The Loophole That Let China Rent America's Most Powerful AI Chips Is Closing, and It Took a Spy Scandal to Do It

The rule: The Commerce Department has issued new export restrictions aimed at stopping Chinese entities from remotely accessing advanced US AI chips housed in data centres outside China. These arrangements have long sat in a gray zone of American export law. The trigger: The move follows an alleged espionage scandal linked to Nvidia. Media reports citing unnamed US officials have tied the case to data centre operators in Southeast Asia that buy high-end Nvidia chips and rent their computing power. The fallout: Jensen Huang, Nvidia's co-founder and its only chief executive since 1993, has stepped down amid the scandal. The departure has shaken the world's dominant AI chipmaker.
For three years, the United States built a wall around its most advanced artificial intelligence chips. It was a wall with a door in it.
Washington barred Nvidia and its rivals from shipping their most powerful processors to China. It tightened the rules again and again. It added Chinese companies to trade blacklists, restricted chipmaking equipment, and pressured allies to follow suit. Yet throughout that campaign, a Chinese company did not need to own an American AI chip to use one. It only needed to rent one.
That door is now being pushed shut. The Trump administration has introduced new export restrictions designed to block China from remotely accessing advanced US AI chips through data centres located outside its borders. The Commerce Department's move targets arrangements that let Chinese firms tap the computing power of servers in Southeast Asia and elsewhere. That power is used to train and run AI models that American policy was meant to keep out of reach.
The timing is no accident. The rule arrives amid an alleged espionage scandal linked to Nvidia, the company whose chips power most of the world's AI. According to media reports citing unnamed US officials, the case involves the network of overseas data centre operators that buy high-end Nvidia hardware and rent out its computing capacity. The scandal has already claimed the most prominent figure in the global technology industry. Jensen Huang, who co-founded Nvidia in 1993 and led it to become one of the most valuable companies on Earth, has stepped down as chief executive.
Beijing has dismissed the spying allegations as "entirely groundless" and "speculative." Officials in Washington, meanwhile, are confronting an uncomfortable question. Their export control regime was designed to slow China's progress in AI. How much of it was working at all?
The numbers suggest the answer is not reassuring. Research group Epoch AI estimates a median of 660,000 Nvidia H100-equivalent chips were smuggled into China through 2025. Its possible upper range is about 1.6 million. Those figures cover smuggling alone. They do not account for the computing power China accessed legally, or nearly legally, through rented servers abroad.
The new rule is the most direct attempt yet to close that gap. Whether it comes too late, and whether it can be enforced at all, will shape the next chapter of the technology rivalry between the world's two largest economies.
A Rule Aimed at the Cloud
American export controls were written for a world of physical goods. They govern what can be shipped, to whom, and under what conditions. A chip crossing a border is something customs officials can inspect, licensing officers can approve or deny, and prosecutors can trace.
Cloud computing broke that model. When a company in Shenzhen or Beijing rents processing time on a server in Singapore, Kuala Lumpur, or Jakarta, no chip crosses any border. The hardware stays in the data centre. Only data moves: training sets go in, trained models come out. Under the traditional reading of US export law, that kind of remote access largely fell outside the rules that governed the physical export of the chips themselves.
The gap was not secret. Security analysts, congressional committees, and former export control officials warned about it for years. Lawmakers introduced legislation aimed at bringing remote access under export control authority. The Commerce Department's Bureau of Industry and Security, which administers the controls, signaled growing concern. In May 2025, the bureau issued guidance warning that providing computing services to train AI models for Chinese entities could raise red flags under existing rules. Guidance, however, is not the same as a binding prohibition. The core loophole remained open.
The new restrictions aim to change that. The rule is designed to stop Chinese entities from reaching advanced US AI chips through overseas data centres. It extends the logic of export controls from the hardware itself to the computing power that hardware provides. In practical terms, it puts data centre operators and cloud providers outside China on notice. Renting top-tier American AI capacity to Chinese customers is now an export control problem, not just a business decision.
