Washington cleared this sale months ago. The holdup was never really on the American side.
ByteDance and Tencent have each received about 10,000 Nvidia H200 chips in recent weeks, the Financial Times reported, the first meaningful movement of the chips into mainland China since the Trump administration approved their export last December.
A Small Batch, After a Long Wait
The scale here is modest relative to what's actually been approved. The US has cleared roughly 10 Chinese companies to buy up to 100,000 H200 chips each, and Nvidia has been sitting on an inventory overhang of around 500,000 units.
The H200 isn't Nvidia's most advanced chip. It's two generations behind the hardware still barred from Chinese buyers entirely, which is exactly why Washington was comfortable letting it through.
The timeline here matters too. Reuters reported back in January that China had agreed in principle to import several hundred thousand H200 units, and the US granted the specific purchase approvals to ByteDance, Tencent and other firms in May. It's taken until this month for any of that approved hardware to actually start moving.
Nvidia's own supply chain is responding to the shift. Partners including Lenovo told Chinese customers last week they could resume ordering products built around the H200, a signal the channel is opening for real rather than staying a one-off shipment.
Why Beijing Doesn't Want Them Onshore
Here's the part that inverts the usual framing. Chinese regulators aren't rushing to get these chips into the country.
They're telling ByteDance, Tencent and other approved buyers to keep most of the hardware outside mainland China, routing it through Hong Kong instead. The goal is protecting the runway for domestic chipmakers like Huawei, which are still catching up to Nvidia's performance.
Beijing wants its own AI companies using foreign chips as little as possible while it builds a self-sufficient alternative, even when the foreign chips are sitting there, fully licensed and ready to deploy.
That's a notable strategic choice. China could flood its AI labs with Nvidia horsepower right now and close the performance gap with US models faster. Instead, it's accepting a slower near-term pace in exchange for not deepening its own companies' reliance on hardware Washington could restrict again at any time.
The Hong Kong Problem
There's a real logistics wall here too. Hong Kong's entire data center capacity runs on about 581 megawatts of power.
Deploying just one company's full 100,000-chip allowance would require more electricity than a fifth of Hong Kong's total data center capacity currently provides. Neither ByteDance nor Tencent currently operates data centers there at all.
That mismatch means the Hong Kong routing works fine for a 10,000-chip trial batch. It doesn't scale to the full approved volume without a buildout that doesn't exist yet.
What This Means for Miami
This adds a layer of nuance to a story MAIN has already covered: the illegal workarounds Chinese firms have used to access even more advanced, still-restricted Nvidia chips through cloud providers in Southeast Asia.
Here, the chips are legal, licensed, and still barely wanted onshore by China's own government. That tells Miami's investors and enterprise AI buyers something concrete about how China is actually managing its chip dependency: not maximizing access at every opportunity, but rationing it deliberately to protect its domestic semiconductor industry's growth curve.
For Miami businesses weighing how long cheap, competitive Chinese AI models can keep improving, that domestic protectionism is a signal worth watching alongside the pricing. Beijing's own actions suggest it expects to need Nvidia hardware for a while longer, even as it works to need it less.
