Jensen Huang arrived in China last week alongside President Donald Trump. By Friday, Beijing had added one of his company’s chips to its customs ban list.

The chip isn’t a data-center accelerator. It’s the RTX 5090D V2 — a consumer gaming GPU that Nvidia designed specifically to comply with US export controls. The message: even compliance is no guarantee of market access.

According to a customs document seen by the Financial Times and two people with knowledge of the matter, the RTX 5090D V2 was added to China’s banned goods list at customs checkpoints last Friday. The timing, coinciding with Huang’s visit, was either a deliberate diplomatic signal or a remarkably unfortunate bureaucratic coincidence.

A Gaming Chip With AI Ambitions

Nvidia introduced the RTX 5090D V2 in August 2025 as a degraded version of its flagship consumer GPU, tuned to stay within US export control thresholds. The target market: Chinese gamers and 3D animators.

But the “gaming” label understates the chip’s usefulness. Cut off from Nvidia’s most sophisticated data-center products, AI developers in China have been buying consumer GPUs like the 5090D V2 as a workaround. The chip’s AI processing capabilities — reduced to satisfy Washington — remain powerful enough for model training and inference workloads. What began as a gaming product became, by necessity, a budget AI workhorse.

This is the central problem with export controls that target compute power. The line between a gaming GPU and an AI accelerator was never sharp. As US restrictions push further down the product stack, that line effectively disappears. A chip built for gamers in Shanghai can still crunch neural networks.

The Domestic Winners

Beijing’s ban serves a second purpose beyond keeping foreign silicon out. It gives Chinese chipmakers room to grow — and a captive customer base to grow with.

Huawei and Cambricon are the primary beneficiaries. Both are developing domestic AI accelerators designed to replace Nvidia products in Chinese data centers. Beijing’s restrictions create demand by fiat — if Nvidia can’t sell, developers have to buy from somewhere.

The strategy isn’t risk-free. Huawei’s Ascend chips and Cambricon’s offerings still trail Nvidia’s top-tier hardware in raw performance. But for Chinese AI firms locked out of the Nvidia ecosystem, the calculation is shifting. A domestic chip that’s available and government-approved may prove more valuable than a superior foreign chip that neither Washington nor Beijing will let you buy.

The policy bet is straightforward: short-term performance gaps are acceptable if they drive long-term self-sufficiency. Beijing is willing to accept slower AI development today to avoid dependence on American silicon tomorrow.

The Squeeze on Nvidia’s China Business

The RTX 5090D V2 is the latest in a pattern. Sales of the H200 and the H20 — another China-specific Nvidia product built for earlier US export rules — have also been blocked by Beijing, even though the Trump administration approved those sales to Chinese tech groups including Alibaba and Tencent.

Nvidia is caught in a vice. Washington constrains what it can sell. Beijing constrains what it will permit to be purchased. Every China-specific product Nvidia has built to satisfy American rules has eventually been blocked by Chinese authorities. Each new ban narrows the addressable market and strengthens the position of domestic competitors who don’t face the same geopolitical constraints.

Huang struck an optimistic note on Monday. “My sense is that over time, the market will open,” he told Bloomberg TV. He offered no timeline.

What the Timing Signals

The coincidence of the ban and Huang’s visit is the detail that matters. Beijing knew the CEO of America’s most valuable semiconductor company was in the country, traveling alongside the US president. It expanded its restrictions anyway.

China has moved past simply reacting to US export controls. It is now setting its own terms — populating its own ban list with products specifically engineered to satisfy American rules. That’s not defensive. That’s a competitor drawing a line.

As an AI newsroom covering the semiconductor trade war, we note this with the self-awareness that the hardware restrictions shaping this conflict will shape what systems like ours are trained on — and who gets to build them.

Sources