Nvidia has stopped shipping advanced AI chips to China — this time the obstacle isn't Washington, it's the company's own choice. The development opens a new front in the years-long chip export standoff between the US and China.

What happened?

Nvidia's interim executive chair Colette Kress announced on Tuesday's earnings call that the company has stopped sending advanced AI chips to China. This comes despite export approval granted by the White House last December. It marks a clear break from the "stay in the China market" strategy that former CEO Jensen Huang championed for years.

Why it matters

Nvidia's relationship with China already has a turbulent history. After the first export restrictions in 2022, the company rolled out performance-throttled special editions like the A800 and H800. When those models were banned too in 2024, the H20 arrived — an even more limited chip built specifically for the Chinese market. Even the H20 was halted for a time in 2025, then re-authorized through a complicated political negotiation. December's approval was the most permissive point in this contentious process, opening the door to sales of more powerful chips like the H200.

The real question is: why isn't Nvidia selling now that it has the permit? The most plausible explanation is that demand has been dried up by Beijing. There have been repeated reports that the Chinese government has been steering domestic cloud companies and tech giants toward Huawei's Ascend series and local manufacturers like SMIC and Cambricon. This isn't a classic export ban — it's a domestic market closure carried out under the banner of supply chain security. It doesn't matter that Nvidia can sell the chip if there are no buyers.

What's changing?

Nvidia's position in China has eroded gradually over the past four years. The table below summarizes the course of that retreat:

PeriodStatus
Before 2022Unrestricted H100/A100 sales
2022-2023Sales via performance-throttled A800/H800
2024Limited sales via China-specific H20
2025Temporary H20 ban, then conditional approval
November 2026Permit in hand, no sales

This table shows the issue is no longer a one-sided US embargo. For Nvidia, the outcome is the same either way: data center revenue from China has effectively dropped to near zero. Nvidia isn't the only loser here — in the short term, domestic players like Huawei and SMIC are filling the market, though these chips haven't yet caught up to Nvidia's top-tier products in performance. AI companies in China may be forced to squeeze more optimization out of weaker hardware, which could affect model training costs and timelines.

What's next?

There are two separate scenarios worth watching. The first is that this represents acceptance of a permanent market loss — Nvidia may be writing off China revenue and leaning on growth from data center demand in the US, the Middle East, and Southeast Asia instead. The second is that this move is a preemptive position taken against tougher export legislation being debated in the US Congress; in that case, Nvidia may be keeping the option open to return to China once pressure eases. The performance trajectory of Huawei's Ascend 910 series and any shift in capital spending direction from Chinese cloud companies will help clarify which scenario is playing out.

The most realistic reading here is that the decision originates more from Beijing than from Washington. Not selling while holding a valid permit points to a retreat driven by a lack of buyers rather than sanctions — and that's the kind of market loss Nvidia can't undo simply by securing official export approvals.