A person in a textured jacket stands with a computer chip featuring a visible logo, set against a background resembling the Chinese flag.

U.S. Expands China Chip Crackdown as Nvidia Faces 10-Day Revenue Probe Deadline

U.S. Tightens Export Rules to Block China’s Remote Access to Advanced NVIDIA AI Chips

The U.S. government is moving to close major loopholes in its semiconductor export controls, targeting the ways restricted artificial intelligence chips may still be reaching Chinese companies through overseas data centers, cloud access, and third-party intermediaries.

The latest action comes from the U.S. Department of Commerce’s Bureau of Industry and Security, which has introduced new licensing requirements aimed at preventing China and Macau-based entities from accessing advanced AI computing power, even when the physical chips are located outside their borders.

The move is especially significant for high-end NVIDIA AI chips, which remain central to global AI development and data center expansion. While China has been restricted from directly purchasing many of these chips, U.S. officials are increasingly concerned that companies can still gain access by renting compute capacity from foreign data centers or routing purchases through partner countries.

New rules target remote access to AI data centers

The new interim rule focuses on what the Commerce Department describes as remote access to advanced AI chips and servers.

In simple terms, the U.S. is no longer only regulating who can buy the chips. It is also trying to regulate who can remotely use large clusters of those chips.

Under the new framework, companies that want to export, transfer, or install advanced U.S.-origin AI chips in data centers abroad may need special U.S. government authorization. This becomes especially important if those data centers provide large-scale computing power to customers in China or Macau.

The rule is designed to stop restricted entities from bypassing export controls by using cloud services or foreign data centers instead of purchasing the hardware directly.

Foreign data centers must perform tougher customer checks

The Commerce Department is also requiring stronger “Know Your Customer” checks from foreign data centers, chip exporters, semiconductor foundries, and related companies.

These firms must now look more closely at whether their customers are directly or indirectly owned or controlled by entities based in China, Macau, or other restricted jurisdictions. This includes ownership hidden through subsidiaries, corporate restructuring, spinoffs, or other indirect arrangements.

The goal is to prevent companies from disguising their links to restricted end users while still gaining access to advanced AI computing resources.

Data centers must report changes in end use

Another important part of the rule involves changes in how chips are used after they are approved for export.

If a foreign data center initially receives advanced AI chips for an approved purpose but later shifts toward providing large-scale compute services to Chinese or Macau-based companies, it must seek new authorization from the Bureau of Industry and Security.

This provision gives U.S. regulators more control after the chips leave the country, reducing the chance that approved exports are later repurposed in ways that violate national security restrictions.

Foreign fabs face tighter controls on China-designed chips

The new rules also tighten restrictions on semiconductor manufacturers and chip packagers.

Previously, a foreign semiconductor fab producing chips for a Chinese chip designer could sometimes submit technical documentation to show that the product was intended for ordinary consumer electronics and should not be treated as a controlled advanced chip.

That pathway is now being narrowed. If the chip designer is based in China or another country of concern, the product will be treated more strictly under U.S. export controls. In practice, this means foreign fabs will have a harder time manufacturing and shipping certain chips for Chinese customers without going through a U.S. licensing process.

This change is intended to stop Chinese chip designers from using overseas manufacturers to work around U.S. restrictions.

New anti-diversion controls affect key transit countries

The Commerce Department is also adding destination-based license requirements for exports to several countries that officials believe could be used as diversion points for restricted semiconductors.

The countries named in the new controls include Indonesia, Malaysia, Singapore, Thailand, and Brazil.

These locations are important because advanced AI chips and servers could potentially be shipped there before being redirected to restricted end users. By increasing scrutiny on these destinations, U.S. officials aim to reduce the risk of indirect shipments to China.

The “de minimis” loophole is narrowed for AI servers

The new rule also limits the use of the “de minimis” exemption for advanced AI computing products.

Under normal U.S. export rules, a foreign-made product that contains only a small amount of U.S. technology may sometimes avoid strict U.S. controls. But for advanced AI servers containing restricted U.S. chips, that exemption is being curtailed.

