The Trump administration approved shipments of about 879,000 advanced AI chips from Nvidia and AMD to Chinese companies between December 2025 and May 2026, according to data the House Foreign Affairs Committee released on Thursday, November 19. Some of the approvals went to firms on the Pentagon’s list of alleged Chinese military-backed companies, which turns a licensing dispute into a national-security question. Reuters was among the outlets to report the figures, and the committee’s headline number is nearly 900,000.

This is a new development in the H200 export story that USA 24 has followed since the December 2025 reversal, not a repeat of it. The committee’s data covers advanced chips from both Nvidia and AMD, so the question reaches beyond a single product line.

Nvidia AI chips China approvals, company by company

The committee’s itemized list names four Chinese companies, each on the Pentagon’s list of alleged Chinese military-backed firms:

  • Tencent: 150,000 chips
  • Alibaba: 75,000 chips
  • Baidu: 43,000 chips
  • BYD: 4,000 chips

Together those four total 272,000 chips, about 31 percent of the 879,000 approved. Former US official Chris McGuire, cited in the Reuters-sourced report, put the share going to Pentagon-listed firms at over 30 percent, so the two estimates line up. Tencent’s allocation alone is larger than Alibaba and Baidu combined, which is the detail most likely to draw questions from lawmakers.

How 879,000 approvals compare with what companies requested

Companies had asked to ship about 3.4 million chips to China in the same six-month window, according to committee documents. Commerce therefore approved roughly one chip in four of the requested volume, or about 26 percent. The gap shows that Commerce did not approve every request, but the total still dwarfs earlier reporting.

The approvals follow a major policy shift. Trump approved Nvidia’s H200 sales to China in December 2025, reversing Biden-era restrictions on access to the most advanced chips. That decision drew bipartisan criticism from China hawks. By May, Reuters reported that about 10 companies had been cleared to buy H200s, but no deliveries had occurred.

The committee’s figures go well beyond that early count. They also cover a broader set of advanced chips than the H200 alone, which means the licensing record is larger than the public debate around one chip model suggested.

What Commerce says about its export license review

Commerce defended its process. According to Reuters, via Mahoning Matters, the department said its H200 licensing “is based on a rule published in the Federal Register in January 2026,” which adds national-security requirements for applicants. That is the government’s core defense: the rules exist, and the licensing process is supposed to apply them.

Critics are not satisfied. Chris McGuire, a former US official and expert on the US-China technology contest, put it sharply. He said the data “calls into question whether Commerce conducted any serious national security review before approving these licenses.”

Republican Brian Mast, the committee chair, defended the release itself. “This is the most detailed public disclosure in the history of the Export Control Reform Act.” That framing matters because it presents the release as a transparency step, even as it puts the approvals under a harsher light. The dispute is about what the numbers prove, and the public record so far does not settle it.

Did the approved Nvidia chips actually reach China?

Approval is not delivery, and the public record does not settle that question. The committee’s statement says, “despite receiving some licenses, we have since stopped shipping H200s to China.” That line suggests shipments have halted, but it does not say how many chips reached Chinese customers before they stopped.

Readers should treat 879,000 as approved volume, not as chips installed in Chinese data centers. The distinction matters for national-security questions, and it matters for Nvidia’s revenue.

Nvidia’s latest results frame the commercial stakes. The company reported fiscal third-quarter 2027 revenue of $115.8 billion on November 17, up 103 percent from a year earlier, with data center revenue of $107.6 billion, according to its SEC filing. Proactive, via Yahoo Finance UK reported that Hopper-generation shipments to China totaled $2.4 billion in the quarter, while China accounted for only about 2 percent of sales.

For the fourth quarter, Nvidia guided to $130 billion in revenue, plus or minus 3 percent, excluding China data center compute sales. CNBC reported that the guidance already assumed no China data center revenue, so the direct earnings effect of the committee’s findings appears limited. The larger exposure is uncertainty, which can weigh on revenue forecasts and investor sentiment. For details on the shipment stop, see USA 24’s coverage of Nvidia’s H200 halt.

A new Commerce rule targets remote access through overseas data centers

The same week, Commerce issued a separate rule that restricts Chinese companies from remotely using advanced AI chips through data centers in other countries, according to TokenPost. Shipment controls alone leave one gap open. A company that cannot import a chip could still try to rent computing capacity in a facility abroad, and this rule targets that route.

Chinese companies are already responding. Alibaba said it is assessing the effect on its global data center plans, and Tencent warned of reduced computing capacity for AI training. Those statements are early and do not quantify the impact, but they show the rule reaches cloud businesses that serve customers far outside China.

The rule’s text, scope and effective date still need confirmation from the Federal Register or official Commerce notices. Until then, readers should treat the TokenPost summary as a starting point rather than the final word.

Who gains, who loses and what to watch next

Nvidia and AMD carry the most immediate exposure. Their revenue forecasts depend on whether licenses for China keep moving, and the committee’s disclosure adds pressure to that question without changing the rules themselves.

US cloud and AI companies face a different set of effects. If Chinese rivals gain compute through approved chips, they could compete harder for AI customers in markets where US providers already operate. If they lose compute through the remote-access rule, business could shift toward US and allied providers. Alibaba and Tencent both flagged that possibility, though neither has quantified it.

Congress is the other variable. Mast’s committee has put the licensing record in public view, and lawmakers could push for tighter review or new reporting requirements. Congress also has other AI oversight proposals in play, including the SAFE AI Act outline for federal safety rules for frontier AI labs, though that proposal addresses model safety rather than chip exports.

Three questions remain open: whether Commerce will reconsider any approvals, whether Congress will act on the licensing decisions, and how many approved chips actually reached customers.

What businesses should check now

Consumers have no direct steps here. Businesses with AI supply chains, overseas data center exposure or cross-border cloud customers should review their compliance posture now:

  • Confirm whether any customer, partner or cloud tenant is a Chinese company or owned by one, and document how you screen for it.
  • Map where Nvidia or AMD accelerators are deployed, including any capacity rented to third parties abroad.
  • Review contracts for remote-access terms, and ask counsel how the new Commerce rule applies once its text is published.
  • Track the Federal Register and Commerce notices for the rule’s effective date and any change to the H200 licensing standard.

The next signals will be concrete. Commerce will need to publish the remote-access rule’s text and effective date, and the committee may press for an explanation of how each approval was reviewed. Watch for Alibaba and Tencent to quantify any capacity loss, and for Nvidia to disclose how much of the approved volume actually shipped. Those answers will show whether these approvals were a one-time reversal or the start of a larger flow of advanced compute to China.