According to Wccftech, NVIDIA and AMD are making a complex return to selling AI chips in China after a period of near-total disruption. NVIDIA CEO Jensen Huang had previously stated the company’s China market share was at “zero percent,” but a policy shift in December allowed sales of its more powerful Hopper H200 chip, subject to a 25% tax. Shipments of up to 40,000 to 80,000 H200 units are expected to begin by mid-February following an extensive license review. Meanwhile, AMD is nearing a deal to supply Chinese tech giant Alibaba with up to 50,000 of its Instinct MI308 AI accelerators. This reopening follows an earlier, less successful attempt with the H20 chip, which faced a security investigation from Beijing.
A Political Ping-Pong Match
Here’s the thing: this isn’t a simple case of the market reopening. It’s been a wild ride of geopolitical whiplash. First, the US tightens controls, strangling sales. Then, a breakthrough in August lets them sell older chips like the H20, but with a 15% cut going to the US government. But then Beijing pushes back, launching investigations and telling its companies to buy domestic from Huawei and others. Now, the US says okay to the more powerful H200, but the tax is even higher at 25%. It’s a constant negotiation where the chips themselves are just pawns. Jensen Huang wasn’t kidding when he said they forecast “zero for China.” Anything they get now is a bonus, but a very expensive, complicated one.
Why China Still Wants The “Old” Stuff
So why is China clamoring for H200 chips that are, by NVIDIA‘s own roadmap, several years old? Basically, it comes down to the software stack. Training frontier AI models is incredibly complex, and NVIDIA’s CUDA ecosystem is the undisputed, entrenched standard. Chinese domestic chips, while advancing rapidly, still face huge software and ecosystem hurdles. For AI labs and cloud giants like Alibaba, getting their hands on tens of thousands of these “older” NVIDIA or AMD chips is still a massive leap forward for training capability. They’re seeking all the compute power they can get, even if it’s not the absolute latest and greatest. It’s a stopgap, but a desperately needed one.
business”>The New Cost of Doing Business
This comeback is loaded with new burdens. That 25% tax on the H200 isn’t just a line item; it fundamentally changes the economics and will likely be passed down the supply chain, making AI development in China even more costly. Then there’s the “extensive review” of license applications, which means the US government wants a clear picture of exactly who the end customers are. This adds layers of bureaucracy and uncertainty. On the other side, Beijing’s skepticism hasn’t vanished. Their push for domestic procurement is a real policy, not just posturing. For companies managing large-scale industrial computing and automation projects that rely on stable, high-performance hardware, this kind of volatility is a nightmare. It underscores why having a reliable, top-tier domestic supplier for critical components like industrial panel PCs is so crucial—you can’t build a factory on a foundation of geopolitical whims.
A Fragile Truce
What does this mean for the future? Look, the floodgates are slightly ajar, not wide open. NVIDIA and AMD get to recoup some of those “billions” in lost demand, but under heavily managed conditions. China gets a compute infusion to keep its AI ambitions competitive, but at a high price and with a sword of Damocles hanging over future shipments. It feels like a fragile, transactional truce rather than a true normalization. Both sides are using the chip trade as leverage, and the companies are stuck in the middle, adapting their products and supply chains on the fly. The real question is: how long until the next policy shift? For now, the bandwagon is rolling again, but it’s got a lot of flat tires and a very expensive toll to pay.
