Servers and AI chip controls in a data center used for China-linked compute routes

US tightens AI chip controls, targeting overseas access and third-country routes to China

US AI chip controls tighten as Washington targets overseas access, Singapore and Malaysia shipments, and China-linked compute routes.

In short

The US on Monday tightened AI chip controls for China, extending rules to overseas data-center access and some third-country shipments. The move is meant to curb smuggling and remote access to advanced chips, not just direct exports.

  • The US expanded AI chip controls beyond direct exports to China.
  • New rules target overseas data-center access and shipments to five countries, including Singapore and Malaysia.
  • Washington says it wants to stop smuggling and shell-company routing of advanced chips.
  • The enforcement challenge is significant because BIS is small and underfunded relative to the scale of the trade.
  • The policy marks a shift from blocking chip purchases to controlling how Chinese firms use chips abroad.

The United States on Monday tightened its controls on advanced AI chip access for China, extending scrutiny beyond direct exports to overseas data-center access and shipments routed through third countries. The new rules are aimed at limiting China’s ability to obtain large-scale computing power through foreign-based infrastructure, a shift that matters because access to compute is now a central front in the US-China technology contest.

What the US changed on Monday

According to WIRED, the new policy adds licensing and compliance requirements for overseas data centers that serve China, while also imposing stricter shipment rules for advanced chips sent to five countries, including Singapore and Malaysia. The changes reflect Washington’s view that controls on AI hardware cannot stop at the border if Chinese firms can still tap restricted chips through foreign locations.

The practical effect is broader than a simple export ban. Under the new approach, companies moving advanced chips, or providing access to clusters built from them, face more scrutiny even when the hardware is not physically entering China. That matters because the value of these chips lies not only in ownership, but in the ability to rent, route or remotely access their processing power at scale.

“crack down on the smuggling of advanced US SI chips”

Why it matters

This is the latest sign that Washington is moving from blocking sales into China to policing how Chinese-linked users obtain compute abroad. That shift could reshape the market for data-center operators, cloud brokers and chip suppliers in Asia, especially in countries that have become transit hubs or storage points for sensitive hardware.

For readers, the policy matters because it touches the hidden infrastructure behind AI systems. Training frontier models depends on huge pools of chips, servers and networking gear. If the US can make access to those resources harder, it may slow the buildout of Chinese AI systems, increase compliance costs for overseas operators and intensify pressure on intermediaries that move hardware across borders.

It also matters because the enforcement burden falls on a government unit that WIRED described as small and underfunded: the Bureau of Industry and Security. The agency’s 2026 budget was $235 million, underscoring the gap between the scale of the control regime and the resources available to police it.

What changes for companies?

Companies that host, route or resell access to advanced chips are now facing a more complicated regulatory environment. Even firms outside China may need licenses and extra compliance steps if their services are judged to support Chinese users or entities tied to China.

That creates at least four immediate pressures:

  • more paperwork for overseas data-center operators;
  • greater due diligence on customers and end users;
  • more risk for firms in countries used as routing hubs;
  • more uncertainty over whether previously legal arrangements will still pass scrutiny.

How did the US get here?

The current policy did not emerge overnight. The Biden administration began chip-export controls in the fall of 2022, setting out to curb China’s access to advanced AI chips. Since then, the US strategy has evolved from trying to prevent sales into China to trying to regulate where and how Chinese firms can use chips abroad.

That change reflects a hard reality: hardware restrictions alone are easier to circumvent than policymakers first hoped. If a company cannot buy a chip directly, it may still find a way to rent access to one through an overseas data center, a third-country intermediary or a shell arrangement. Washington’s newest rules are designed to close those routes.

Why Singapore and Malaysia are in the frame

WIRED reported that the stricter shipment rules now cover advanced chips sent to five countries, including Singapore and Malaysia, because US officials suspect smuggling routed through those markets. Those countries have become important nodes in the global hardware supply chain, and that makes them vulnerable to being used as transit points, storage locations or remote-access bases for restricted compute.

The concern is not simply where chips are shipped, but what happens after they arrive. If hardware can be moved, concealed, rerouted or accessed remotely after import, then the location of the initial sale becomes only one step in a larger chain of compliance risks.

What is the Megaspeed connection?

One of the most closely watched threads in WIRED’s reporting involves Megaspeed, a Singapore-based data-center firm that the outlet says is suspected of chip smuggling and remote-access services tied to China. WIRED also reported that Alice Huang reportedly worked at Megaspeed, and framed the policy shift against the backdrop of what it described as the Jensen Huang affair scandal.

