The United States has opened the door for Nvidia’s powerful H200 artificial intelligence chips to reach Chinese buyers again, but the door is narrow and heavily guarded. New export rules allow shipments only under strict volume caps, third‑party testing, and bans on military use, turning every sale into a closely watched policy decision.
What the US just approved
- The US Commerce Department’s Bureau of Industry and Security has formally cleared Nvidia to export its H200 AI processors to China.
- Shipments are allowed only if the total number of H200 chips going to China stays at or below 50% of the volume sold to customers inside the United States.
- Each China‑bound batch must go through a US‑headquartered independent lab to verify the chip’s AI capabilities before export.
That marks a shift from earlier rules that treated advanced AI chips for China under a “presumption of denial,” effectively blocking such exports. Now, approvals move to a case‑by‑case model but with hard guardrails baked in.
The fine print: conditions on Nvidia and Chinese buyers
The new framework tries to hold Nvidia close while keeping Beijing at arm’s length.
Key conditions include:
- Nvidia must certify there is a “sufficient” supply of H200 chips available in the US before any shipment leaves for China.
- Chinese customers have to show “sufficient security procedures” and agree not to use the H200 in military or other restricted applications.
- US authorities reserve the right to scrutinize cloud providers to prevent indirect or “re‑routed” access to the chips for banned use cases.
Analysts note that enforcing know‑your‑customer checks across complex Chinese cloud and AI ecosystems will be difficult in practice. Still, the rules send a clear signal that Washington wants visibility into where each advanced chip lands.
Why the H200 matters so much
The H200 is Nvidia’s second‑most powerful data‑center GPU, positioned just below its Blackwell‑class flagships but far ahead of older AI accelerators. It targets demanding workloads such as large language models, recommendation engines, and generative AI systems that drive the current AI arms race.
At CES in Las Vegas, Nvidia CEO Jensen Huang said demand for H200 is intense globally, with rental prices in cloud data centers pushed up by limited supply. Chinese orders for H200 chips are reported to exceed 2 million units, roughly comparable to the compute footprint of a top‑tier US frontier AI lab, raising alarms among US security officials.
US–China power play over AI chips
This approval comes after years of tightening US export controls on high‑end silicon, including blanket restrictions on earlier Nvidia AI chips for Chinese buyers. The Trump administration is now trying a narrower approach: let some advanced GPUs flow, but limit how many and how they are used.
- Supporters say the policy protects US firms like Nvidia and keeps Chinese customers tied to American suppliers instead of fully pivoting to domestic chips from players such as Huawei.
- Critics argue that even with caps and testing, each shipment still boosts China’s long‑term AI capabilities, potentially undermining US strategic advantages.
Tensions did not stop on the US side. Chinese regulators have recently told customs agents that Nvidia’s H200 chips are “not permitted” to enter the country except in limited cases, effectively creating a temporary import freeze while Beijing reviews security and industrial policy risks.
What it means for Nvidia, China, and global AI
For Nvidia, the compromise preserves access to the world’s second‑largest economy, even if every container of chips now moves under a microscope. The company still faces volume ceilings, extra testing costs, and the risk that Beijing leans harder on domestic GPU champions.
For China’s tech sector:
- Big cloud and internet platforms get a partial path back to cutting‑edge US GPUs, but only in quantities capped by US domestic sales and subject to political swings in Washington.
- At the same time, regulatory hesitancy in Beijing and new pushes for homegrown chips could slow or fragment adoption of the H200, even with US approval in place.
For the broader AI ecosystem, the move highlights how access to a single GPU model can now hinge on geopolitics as much as on engineering or price. Supply chains, cloud build-outs, and AI roadmaps across Asia‑Pacific will have to account for both US export rules and China’s evolving stance on foreign chips.
(Source: techwireasia, theglobeandmail, reuters)