For the Commerce Department, the move also serves another purpose. The agency has faced increased scrutiny over its handling of semiconductor export curbs, particularly as details of the Nvidia-linked case have emerged. Critics in Congress and the national security community have argued that the department moved too slowly and too inconsistently. They say the gap between policy on paper and reality on the ground grew wide enough for a significant share of America's most advanced AI hardware to end up serving Chinese interests.
The rule is, in part, a response to that criticism. It is also an admission that the old framework was not built for how AI is actually developed.
The Scandal That Forced Washington's Hand
Export control debates usually unfold in technical language, inside agency rulemakings and committee hearings that attract little public attention. The Nvidia-linked espionage scandal changed that.
The full details of the case remain unclear, and much of what is known comes from media reports citing unnamed US officials. Those reports describe an alleged spy connected to a company that is one of several data centre operators in Southeast Asia. These operators purchase high-end Nvidia chips and then rent their computing power to customers. The allegations go to the heart of the remote access problem. They suggest that the overseas infrastructure built around American chips may have served not only commercial clients but also intelligence interests.
Beijing has denied the allegations. No court has ruled on them, and many questions remain unanswered. These include who was involved, what was accessed, how long the alleged activity lasted, and what Nvidia knew. Those questions will likely be pursued in congressional inquiries, regulatory reviews, and possibly legal proceedings in the months ahead.
Yet the political impact has been immediate. For years, the debate over AI chip exports pitted two arguments against each other. National security hawks warned that China was getting around the controls. Industry voices warned that overly aggressive rules would hand market share to Chinese competitors and weaken American companies. The scandal has shifted the balance of that argument sharply toward the hawks.
It has also changed how the problem is framed. Smuggling and remote access were once treated mainly as questions of commercial leakage, in which advanced chips ended up powering Chinese AI labs that were not supposed to have them. An espionage dimension raises the stakes. It suggests that the global network of AI data centres may be vulnerable not only to evasion but also to exploitation by state actors.
That shift in framing helps explain why the administration moved now. A loophole that had been tolerated for years became politically untenable almost overnight.
Huang's Exit and Nvidia's Reckoning
Few corporate departures carry the weight of Jensen Huang's.
Huang co-founded Nvidia in 1993 and led it for more than three decades. He took it from a maker of graphics cards for video gamers to the backbone of the global AI industry. Under his leadership, Nvidia's chips became the default hardware for training large AI models. Its CUDA software platform created a moat that competitors spent years trying to cross. In October 2025, the company became the first in history to reach a market value of $5 trillion.
Huang himself became the face of the AI boom, a fixture in his signature leather jacket at product launches and a frequent visitor to world capitals. He was also a central figure in the debate over chip exports. He argued repeatedly that China represented an enormous market for American technology. He warned that cutting Nvidia off would only accelerate the rise of Chinese rivals such as Huawei. He said that export controls had, in some respects, backfired by pushing Chinese firms to build their own alternatives.
That position made Huang influential in Washington. It also made him a target for critics who believed Nvidia was too eager to keep selling into China, even as the US government tried to limit its access to advanced computing.
His departure amid the espionage scandal marks a turning point for the company. The circumstances surrounding his exit are still being examined, and the public record remains incomplete. But the symbolism is hard to miss. The executive who most forcefully argued for keeping China as a customer has left as Washington moves to tighten the rules further.
For Nvidia, the challenges ahead are substantial. Its new leadership must reassure investors that the company's growth story remains intact. It must rebuild trust with policymakers who now view the firm's overseas sales channels with suspicion. And it must address how its chips were bought, deployed, and rented through a global network of distributors, server makers, and data centre operators. The company has historically argued that it does not control how customers use its products once they are sold. That argument is likely to face much tougher scrutiny.
Nvidia also faces pressure from Beijing. In 2025, Chinese regulators questioned the company over alleged security risks in its China-specific chips, which Nvidia denied. Chinese authorities also opened an antitrust probe tied to its earlier acquisition of Mellanox. The company now finds itself squeezed between two governments, each suspicious of how its technology is being used.