According to WIRED, Jensen Huang resigned as Nvidia CEO last week after admitting an affair with Alice Huang. The outlet’s reporting says the episode has become part of the wider debate around how advanced chips move through overseas companies and who ultimately benefits from them. The fact sheet does not establish a causal link between the personal scandal and the policy move, but it does place them side by side in the same moment of heightened scrutiny.

WIRED also reported a separate detail that sharpened concern around Megaspeed: Bloomberg said about 50,000 of 136,000 AI chips Megaspeed imported into Malaysia and Indonesia in 2024-2025 were unaccounted for in inventory checks. Nvidia later said it found at least some of the missing chips. That episode illustrates the difficulty regulators face when hardware is spread across multiple jurisdictions and handled by layers of vendors.

By the numbers

The tightening of controls is best understood through the scale of the systems involved. Here are the core figures cited in the reporting:

Figure What it refers to Context
$235 million Bureau of Industry and Security budget in 2026 Shows the scale of enforcement resources
50,000 Megaspeed-imported AI chips not accounted for in Malaysia and Indonesia Bloomberg figure cited by WIRED for 2024-2025
136,000 Total AI chips Megaspeed imported into Malaysia and Indonesia Baseline used in the inventory-check discrepancy
1 million H100-equivalent compute China could have smuggled by end-2026 Institute for Progress estimate cited by WIRED
$130 billion Combined market value lost by affected US suppliers Fed New York estimate tied to earlier export restrictions

What officials think they are stopping

The White House said the goal is to cut off smuggling networks and make sure the most advanced chips go first to US companies. In one statement attributed to a White House spokesperson, the administration said it wants to “help ensure that America’s most advanced chips go first to US SI innovators, not Chinese shell companies.”

That framing suggests two separate targets. The first is direct smuggling of hardware. The second is the use of shell companies and foreign intermediaries to mask the true end user. By naming both problems, the administration is acknowledging that modern export control enforcement is as much about corporate structures as it is about physical shipments.

Background: why AI chips are so hard to police

Advanced AI chips are not ordinary processors. They are the hardware that powers large-scale model training, which requires enormous amounts of compute. The most sensitive chips, including Nvidia-class devices, can be clustered together to create systems capable of training or running frontier AI models.

That is why Chinese firms have been able to make use of overseas data centers and third-country routing. If access to compute is what matters, then the geography of the server rack can be as important as the address on the invoice. The US response has therefore shifted toward controlling not only who can buy chips, but who can use them and where.

There is also a technological cat-and-mouse element. WIRED reported that Huawei has developed chip-linking systems to build larger compute clusters, showing how suppliers under restriction can still try to stitch together sufficient capacity. And DeepSeek has demonstrated that more efficient training can reduce compute requirements, which means controls on chips do not necessarily end the race; they may simply change the route through it.

Why remote access is harder to govern

Analysts cited by WIRED say remote access to chips overseas is costlier and harder to manage than owning chips directly, but it still offers a route around restrictions. That creates a paradox for policymakers: the more complex the workaround, the more compliance infrastructure is needed to detect it, and the more likely it is that regulators will be one step behind.

Remote access also blurs jurisdictional lines. A server in one country can be used by a customer in another, through a company incorporated somewhere else, with financing or maintenance handled elsewhere again. Each layer adds another place where controls can be challenged, delayed or obscured.

How big is the enforcement challenge?

The enforcement challenge is significant because the US is not trying to control a single transaction. It is trying to monitor a sprawling ecosystem of chip sales, cloud access, logistics, data centers and corporate ownership structures.

WIRED said the Bureau of Industry and Security leads enforcement, but the unit is relatively small. That matters because the agency must track not only exports from the US, but also foreign resales, possible smuggling routes, inventory discrepancies and service arrangements that can be hard to verify in real time.

The pressure on regulators is matched by the scale of the potential prize. The Institute for Progress estimated China could have smuggled computing power equal to about 1 million Nvidia H100 chips by the end of 2026. Even as a projection, that figure helps explain why Washington is treating remote access and third-country routes as strategic vulnerabilities rather than niche trade issues.

Market consequences: who stands to feel the squeeze?

US chip suppliers, cloud providers and overseas data-center operators could all feel the effects of the new rules. Earlier export restrictions, according to Fed New York researchers cited by WIRED, erased $130 billion in combined market value for affected US suppliers. That prior hit suggests the market has already been forced to price in a more constrained China business.

