US chipmakers are getting another shot at China, and this time the comeback is happening under a very different set of rules. After months of tightened export controls and shifting political winds that squeezed sales to one of the world’s biggest AI markets, NVIDIA and AMD are preparing to ship new, China-approved AI accelerators again—though the path back is filled with extra taxes, licensing scrutiny, and fresh pressure from China’s push for domestic alternatives.
Earlier in 2025, the outlook for American AI chip vendors in China turned bleak. Policy changes tied to geopolitical tensions disrupted shipments so severely that NVIDIA CEO Jensen Huang publicly described the company’s China presence as essentially wiped out, saying its market share had fallen to “zero percent.” For companies that once relied on China as a major source of AI accelerator demand, the stoppage wasn’t just a temporary slowdown—it reset expectations for an entire region.
Behind the scenes, the next phase became a mix of lobbying and product strategy. To keep any foothold in the Chinese market, NVIDIA and AMD have had to ensure their offerings meet evolving US regulatory requirements. A key development arrived in August, when the US government outlined a framework that allowed certain AI chips to be sold into China under a condition that a percentage of shipment value would be taken as a cut—reported as 15% at the time—opening the door for limited approved products such as NVIDIA’s Hopper H20 and AMD’s Instinct MI308.
But the reopening didn’t turn into a clean return. Beijing pushed back on NVIDIA’s H20 accelerators, including a regulatory investigation focused on whether the chips could contain a security backdoor. At the same time, China reinforced a national-level priority to steer major AI firms toward domestic chips from companies such as Huawei, Cambricon, and BirenTech. That combination—political friction plus a homegrown procurement push—made it look like NVIDIA and AMD might struggle to regain meaningful momentum in China even if US licensing allowed it.
Meanwhile, China’s AI ecosystem kept moving quickly. Domestic players accelerated development schedules and promoted self-reliance, fueling new debate across the industry about how close China is getting to the cutting edge. Huang himself added to that conversation by warning that China was now only “nanoseconds” behind the US in AI progress—comments that helped underscore how much is at stake for both competitiveness and supply.
By December, the policy atmosphere shifted again. President Trump announced that NVIDIA would be permitted to sell its Hopper H200 AI chips to China, but with a higher reported tax rate of 25%, above the earlier figure tied to the H20 framework. On one hand, the approval signaled that NVIDIA could return with a significantly more capable option. On the other, the higher cost and more complicated compliance process add new supply chain and pricing hurdles, meaning any rebound will likely come with thinner margins and more operational complexity.
Now the next question is timing—and early indicators point to movement soon. A Reuters report says NVIDIA’s H200 shipments are expected to begin around mid-February, after what was described as an extensive review of license applications. The first wave is estimated at roughly 40,000 to 80,000 H200 chips, and NVIDIA is reportedly preparing to expand capacity if Chinese demand stays strong.
Even though the H200 is not NVIDIA’s newest architecture, it remains highly attractive in China for a straightforward reason: training top-tier AI models is extremely compute-hungry, and many teams still find it difficult to match NVIDIA’s performance and software ecosystem with purely domestic hardware. NVIDIA’s platform remains widely viewed as the default choice for large-scale training workloads, which keeps demand alive whenever supply becomes possible.
AMD is seeing interest as well. A report from MLex suggests Alibaba may be set to receive as many as 50,000 AMD Instinct MI308 AI chips—potentially one of AMD’s largest China-related orders, even compared with the period before export controls tightened. If that demand materializes, it signals that China’s biggest tech firms are seeking every available source of compute, balancing domestic chip adoption with any remaining opportunities to buy US accelerators that meet regulatory requirements.
For NVIDIA, the reopening matters financially as well as strategically. The company has previously indicated it had effectively ruled out billions of dollars in potential China demand under the assumption that sales would remain at zero. With H200 shipments now reportedly moving toward launch and large first-batch estimates circulating, the return to China—while constrained—could offer meaningful relief.
What happens next will depend on how smoothly licensing approvals continue, how Chinese buyers manage political and security scrutiny, and whether domestic chip options can close performance and software gaps fast enough to reduce reliance on US hardware. For now, though, the “floodgates” appear to be cracking open again—and both NVIDIA and AMD are lining up to supply China’s next wave of AI compute.