Southeast Asia's Compute Boom
The new rule places Southeast Asia at the center of the story, and for good reason.
Over the past several years, the region has become one of the fastest-growing data centre markets in the world. Malaysia, particularly the southern state of Johor, has attracted billions of dollars in investment from global technology companies. Johor sits just across the causeway from Singapore, where land and power constraints have pushed data centre development outward. Thailand, Indonesia, and Vietnam have also courted data centre investment as part of broader strategies to build digital economies.
Much of that growth is legitimate and strategically valuable. Global cloud providers have built facilities in the region to serve local and international customers. Governments have welcomed the investment as a source of jobs, infrastructure, and technological prestige.
But the boom also created opportunities for evasion. A data centre in Southeast Asia could legally acquire advanced American AI chips under rules that applied mainly to direct shipments to China. Once those chips were installed, the operator could rent computing capacity to a wide range of customers. Some of those customers were Chinese companies that could not buy the same hardware directly.
The arrangement was attractive for everyone involved. Chinese firms gained access to cutting-edge computing without needing to smuggle physical chips. Data centre operators gained high-paying customers. And the chips stayed outside China, technically consistent with the letter of the export controls.
Regional governments have already begun to respond to the scrutiny. In early 2025, Singaporean authorities charged three men with fraud in a case involving servers that may have contained Nvidia chips and were shipped to Malaysia. In July 2025, Malaysia introduced a permit requirement for exports and transshipments of high-performance US-origin AI chips. The move signaled that the country was aware of its role as a potential conduit.
The new US rule raises the pressure further. Data centre operators across the region now face a direct question. Are they willing to risk access to American technology in order to keep serving Chinese customers? For many, the answer will be no. For some, the economics of the Chinese market may still prove tempting. That is exactly the kind of behavior enforcement officials will be watching for.
660,000 Chips: Measuring the Leak
Remote access is only one part of the leakage problem. The other is smuggling, and the scale of that problem has grown clearer.
Epoch AI's estimate that a median of 660,000 Nvidia H100-equivalent chips were smuggled into China through 2025 is striking on its own. The upper range of roughly 1.6 million is even more so. The H100 was the flagship chip of the generative AI boom. It powered the training of many of the world's most advanced AI models, and it was one of the primary targets of US export controls.
To put those numbers in context, hundreds of thousands of H100-class chips represent a level of computing power on the scale of the largest AI clusters operated by leading American technology companies. Even at the lower end of the estimate, the smuggled supply would amount to a significant national AI resource.
Smuggling has taken many forms. Chips have reportedly moved through shell companies, false end-user declarations, and transshipment hubs. They have traveled hidden inside larger server shipments and through intermediaries in third countries. The US Justice Department has brought cases against individuals accused of illegally exporting AI chips to China. The high value and small size of the hardware make it difficult to track once it leaves authorized channels.
These estimates also complicate a central claim of US policy. Supporters of export controls have argued that the restrictions were slowing China's AI development by limiting access to advanced computing. If hundreds of thousands of top-tier chips reached China anyway, and if Chinese firms could rent even more capacity abroad, the controls may have been far less effective than their architects intended.
The success of Chinese AI developers reinforced that concern. In January 2025, the Chinese startup DeepSeek released a reasoning model that rivaled leading American systems at a fraction of the reported cost. The announcement shocked global markets. Nvidia lost roughly $590 billion in market value in a single trading day, the largest one-day loss for any company in US history at the time. DeepSeek's success raised questions about whether China needed the most advanced chips at all, or whether it was obtaining them in greater numbers than Washington realized.
The truth is likely a mix of both. Chinese engineers have become highly skilled at extracting performance from limited hardware. But access to large volumes of advanced American chips, whether smuggled, stockpiled, or rented, has clearly helped.
Beijing Pushes Back
China's response to the latest developments has followed a familiar pattern.
