A new report says the Chinese government has given DeepSeek conditional approval to buy NVIDIA's H200 AI chips, with the details of those conditions still being finalised. The same reporting also says three other major buyers in China have been cleared under similar terms: ByteDance, Alibaba, and Tencent. Collectively, they were said to be approved for more than 400,000 H200 units.

That "conditional" part is doing a lot of work here. The story isn't simply "China can now buy chips." It's more like: approvals are possible, but only inside a policy box that keeps changing on both the China side and the US side.

Why it's conditional in the first place

From what Reuters described, China's approvals come with regulatory conditions overseen by the country's economic planning apparatus, and some terms were still being negotiated at the time of reporting.

One idea that has come up in related reporting is a "bundle" approach: if a company imports H200 chips, it may also be expected to buy a certain quota of domestically made chips as part of the deal. If that sounds like industrial policy 101, that's because it is: keep the AI race moving forward, but don't let local chip efforts wither on the vine.

The other half of the story: US export licensing

Even if Beijing says yes, the export still sits under US rules.

In January 2026, the US Commerce Department's Bureau of Industry and Security (BIS) announced a revised license review policy for certain semiconductors going to China. The BIS release describes conditions applicants need to demonstrate (for example, compliance procedures and independent third-party testing in the US).

Reuters also reported around mid-January 2026 that the US was allowing exports of the H200 to China with conditions, including third-party review requirements.

So when Jensen Huang says he hadn't been formally notified and that licensing was still being finalised, that fits the reality of how multi-layered this is: approvals can exist "in principle," while the paperwork and enforcement details are still messy.

Why DeepSeek is always in the spotlight

DeepSeek grabbed attention because it positioned itself as "high capability, surprisingly low cost." The original narrative (that a strong model could be trained on a relatively small budget) is exactly the kind of claim that triggers debate in the AI world, because it challenges the assumption that only mega-spenders can compete.

But there's also been pushback. Some reports and analyst commentary have argued that the headline numbers may represent only a slice of the true costs (for example, focusing on one portion of compute rather than full infrastructure, R&D, and total cost of ownership).

This matters because it changes how people interpret the H200 purchase story:

Why the H200 specifically is such a big deal

The H200 sits in the category of chips that can materially change how fast large AI models can be trained and served. That's why it keeps showing up in trade headlines: it's not just another GPU; it's a capability multiplier. Reuters' reporting frames it as part of a broader reality China is navigating: domestic chips are improving, but the top-tier NVIDIA stack still has a meaningful performance edge for many workloads.

What to watch next (the practical version)

Final thoughts: this is a controlled "release valve," not a free-for-all

If the Reuters reporting is accurate, China's conditional approval for DeepSeek (and other big names like ByteDance, Alibaba, and Tencent) isn't some sudden policy flip where everything is open again. It reads more like a controlled release valve: keep the country's AI progress moving, but do it in a way that still protects local chip ambitions and stays within the constantly shifting boundaries of export licensing.

And that's the real headline here. The H200 isn't just another GPU purchase, it's a signal of what China wants right now: practical AI horsepower, fast. At the same time, the "conditional" label is a reminder that politics, compliance, and supply chains can still slow the whole thing down even after approvals exist on paper.

So the most realistic takeaway is this: expect progress, but expect friction. The next few weeks will matter more than the headline number of "400,000 chips," because the real story will be whether those approvals translate into shipments, deployments, and measurable AI momentum, or whether conditions on both sides turn it into a long, slow negotiation disguised as a tech purchase.