA’s advanced AI chips were blocked from the Chinese market. US export restrictions and Beijing’s countermeasures had ground trade to a near halt. For NVIDIA, that meant losing a massive market — China is the second-largest buyer of AI hardware after the US.
What Changed
Huang said at GTC that NVIDIA has now received licenses for many Chinese customers. Purchase orders are already in hand, and manufacturing is ramping up. Both Washington and Beijing have apparently given the green light.
But with strings attached: the US takes 25% of chip sale proceeds, shipments are capped, and all sales must go through third-party verification. Not a free market — controlled trade.
Why This Matters
The H200 is NVIDIA’s most powerful inference chip before the Vera Rubin generation, still carrying the workload in data centers worldwide. For Chinese AI companies like Baidu, Alibaba, and ByteDance, access to these chips was existentially important — and their absence accelerated the rise of domestic alternatives like Huawei Ascend.
Whether NVIDIA can regain lost ground remains to be seen. China’s chip industry made massive strides during the ten-month blockade. But the restart signals something important: neither side wants a complete decoupling.
What It Means for the AI Industry
More H200 chips in China means more compute for Chinese AI models. That will further intensify the global race for the best models. At the same time, it shows that technology embargoes have limits as a geopolitical tool — sooner or later, the market dictates.
Sources: