The background

The chip war is a fight over the small, powerful processors that train artificial intelligence systems. The most advanced ones are made by Nvidia, an American company, and the United States has spent four years deciding how many of them China is allowed to buy.

It started on October 7, 2022, when the Commerce Department's Bureau of Industry and Security published its first broad package of controls. Sales of high-end AI and supercomputing chips to China needed a licence. Equipment for making advanced chips was restricted too, and Americans were barred from helping advanced chip factories in China without a licence. Nvidia answered by building cut-down versions of its data center chips, the A800 and H800, designed to sit just under the new thresholds.

A year later, on October 17, 2023, Washington closed that gap. The rules switched to total processing performance, which caught the A800 and H800. Nvidia's next China chip, the H20, was built to sit below the new line. In December 2024 the controls widened again to high-bandwidth memory, the stacked memory that sits beside AI accelerators, and about 140 more entities were added to the Entity List. China answered within days by banning exports of gallium, germanium and antimony to the United States.

The pattern kept repeating in 2025. In January, the outgoing Biden administration issued the AI Diffusion Rule, a global three-tier framework for AI chips. The Trump administration rescinded it in May and warned that using Huawei's Ascend chips anywhere in the world could breach US rules. In April 2025, the H20 itself needed a licence, and Nvidia took a multibillion-dollar charge on inventory. Licences for the H20 began flowing again that summer, under a reported arrangement in which Nvidia and AMD would pass a share of China revenue to the US government — while Chinese regulators discouraged local companies from buying the chip.

Then came the first real loosening. On December 8, 2025, President Trump said Nvidia could sell the H200 to approved Chinese customers. The H200 is a big step above the H20, though still behind Nvidia's Blackwell generation. On January 15, 2026, the Commerce Department moved H200 licences from a presumption of denial to case-by-case review, with conditions: the sale must not cut into supply for US customers, the buyer must screen its own customers, and the chip must pass independent testing in the United States. A presidential proclamation the same day put a 25% tariff on such chips. On May 31, 2026, Washington confirmed the rules follow Chinese-headquartered companies wherever they operate, closing the offshore-subsidiary route for Blackwell chips.

Where things stand

Four years in, the controls are still tightening in some places and loosening in others — and the actual flow of chips is far smaller than the headlines suggest.

H200 shipments to China began in July 2026. Jeffrey Kessler, the Commerce Department's under secretary for industry and security, told Congress on July 14 that the volumes shipped so far were "very few" and "trivial." Roughly $10 billion of export licences have been approved since the conditional clearances began in December 2025, covering about ten Chinese firms including Alibaba, Tencent, ByteDance and JD.com, according to reports. But in August 2026, Nvidia wrote down $0.4 billion of H200 inventory and said Beijing had restricted licensed sales.

Enforcement is also becoming the story. In October 2026, a security firm said it had found 32 barred Blackwell B300 chips inside China Telecom's network; Nvidia called it a drop in the bucket. And this week, two reports put Nvidia itself at the center of the rivalry. The New York Times reported on a US intelligence memo claiming that Alice Huang, an executive at the Singapore-based company Megaspeed, was recruited by China's Ministry of State Security to spy on Nvidia CEO Jensen Huang. The Financial Times reported the memo also claimed the two had a romantic relationship. Nvidia told the Times that Huang's interactions with Alice Huang were "not unusual" and that the company does not share classified information with its customers. You can read our full report on [the memo's claims](/article/us-memo-claims-china-recruited-exec-to-spy-on-nvidia-ceo).

A separate Times report examined New Matrix, a Chinese company authorized by the Commerce Department to buy advanced Nvidia chips. At the office address New Matrix listed, reporters found only "a hollow gray shell," raising concerns it could be a front for a blacklisted private equity firm. Our report on [that empty office](/article/us-let-nvidia-sell-chips-to-empty-office-in-china) has the details.

What it means for the US

For American readers, the chip war cuts both ways. Keeping Blackwell and the H100 out of China protects the US lead in AI — the chips that train the most capable models stay in American data centers first. The licensing regime also turns the controls into leverage: the 25% tariff and the revenue-share arrangement mean the US government takes a cut of the China sales it does allow.

But the costs are real. Nvidia is the world's most valuable chipmaker, and China was one of its largest markets; every tightening costs it sales, and its stock has swung on each policy turn. Smuggling and front companies keep testing enforcement, and the controls push China to build its own alternatives — Huawei's Ascend chips are already the target of US warnings. Meanwhile Congress is watching closely: the same House committee that heard Kessler's testimony in July will want answers on the New Matrix case and the spy memo. The new Democratic majority in Congress, which took power after the midterm elections, has already signaled wider oversight of the administration — read [what happens next after the Democrats' win](/article/democrats-won-congress-what-happens-next-explained).

What to watch next

The most immediate date is November 10, 2026. The Affiliates Rule — which automatically extends Entity List restrictions to any company at least 50% owned by a listed party — was suspended for a year after the Trump-Xi meeting in Busan. The suspension ends on November 9, and the rule returns on November 10 unless the administration extends it. China's own suspended mineral controls expire around the same time, so the two are likely being negotiated together.

Beyond that, the Commerce Department has said it is drafting a replacement for the scrapped AI Diffusion Rule. That framework will decide whether chip access abroad is governed by one global system or a string of bilateral deals. And the practical question remains whether H200 volumes ever grow past "trivial" — or whether Beijing's restrictions and Washington's enforcement keep the licensed trade small while the rivalry moves to the next chip.