Chinese officials have rejected the spying allegations as "entirely groundless" and "speculative." They have noted that the claims have come largely from unnamed US officials quoted in media reports. Beijing has long described American export controls as an abuse of national security arguments, designed to suppress China's technological development and protect US commercial dominance.
That framing resonates with many in China and with some observers abroad. Chinese officials argue that the United States has repeatedly broadened the definition of national security to justify trade restrictions. They say the latest rule will harm global supply chains, raise costs for businesses worldwide, and undermine the openness that once defined the technology industry.
Beijing has also been building alternatives. Huawei has developed its Ascend line of AI processors, and Chinese authorities have encouraged domestic companies to use local chips where possible. In 2025, Chinese regulators reportedly discouraged firms from buying certain Nvidia products designed for the Chinese market. They cited security concerns and the need to support domestic suppliers.
Those efforts have produced real progress, though significant gaps remain. Chinese chipmakers still face restrictions on access to the most advanced manufacturing equipment, much of which comes from the United States, the Netherlands, and Japan. Producing cutting-edge AI chips at scale without that equipment is difficult. Software ecosystems built around Nvidia's platform are also hard to replace.
The new US rule will likely accelerate China's push for self-reliance. That is the central dilemma of export controls. Every restriction that limits China's access to American technology also strengthens Beijing's incentive to build its own. Washington is betting that it can stay ahead long enough for the lead to matter. Beijing is betting that it can close the gap faster than Washington expects.
From Biden to Trump: A Policy That Kept Shifting
The remote access loophole did not survive by accident. It persisted through years of shifting and sometimes contradictory policy under two administrations.
The modern era of AI chip controls began in October 2022. That month, the Biden administration imposed sweeping restrictions on the export of advanced chips and chipmaking equipment to China. Nvidia responded by designing modified chips, such as the A800 and H800, that fell just below the performance thresholds. In October 2023, the administration tightened the rules to close those gaps. Nvidia again adapted, creating the H20, a chip designed specifically to comply with the new limits.
In January 2025, in its final days, the Biden administration issued the AI Diffusion Rule. It was an ambitious framework that would have divided the world into tiers, setting limits on how much advanced computing power countries could import. The rule was designed in part to address the problem of chips flowing to China through third countries.
The Trump administration rescinded the AI Diffusion Rule in May 2025, before it took effect. Officials argued that it was overly complex, would burden American companies, and would damage relationships with allies. They promised a simpler replacement.
What followed was a series of reversals. In April 2025, the administration required licenses for exports of Nvidia's H20 to China, effectively blocking sales. Nvidia took a multibillion-dollar charge as a result. By July, the administration said it would grant licenses for the chip. In August, reports emerged of an unusual arrangement under which Nvidia and AMD would pay the US government a share of revenue from certain China sales. Late in 2025, President Trump said Nvidia would be allowed to sell its more powerful H200 chips to approved customers in China, with the US government taking a cut.
Supporters described that approach as pragmatic. In their view, it kept American companies competitive, generated revenue for the government, and maintained Chinese dependence on US technology. Critics called it incoherent. They argued that treating export controls as a bargaining chip undermined their national security purpose and sent mixed signals to allies, companies, and Beijing alike.
The remote access loophole sat in the background of all these debates. While Washington argued over which chips could be sold directly to China, the question of who could rent advanced computing abroad received far less decisive action.
The new rule reflects a shift in emphasis. Whatever the administration's position on direct sales, it is now signaling that access through overseas data centres must be controlled. Critics will argue that the shift is overdue. Supporters will argue that it shows the administration is willing to act when evidence of abuse emerges. Both arguments contain some truth.
Congress Moves to Tighten the Screws
Capitol Hill is not waiting for the Commerce Department to finish the job.
Lawmakers are moving to cut AI chip sales to China further in the wake of the scandal and Huang's departure. Concern over chip exports has long been one of the few areas of genuine bipartisan agreement in Washington. Republicans and Democrats alike have criticized the executive branch, under both parties, for moving too slowly and leaving gaps in the rules.
