China Just Quietly Unblocked Nvidia's H200 — But the Real Story Is Who Gets Cut Out

China is preparing to let its biggest AI companies buy a limited number of Nvidia’s H200 chips, easing a self-imposed blockade that had kept the US-approved processors out of the country for months — and revealing a quieter truth about who actually controls the global AI chip supply chain.

Chinese officials have told Alibaba, ByteDance, and DeepSeek they will receive permission to purchase some H200 processors, according to The Information, which broke the story on Wednesday. The companies must explain how many chips they need — and why — to get approval. The total may end up being fewer than 200,000 units, less than half of what the firms originally requested earlier this year.

🔍 THE BOTTOM LINE

The H200 was never banned by Washington — Trump granted Nvidia permission to sell it to China in December. Beijing blocked it anyway, because a flood of American-designed chips would undermine China’s push for an indigenous chip industry. Now, with AI labs starved for compute, Beijing is relenting — but on its own terms, at its own volume, and with its own political logic. The export control narrative has always been more complicated than “US restricts, China suffers.” China restricts too.

The Two-Sided Blockade

The chip story most people know is the American one: Washington restricts sales of advanced AI chips to China on national security grounds, forcing Nvidia to design compliant products like the H200. That’s true, but it’s only half the picture.

The other half is Beijing’s own restriction. China had previously blocked the import of Nvidia’s H200 chips on two stated grounds: first, that a flood of American-designed AI processors would hinder the government’s long-held goal of developing an indigenous chip industry; second, that allowing American-made chips into the local market could pose a cybersecurity risk.

So the H200 sat in limbo. The US said Nvidia could sell it. China said its companies couldn’t buy it. Nvidia’s CFO Colette Kress confirmed in May that the H200 had generated zero revenue in China. “We are uncertain whether any imports will be allowed into the country,” she said at the time. Nvidia didn’t include any China revenue from that market in its outlook.

That deadlock is now breaking — but not because China surrendered. It’s breaking because Chinese AI labs can’t wait any longer.

Why Beijing Relented

The trigger was compute hunger. According to The Information’s reporting, cited by two people with direct knowledge, labs in China have been struggling — like their US counterparts — to find enough computing power to meet growing needs to train and run new models. The difference is that US labs can buy Blackwell GPUs by the hundred-thousand. Chinese labs couldn’t even buy the H200, which is already a generation behind.

Nvidia shares climbed 3.7% to $204.12 on the news, the stock’s biggest one-day gain in more than a month. But the enthusiasm should be tempered. Fewer than 200,000 H200 chips is not a large number in AI infrastructure terms. A single data center can use more than 400,000 Blackwell processors. The allocation China is considering is roughly what one hyperscaler campus burns through in a quarter — distributed across the entire Chinese AI industry.

Meanwhile, Nvidia is already working to release its next-generation Vera Rubin lineup in the second half of this year, which Washington continues to restrict from Chinese sale. The H200 is yesterday’s chip. China is being allowed to buy yesterday’s chip, in limited quantities, after a year of diplomatic wrangling.

The DeepSeek Factor

DeepSeek’s inclusion on the approved list is the most politically charged detail. DeepSeek has been the poster child for China’s ability to build competitive AI models despite chip restrictions, releasing open-weight models that matched or beat frontier offerings from Western labs. The company was already building its own chips as a hedge against exactly this kind of blockade.

Now Beijing is telling DeepSeek it can buy Nvidia H200s alongside that domestic chip program. The message is pragmatic: self-sufficiency is the goal, but starving your best labs of compute in the meantime is not a strategy for staying competitive. The export control regime has always assumed that restricting chips would slow China down. In practice, it pushed Chinese firms to build alternatives faster — and now China is buying the restricted chips anyway, just at a volume it controls.

What This Means for the Chip War

The H200 easing is not a reversal. It’s a controlled compromise. Beijing is not abandoning its indigenous chip industry push — it’s supplementing it with capped Nvidia purchases where the alternative is falling behind. The companies approved are the ones Beijing trusts: state-adjacent giants like Alibaba and ByteDance, and national champion DeepSeek.

The broader pattern is clear. US export controls and Chinese import controls are both flexible, both negotiated, and both ultimately subordinate to the practical reality that AI compute demand is outstripping every political constraint designed to manage it. The Singapore chip-smuggling case showed what happens when the formal channels break down — chips find back doors. The H200 easing shows what happens when the formal channels open slightly — the back door traffic drops.

For New Zealand, the signal is indirect but real. Every chip that enters China through legitimate channels is one less chip driving grey-market demand through transit hubs in the region. And every chip China buys is a data point in the broader argument about whether export controls are working at all — or just reshaping the supply chain without actually constraining the technology.

❓ FAQ

Will this mean Chinese AI models get dramatically more powerful?

Not immediately. Fewer than 200,000 H200 chips is a meaningful but modest allocation — enough to relieve compute pressure at the largest labs, not enough to match the scale of US hyperscaler buildouts. DeepSeek and Alibaba will likely use the chips to maintain parity, not to leapfrog.

Is this a win for Nvidia?

Financially, modestly — the H200 is already a legacy product line. Strategically, it’s more complicated. Nvidia gets some China revenue it wasn’t counting on, but the volume is capped and the next-generation chips remain restricted.

Why did China block chips the US had already approved for sale?

Because China has its own chip sovereignty agenda. Allowing a flood of American-designed processors would undermine domestic chipmakers like Huawei’s Ascend line. Beijing’s restriction was as much about industrial policy as national security.

Could Beijing reverse this decision again?

Yes. The policy is not finalized — the exact number of chips remains under discussion, and Chinese officials could tighten or loosen the cap depending on how domestic chip development progresses and how US-China relations evolve.

🔍 THE BOTTOM LINE

China is letting its AI champions buy Nvidia’s last-generation chips in limited quantities — not because export controls failed, but because Beijing’s own import controls were starving its labs. The compromise reveals the real dynamic: both governments restrict chips, both governments negotiate exceptions, and AI compute demand outruns every political fence built to contain it. The chip war isn’t ending. It’s just learning to live with the leaks.