The U.S. Cleared Nvidia's H200 for China. Six Months Later, Sales Are Almost Zero

"Loosen the controls and Nvidia will sell in China again" — that piece of conventional wisdom, intact through 2025, collapsed within six months. Washington opened the gate; Beijing shut its own. The U.S. and Chinese AI chip markets have entered a phase where no single government's policy can reunify them.

What is happening?

Here is the timeline (Brookings, CRS, press reports).

The U.S. clearance is not unconditional. Exports require verification testing inside the U.S. and end-use documentation, with each case reviewed individually. Even so, the new reality of the past six months is that the main obstacle to sales is no longer American paperwork — it is Chinese intent.

Why is China refusing to buy?

The official reason Chinese authorities give is security: the risk that American chips carry tracking features or backdoors — hidden mechanisms allowing outside control. But the substance is industrial policy.

According to Brookings, China treats its short-term compute shortfall as an acceptable cost of growing a domestic chip ecosystem. The template is its own past success. By refusing to cede its payments market to foreign players, China incubated Alipay and WeChat Pay into giant platforms. It is now running the same script for AI chips: lock out foreign firms while domestic players capture the standard (Brookings).

The designated beneficiary is Huawei. Its Ascend series has built adoption as the compute platform for the high-profile AI model DeepSeek, with SMIC providing manufacturing at 7nm-class processes. How U.S. sanctions ended up accelerating China's chip drive is covered in our analysis of China's semiconductor strategy (in Japanese), and SMIC's actual capabilities in our SMIC company deep-dive (in Japanese).

How much has Nvidia lost?

Before the export controls, Nvidia's share of China's AI chip market was put at over 90%. It has since ceded a substantial share of that market to domestic players, and Brookings concludes there is no realistic prospect of Nvidia recovering its former dominance (Brookings).

The numbers come from Nvidia's own disclosures. In its FY2026 annual report (Form 10-K), the company stated it is "effectively foreclosed" from China's data-center compute market. Revenue from China including Hong Kong fell from 19.2% of sales to 9.1% — cut in half — and FY2027 guidance assumes zero China data-center revenue (Nvidia). We unpack that filing in our Nvidia FY2026 10-K analysis (in Japanese).

For investors, this is a strange kind of comfort. Because China revenue has already been removed from the forecast, any H200 units that do sell from here are pure upside.

What do the primary sources say? The latest 10-Q disclosures

[Update, July 14, 2026] Nvidia's latest quarterly report filed with the SEC (Form 10-Q, quarter ended April 26, 2026) corroborates the unsold H200 in the company's own legal disclosures. The U.S. government began granting licenses in February 2026 for limited H200 shipments to specific Chinese customers, but revenue from this licensing program stands at zero to date. Shipments require pre-export inspection inside the U.S., and imports into the U.S. incur a 25% tariff (Nvidia 10-Q).

The same 10-Q discloses that data-center Hopper-generation shipments (H100/H200) to China were zero in the quarter. In the year-ago quarter (ended April 2025), China Hopper revenue was $4.6 billion. In one year: $4.6 billion to zero. Meanwhile, press reports suggest China may soon permit imports of fewer than 200,000 units for Alibaba, ByteDance, and DeepSeek — whether that zero moves is the next thing to watch (press reports, unconfirmed).

What does a cleared export that nobody buys actually mean?

The paradox marks a transfer of regulatory control. From 2022, it was U.S. export controls that divided the market. In 2026, it is Chinese procurement controls that keep it divided. The U.S. and Chinese AI chip markets have entered a double lockout — a state that concessions from either side alone cannot undo.

That is a heavy precedent for every other chipmaker. Winning the lobbying fight for deregulation no longer guarantees the market on the other side will come back.

What should we watch next?

First, the real capability of China's domestic chips. How far Ascend's performance and supply volume can fill Nvidia's void will set the pace of Chinese AI development. Second, the swing of the U.S. regulatory pendulum: the 25% tariff and the arrangement remitting a share of revenue to the government remain in flux, court battles included — see our Section 232 chip tariff explainer (in Japanese). Third, the durability of Nvidia's growth without China. Can it keep compounding at 65% a year on a China-zero assumption? The next earnings report is the test.

The chip performance race continues, but the deciding variable is no longer technology alone. Who is permitted to sell, and who is permitted to buy. The 2026 AI chip market rests on that double set of permissions (Brookings, Nvidia).