Several legislative ideas have circulated for years and are now gaining momentum. One is location verification. In 2025, lawmakers introduced the Chip Security Act, which would require advanced AI chips to include mechanisms that verify where they are physically located. Supporters argue that such technology would make smuggling and unauthorized diversion far easier to detect. Opponents in the industry have raised concerns about technical feasibility, cost, and the risk of creating security vulnerabilities.
Other proposals would bring remote access under clear statutory authority rather than relying on regulatory interpretation. Writing the rules into law would make them harder for future administrations to loosen and would give enforcement agencies stronger tools.
Lawmakers have also pushed proposals that would require chipmakers to prioritize American customers before selling to countries of concern. The argument behind these measures is straightforward. If computing power is a strategic resource, the United States should make sure its own companies and researchers have first access to it.
The scandal has given these efforts fresh energy. Members of Congress who once faced intense industry lobbying against tighter controls now have a powerful counterargument. Leakage is no longer a theoretical risk. It is a documented problem with a potential espionage dimension.
The industry will still push back. But the political climate has shifted, and lawmakers sense an opening to lock in restrictions that have long been debated.
The Industry's Counterargument
Not everyone in the technology sector welcomes the new restrictions, and their concerns deserve a hearing.
The first concern is commercial. China has historically been one of the largest markets for American semiconductors. Restrictions on direct sales have already cost US chipmakers billions of dollars in revenue. Extending controls to remote access could reduce demand further, particularly among overseas data centre operators that counted Chinese firms among their customers.
The second concern is competitive. Industry leaders, Huang chief among them, have argued that cutting China off from American chips does not stop Chinese AI development. Instead, it hands the Chinese market to domestic suppliers such as Huawei. Over time, that could help Chinese chipmakers build scale, improve their products, and eventually compete globally. In this view, export controls risk creating the very competitor they are meant to contain.
The third concern is compliance. Cloud providers and data centre operators will need to know their customers in far greater detail than before. Determining whether a customer is ultimately controlled from China can be difficult when ownership structures span multiple jurisdictions, holding companies, and intermediaries. Compliance costs will rise. Some legitimate customers may be turned away out of caution.
The fourth concern is diplomatic. Many of the data centres affected by the rule are in countries that are not adversaries of the United States. They are partners and, in some cases, close allies. Aggressive extraterritorial rules can strain those relationships. Governments in Southeast Asia have worked hard to attract data centre investment. They may resent rules written in Washington that limit how facilities on their territory can operate.
These arguments have shaped US policy for years and will continue to do so. But the balance has changed. The combination of large-scale smuggling estimates, documented remote access arrangements, and an alleged espionage scandal has made the commercial case harder to sustain politically. Industry arguments that once carried significant weight in Washington now face a skeptical audience.
The Enforcement Challenge
Writing a rule is one thing. Enforcing it is another.
The Bureau of Industry and Security is a small agency relative to the scale of its mission. It is responsible for overseeing exports of a vast range of sensitive technologies, from chips to aerospace components. Former officials and outside analysts have long warned that the bureau lacks the staff, funding, and technical resources to police the global semiconductor trade effectively.
Remote access is especially hard to monitor. A physical chip can, at least in theory, be tracked through shipping records, customs inspections, and serial numbers. Computing power rented through the cloud leaves a different kind of trail. Enforcement depends heavily on the cooperation of data centre operators and cloud providers. It also relies on their willingness to conduct thorough due diligence on customers and to report suspicious activity.
Determined actors have many ways to evade detection. They can use front companies registered in third countries. They can route access through layers of intermediaries. They can split workloads across multiple providers to avoid drawing attention. They can rely on individuals rather than corporate accounts. Each of these tactics makes it harder to identify when a Chinese entity is the true end user.
Technology may offer partial solutions. Location verification, if it proves workable, could help confirm where chips are physically operating. Cloud providers could build more detailed monitoring into their platforms, tracking unusual usage patterns associated with large-scale model training. Governments could share intelligence about suspicious entities more systematically.