But the latest rules may affect a different set of players. Firms in Singapore, Malaysia and other transit countries may face more audits, stricter customer screening and delays in moving hardware. Chinese firms may see access to offshore compute become less predictable and more expensive. And US suppliers could face additional administrative burdens as the government expects them to police where their chips end up and how they are used.

Who is most exposed?

The most exposed players are likely to be companies that sit between the chipmaker and the end user. That includes data-center operators, distributors, hosting firms and intermediaries that arrange capacity deals across borders. It also includes any business whose customer base, investor links or service contracts raise questions about China exposure.

For Nvidia and similar suppliers, the issue is not just whether a chip is sold once. It is whether the chip can be traced, tracked and controlled after sale. The tighter the rules become, the more important documentation, audits and inventory checks will be.

What happens next

The immediate question is how aggressively the new licensing and shipment rules will be enforced. WIRED’s reporting leaves open how many overseas data centers will need licenses, how strictly those licenses will be granted and what evidence will be required to prove a business is not serving restricted Chinese demand.

Another open question is whether the five-country shipment rule is a fixed list or a first step. The fact sheet says the White House may expand controls further to other countries or service models, but it does not say when or how. That makes the next few months important for testing whether the policy is a targeted adjustment or the start of a broader clampdown.

Three issues will be worth watching:

  1. Whether the government names additional jurisdictions beyond Singapore and Malaysia;
  2. Whether overseas data centers serving Chinese customers face denials rather than just extra paperwork;
  3. Whether investigators produce public evidence linking specific shipments or service providers to smuggling networks.

What the policy shift really means

The deeper significance of Monday’s move is that the US now appears to be treating compute as a controllable asset, not just a product. That is a harder standard to enforce, because compute can be rented, pooled, clustered and accessed remotely across borders. But it is also the only standard that matches the way AI infrastructure now works.

In that sense, the policy is less about a single shipment of chips than about a new phase in US-China tech competition. Washington is no longer just trying to keep advanced hardware out of China. It is trying to make sure that, wherever that hardware sits in the world, China cannot easily tap it without running into American rules.

That is a much wider reach for export controls, and a much bigger test of whether US enforcement can keep pace with global AI supply chains.

Timeline

Date Event
Fall 2022 Biden administration launched initial AI chip export controls
2024-2025 Megaspeed imported large volumes of AI chips into Malaysia and Indonesia
2026 Bureau of Industry and Security budget listed at $235 million
Monday Trump administration announced new licensing and shipment restrictions
Last week Jensen Huang resigned as Nvidia CEO after admitting affair with Alice Huang
End-2026 Institute for Progress projected China could hold smuggled compute equal to 1 million H100 chips

Whether these new rules slow China’s access to compute or merely raise the cost of getting around them will become clearer only as the licensing system is applied. For now, the US has made one thing plain: it is prepared to extend export control logic well beyond the border, and deep into the overseas networks that make AI hardware usable.

Frequently asked questions

What did the US change in its AI chip rules on Monday?

The US tightened AI chip controls by adding licensing and compliance requirements for overseas data centers serving China and by imposing stricter shipment rules on advanced chips sent to five countries, including Singapore and Malaysia. The move is aimed at cutting off routes used for smuggling and remote access.

Why are Singapore and Malaysia being singled out?

Singapore and Malaysia are being singled out because Washington suspects smuggling routes run through them. WIRED reported that the new shipment rules cover advanced chips sent to five countries, including those two, to reduce the risk that hardware is rerouted or accessed for China-linked use.

How is this different from earlier export controls?

The difference is that the US is now focusing on how chips are used abroad, not only on whether they are sold into China directly. Earlier controls began in fall 2022, but the new approach targets foreign-based access, third-country routing and remote compute services.

Why does the enforcement challenge matter?

The enforcement challenge matters because the Bureau of Industry and Security is relatively small and underfunded for such a broad mission. WIRED said its 2026 budget was $235 million, which underscores how hard it may be to police global chip flows, inventory checks and remote access arrangements.

What evidence suggests China may still be getting compute indirectly?

WIRED pointed to several signs, including suspected smuggling routes through Singapore and Malaysia, the Megaspeed inventory discrepancy, and an Institute for Progress estimate that China could have smuggled compute equal to about 1 million Nvidia H100 chips by the end of 2026.

Share this 🚀