United States WWC APAC Desk

US Approved More Than 879,000 AI Chip Licences for China – But How Many Reached Their Buyers?

US Approved More Than 879,000 AI Chip Licences for China – But How Many Reached Their Buyers?

The US government approved more than 879,000 export licence applications involving advanced artificial intelligence chips destined for Chinese companies between December 2025 and May 2026, according to information publicly released by the US House Foreign Affairs Committee on Nov. 19, 2026.

The disclosure also identified requests covering 3.4 million chips, placing the spotlight on how Washington has balanced national security restrictions with the commercial interests of American semiconductor companies Nvidia and AMD.

The figures describe regulatory approvals, not proof that every authorised chip was shipped to China. That distinction is central to understanding the scale of the transactions and the continuing scrutiny of advanced American technology exports.

What the congressional disclosure actually establishes

House Foreign Affairs Committee Chairman Brian Mast said he had exercised authority under the Export Control Reform Act to release information concerning Nvidia and AMD export licence applications for China.

The committee’s Nov. 19 announcement stated that the applications involved requested sales of 3.4 million AI chips, with the Bureau of Industry and Security approving more than 879,000.

The announcement also identified approvals involving Chinese companies including Alibaba and Tencent. However, the public announcement does not establish that all approved transactions were completed or that all the processors reached their intended destinations.

An export licence is government authorisation for a specified transaction under applicable conditions. It is not, by itself, a shipping record, proof of delivery or confirmation that a buyer completed a purchase.

That distinction matters because the number of approved applications can differ substantially from the number of chips ultimately delivered. Commercial decisions, destination-country restrictions, compliance requirements and changes in orders can all affect whether an authorised transaction proceeds.

The committee’s disclosure is therefore significant as a record of US licensing decisions, but it should not be treated as a final count of physical chip deliveries into China.

Why Nvidia H200 and AMD MI325X exports matter

The licensing debate centres on advanced processors used in artificial intelligence computing.

Nvidia’s H200 and AMD’s MI325X are among the products covered by the US government’s revised approach to certain semiconductor exports to China. These processors can support demanding AI workloads, making access to them an important commercial issue for technology companies and a national-security concern for Washington.

On Jan. 13, 2026, the US Commerce Department’s Bureau of Industry and Security announced that applications involving the Nvidia H200, AMD MI325X and similar chips would receive case-by-case reviews, provided specified security conditions were met.

Those conditions include safeguards concerning US semiconductor supply, compliance procedures adopted by Chinese purchasers and independent third-party testing in the United States.

The policy did not create an unconditional right to sell these processors to every Chinese customer. Each transaction remained subject to the applicable licensing requirements and review.

This is the distinction between a change in licensing policy and a general reopening of the Chinese market. A company may be permitted to apply for a licence without receiving approval for every customer or transaction.

Why Washington is examining the approvals

The committee’s disclosure comes amid broader congressional scrutiny of how advanced American technology reaches Chinese companies.

The House Foreign Affairs Committee has also sought information from the Commerce Department about export licences and investigations involving Megaspeed and its affiliate Speedmatrix. Those inquiries concern possible chip diversion and the movement of advanced computing equipment through international business networks.

Such allegations and investigations must be distinguished from established findings of wrongdoing. A request for information or a congressional investigation does not, by itself, prove that a manufacturer or intermediary violated export-control laws.

The policy challenge is nevertheless clear: regulators must assess not only the immediate purchaser but also the intended end user and the eventual destination of controlled technology.

This issue has become more complicated as data centres and cloud providers allow customers to access computing power remotely, without physically importing the processors themselves.

WWC ONE MEDIA has previously examined the questions surrounding Nvidia’s export licences and whether authorised China sales can become completed transactions. The latest congressional disclosure adds a wider view of the number of applications and approvals involved in the licensing process.

New US rules extend scrutiny beyond direct shipments

A separate US regulation effective Nov. 19, 2026, addresses access to advanced computing through overseas data centres.

The Bureau of Industry and Security’s interim final rule, published in the Federal Register, introduces licensing requirements for certain advanced computing items when overseas data centres provide significant computing capacity to companies headquartered in China, Macau or other specified restricted destinations.

The rule also identifies Indonesia, Malaysia, Singapore, Thailand and Brazil for additional licensing requirements covering specified advanced semiconductor exports, re-exports and transfers.

These are not blanket bans on every chip shipment or data-centre operation in those countries. The requirements depend on the covered technology, destination, customer, ownership structure and intended use. The regulation also provides an exception for qualifying transactions involving US-headquartered companies, subject to reporting conditions.

For Southeast Asia, the changes place additional attention on supply-chain transparency and the identity of customers using advanced computing resources.

Singapore, for example, has been engaging Washington over the implementation of the new controls and the need to preserve legitimate access to American technology. WWC previously examined Singapore’s discussions with US officials over access to advanced chips.

The Philippines is not among the five destinations specifically named for the additional location-based licensing requirement. However, the separate remote-access provisions can apply to covered data-centre arrangements wherever the facilities are located.

WWC’s earlier report on the impact of new US AI chip controls on Southeast Asian data centres explains why a facility’s physical location does not necessarily settle the export-control question.

What the figures mean for the AI industry

The difference between 3.4 million requested chips and more than 879,000 approvals illustrates the scale of demand placed before US regulators during the six-month period.

It also shows why export licensing has become an important part of the commercial environment for advanced semiconductor manufacturers. Access to a large market such as China can represent a significant business opportunity, but that opportunity is subject to government restrictions and the requirements of individual transactions.

For US policymakers, the issue is whether licensing safeguards can permit legitimate trade without allowing sensitive computing technology to reach prohibited users or destinations.

For manufacturers, the challenge is more practical: an approved licence does not guarantee a completed order, and an order does not necessarily mean a shipment has taken place.

For customers and data-centre operators across Asia, the growing regulatory focus means that verifying the ultimate customer and intended use of advanced computing equipment is becoming increasingly important.

What happens next

The congressional disclosure is likely to keep attention on the Commerce Department’s licensing decisions and the effectiveness of safeguards governing advanced AI chips.

The public record reviewed for this report does not establish how many of the more than 879,000 approved chips were ultimately delivered to China. It also does not provide a complete transaction-by-transaction account of the commercial outcome of every licence.

Further disclosures, company filings or official statements may clarify how many authorised transactions proceeded, which products were involved and whether additional restrictions or compliance measures followed.

For now, the clearest conclusion is that US authorities approved a substantial number of AI chip export applications for China during the period examined, while the actual delivery picture remains a separate question.

Frequently asked questions

1. How many AI chips did the US approve for export to China?

The House Foreign Affairs Committee said more than 879,000 AI chips were approved by the Bureau of Industry and Security between December 2025 and May 2026.

2. How many chips were requested for sale to China?

The committee’s Nov. 19 disclosure stated that the applications involved requested sales of 3.4 million AI chips.

3. Which companies’ chips were covered?

The disclosure concerned AI chip export licence applications involving Nvidia and AMD.

4. Does approval mean the chips were delivered to China?

No. An export licence authorises a transaction under specified conditions. It does not independently prove that the chips were shipped, delivered or accepted by the buyer.

5. Are Nvidia H200 exports to China automatically permitted?

No. The US policy provides for case-by-case review of qualifying applications under specified conditions. Individual transactions remain subject to export-control requirements.

Get our stories first on Google

More in United States