That means a foreign manufacturer cannot place restricted U.S. AI chips inside a server and then claim the final product is mostly foreign-made. If the server is intended for a restricted data center or a customer providing major computing power to China or Macau, U.S. export restrictions can still apply.

In other words, even a small amount of controlled U.S. AI technology inside a foreign-built system may trigger licensing requirements if the final destination or end user is prohibited.

TikTok-related compute may be treated differently

One notable detail is that ByteDance’s search algorithm-related computing footprint for TikTok is reportedly not covered by these particular restrictions.

That carveout could become a point of discussion as regulators continue to refine how export controls apply to cloud computing, algorithm training, and AI infrastructure used by major technology platforms.

Reported NVIDIA chip flows to China draw scrutiny

The rule changes arrive amid growing attention on how restricted NVIDIA hardware may have reached China despite U.S. export controls.

According to institutional supply-chain research cited in the original reporting, more than 1.5 gigawatts of restricted NVIDIA chips may have ended up in China in recent years.

Separate reporting has claimed that Singapore-based Megaspeed, where Alice Huang previously served as an executive, arranged shipments involving nearly $12 billion worth of top-tier NVIDIA chips to Chinese companies between March 2024 and June 2025. It remains unclear how many of those chips were actually delivered.

Other reporting has suggested that at least $7.5 billion in NVIDIA hardware, equal to roughly 240,000 GPUs, moved through Megaspeed and a Malaysian subsidiary for Chinese clients.

These claims have intensified pressure on NVIDIA, its leadership, and its board to explain whether proper compliance systems were in place.

Scott Bessent demands answers from NVIDIA

Treasury Secretary Scott Bessent has publicly criticized NVIDIA and CEO Jensen Huang over the controversy, calling for answers from the company’s board and outside counsel.

Bessent questioned whether NVIDIA’s internal culture allowed questionable sales practices to continue in order to boost revenue and earnings. He also referenced allegations involving Alice Huang and Jensen Huang, while stressing that the central issue is whether restricted sales were improperly encouraged or enabled.

He said NVIDIA’s board and legal advisers should provide answers within ten days.

The timing is especially important because NVIDIA is expected to report its latest quarterly earnings soon. The company remains one of the most important players in the global AI boom, and any regulatory pressure over China-related chip sales could have major implications for investors, data center customers, and the broader semiconductor industry.

Why the new AI chip rules matter

The latest U.S. export controls show that Washington is expanding its focus beyond physical chip shipments. Regulators are now targeting the broader AI infrastructure ecosystem, including cloud computing, foreign data centers, semiconductor manufacturing, server assembly, and remote access to large GPU clusters.

For NVIDIA and other AI chip companies, this creates a more complicated compliance environment. Selling chips to a foreign customer is no longer just about the immediate buyer. Companies may now need to understand who ultimately uses the chips, where the servers are located, who rents the computing power, and whether any restricted entity benefits from that access.

For China, the rules could make it harder to obtain the massive AI compute capacity needed to train advanced models at scale. For global data center operators, the message is clear: hosting advanced U.S. AI chips comes with tighter obligations, especially if customers have links to restricted regions.

The broader impact on the AI chip market

These restrictions could reshape the global AI hardware supply chain. Countries that have become key data center hubs may face more scrutiny. Foreign manufacturers using U.S. AI components may need stronger compliance systems. Cloud providers may have to conduct deeper checks on customers seeking large-scale GPU access.

At the same time, demand for advanced AI chips remains extremely high. NVIDIA’s GPUs continue to dominate AI training and inference workloads, making them strategically valuable not only to technology companies but also to governments concerned about national security and economic competitiveness.

The U.S. government’s latest move signals that export controls are evolving alongside the AI industry itself. As access shifts from buying chips to renting computing power, regulators are adjusting the rules to close gaps that could allow restricted users to benefit from advanced American technology indirectly.