US Chip Export Controls on China: H200 Rule Analysis

Economic Warfare & Trade Policy Tech & export controls

Photo: A researcher holds a finished silicon wafer at Lawrence Livermore National Laboratory. U.S. Department of Energy, public domain, via Wikimedia Commons.

U.S. semiconductor export controls on China changed direction in 2026. After three years of tightening, the Commerce Department’s Bureau of Industry and Security (BIS) on Jan. 15 opened a path to license Nvidia’s H200 and similar AI chips for Chinese buyers, under conditions and with a 25% U.S. government take. Beijing then discouraged the purchases. This investigation reviews the rules in the Federal Register, what has been shipped, how China’s AI chip market has shifted, and whether the controls are meeting the goals Washington set.

Labels used in this report: Verified fact Official claim Expert interpretation Estimate Forecast

Executive summary

The January rule moved one class of chips from a presumption of denial to case-by-case review; the most advanced chips remain barred.
China’s response limited the effect: reports in January said customs officials blocked H200 shipments and firms were told not to buy unless necessary.
Industry data show Nvidia losing share in China while Huawei and other domestic designers gain, the outcome both critics and supporters of the controls argue over.
A wider rule covering affiliates of listed companies is suspended until Nov. 9 and will return the next day unless BIS acts again.

Timeline

Nov. 17, 2023: A license requirement takes effect for advanced computing items going to entities headquartered in China, Macau or other Country Group D:5 destinations, wherever they are located. Verified fact (BIS guidance)
January 2025: The AI Diffusion Rule creates a worldwide license requirement. In May 2025 BIS says it will not enforce the rule’s new compliance requirements.
Sept. 30, 2025: BIS publishes the Affiliates Rule, applying Entity List restrictions to any company at least 50% owned by listed parties.
Nov. 10, 2025: BIS suspends the Affiliates Rule for one year; the rule is written to return on Nov. 10, 2026.
Dec. 8, 2025: The President announces that H200 and similar chips will be allowed to approved Chinese customers.
Jan. 15, 2026: The BIS rule takes effect. A Section 232 tariff of 25% on certain semiconductors is imposed the same week.
May 31, 2026: BIS guidance confirms that the 2023 headquarters-based license requirement is still enforced.

The policy, explained

The January rule covers chips with a total processing performance below 21,000 and total DRAM bandwidth below 6,500 GB/s, such as the Nvidia H200 or AMD MI325X. Exporters must certify that U.S. supply is sufficient, that production for China will not divert foundry capacity from U.S. customers, that the buyer has adequate security procedures, and that each chip passes independent third-party testing in the United States. Verified fact CNBC reported that shipments to China are capped at 50% of the volume shipped to U.S. customers. The chips must enter the United States for testing, where the 25% tariff applies, which is how the government collects its share. The rule keeps a presumption of denial for exports to China-headquartered entities located outside China.

Affected countries

China and Macau are the direct targets. Taiwan is involved because Taiwanese foundries manufacture the chips. Any country hosting subsidiaries of Chinese firms is affected by the headquarters rule: BIS’s May guidance says a license is required for these items “to all destinations outside the United States” when the end user’s ultimate parent is headquartered in China or Macau. The Entity List itself, which we track on the data dashboard, had 1,103 entries with an address in mainland China and 230 in Hong Kong in the screening file we downloaded on Oct. 22. Verified fact

Economic mechanisms

Export controls work by denying access to the most capable hardware, which raises the cost and time needed to train and run large AI models. Licensing an older but still capable chip changes that calculation: U.S. sellers regain revenue, while Chinese buyers get an alternative to domestic chips. Experts quoted by the Guardian were split. Supporters said H200 sales could slow China’s own chip development by keeping companies dependent on U.S. technology; opponents said the chip is powerful enough for military uses. Expert interpretation

Relevant statistics

China data-center AI accelerator market, first half of 2026 (IDC, as reported by Reuters)
Measure	Value
Total AI accelerator shipments	2.008 million units
Nvidia shipments and market share	678,000 units, down 44% from a year earlier; 34% share
Huawei shipments and share	572,000 units, up from more than 300,000; 28% share
GPU shipments	991,000 units, down 26%
Non-GPU accelerators	1.016 million units, up nearly 80%

Estimate Source: Reuters, citing an IDC report (Oct. 12, 2026). IDC did not say whether its figures include barred chips such as the B200 and B300, which Reuters reported have been smuggled into China.

Stakeholder impacts

Nvidia had said the Chinese market could be worth $50 billion a year, CNBC reported, and its chief executive, Jensen Huang, has argued the curbs ceded that market to Huawei, a company view rather than a finding. Suppliers paused H200 parts production in January after Chinese customs blocked shipments, the Financial Times reported, according to the Guardian; Reuters could not immediately verify that report. Chinese chip designers gained: nine of the 11 domestic chipmakers IDC tracks increased shipments. The U.S. Treasury collects tariff revenue on chips routed through U.S. testing, though no public figure for that revenue has been released.

Geopolitical objectives

BIS framed the January change as necessary “to ensure the national security benefits of U.S. leadership in artificial intelligence,” while keeping controls on the most advanced chips. Official claim Under Secretary Jeffrey Kessler said export controls “should evolve with changes in technology, while protecting national security.” China’s apparent objective, judging by reported guidance to its companies, is self-sufficiency in AI chips; Beijing has not published a formal ban, and sources cited by Reuters and the Guardian said the authorities gave no reasons for their directives.

Policy effectiveness

Keeping China behind at the frontier: partly met. Blackwell-class chips remain barred, but smuggling reports and Huawei’s gains show limits. Restoring U.S. sales in China: largely not met so far. Reuters reported in May that the U.S. had cleared H200 sales to 10 Chinese firms, and Bloomberg reported Nvidia’s first H200 sales to China in late August, but IDC’s figures show Nvidia’s China shipments falling sharply in the first half. Slowing China’s domestic industry: the evidence points the other way, though the rise of Huawei and others began before 2026, so the January rule cannot be isolated as a cause. These are correlations across overlapping policies on both sides.

Unintended consequences

The licensing route made U.S. revenue partly dependent on Chinese import approval, giving Beijing a veto over how much the policy delivers. The pass-through tariff design also means Taiwan-made chips cross the Pacific twice. And the headquarters rule leaves cloud and data-center operators abroad facing compliance questions; BIS said in May that bona fide data-center operators need not stop using equipment they already hold “until further notice.”

Future scenarios

Nov. 10, 2026: the Affiliates Rule returns as written unless BIS publishes a further suspension; the trade truce extension to Jan. 10, 2027 makes a new suspension possible but not certain.
Chinese approval: Beijing allows larger H200 purchases as part of the trade arrangement, lifting Nvidia’s China sales.
Domestic substitution: IDC expects total AI accelerator shipments in China to grow 19% this year while GPU shipments fall 9%. Forecast

Additional details

Huang said in January that he expected purchase orders rather than formal announcements from China. Alibaba’s T-Head held 7% of the Chinese accelerator market in IDC’s count, Cambricon 5%, Baidu’s Kunlunxin 4% and Hygon 3%. For U.S. chipmaking capacity, see our map of AI chip production and new U.S. chip plants, and for allied subsidies, Japan’s funding for GlobalFoundries fabs in New York.

Sources

Like this story? Get the newsletter

The day's Boston news, weather and local business in one short email, with links to the full stories.

Latest

All latest news →

US Chip Export Controls on China: The H200 Reversal Explained