But none of these measures will be foolproof. The history of export controls suggests that restrictions raise the cost and difficulty of evasion rather than eliminating it. The real question is whether the new rule raises those costs enough to meaningfully slow China's access to advanced computing.
Allies, Partners, and the Data Centre Map
The new restrictions will reshape the global data centre map in ways that extend far beyond China.
Countries hoping to become AI hubs now face a clearer choice. They can align closely with US export controls, gaining access to the most advanced American chips and the investment that follows. Or they can maintain looser policies that may attract Chinese customers but risk losing access to American technology.
For many governments, the choice is straightforward. The United States remains the dominant source of advanced AI chips, and American cloud providers are the largest investors in global data centre infrastructure. Few countries will want to jeopardize that relationship.
Still, the dynamics are complicated. Southeast Asian governments have generally sought to maintain strong economic ties with both the United States and China. They are wary of being forced to choose sides in a technology conflict. Rules that push them toward explicit alignment with Washington may be accepted reluctantly, and enforcement may vary in practice.
The Middle East presents another dimension. The United States has struck major agreements involving AI chips and data centres with Gulf states, including the United Arab Emirates and Saudi Arabia, which aim to become global AI centers. Those arrangements have included commitments around security and preventing diversion to China. The new remote access rule will likely raise expectations for how such commitments are monitored and enforced.
Europe, Japan, South Korea, and Taiwan also have roles to play. These partners host significant semiconductor industries and data centre markets. Coordinated controls among them would be far more effective than unilateral American action. Achieving that coordination has historically been slow, but the scandal may create new urgency.
Markets Weigh the Fallout
Investors are confronting a rare combination of risks: leadership change, regulatory tightening, and reputational damage at the most important company in the AI supply chain.
Nvidia's value rests on its dominance in AI hardware and on expectations of continued rapid growth. The company's leadership transition introduces uncertainty at a sensitive moment. Huang's vision, relationships, and public profile were deeply tied to Nvidia's identity. His successor will need to establish credibility quickly with investors, customers, and governments.
The new export restrictions add another layer of risk. Direct sales to China were already constrained and uncertain. Remote access controls could reduce demand from some overseas data centre operators. Tighter compliance requirements across the supply chain could slow sales cycles and raise costs for customers.
At the same time, demand for AI computing outside China remains enormous. American technology companies continue to spend heavily on data centres. Governments around the world are pursuing national AI strategies, and enterprises are building AI into their operations. For many investors, the China question, while significant, is secondary to that broader demand.
The ripple effects extend beyond Nvidia. Server makers, cloud providers, data centre developers, and other chip designers all have exposure to the rules. Companies with significant operations in Southeast Asia may face particular scrutiny. Those that can demonstrate strong compliance practices may gain an advantage as customers and governments look for trusted partners.
What Comes Next
The coming months will determine whether the new rule marks a genuine turning point or another chapter in a long cycle of loopholes and patches.
The Commerce Department will need to show that it can enforce the restrictions. That will likely mean investigations, penalties, and additions to trade blacklists for companies found to be violating the rules. Visible enforcement actions will send a stronger signal than any rulemaking.
Congress will continue to press for legislation. Measures on remote access, location verification, and limits on chip sales to China are likely to see renewed debate. The scandal gives lawmakers political cover to pursue restrictions that industry has resisted.
Nvidia's new leadership will face pressure to demonstrate cooperation with Washington. That could include stronger due diligence on distributors and customers, support for chip location verification, or changes in how the company approaches the Chinese market. The company's response will shape how policymakers view it for years to come.
Beijing will respond as well. China could take further regulatory action against American firms, accelerate support for domestic chipmakers, or retaliate through restrictions on critical minerals and other inputs. China has used its dominance in rare earths and other materials as leverage before, and it could do so again.
Southeast Asian governments will need to decide how closely to align with the new rules. Their decisions will shape the region's role in the global AI economy.
The espionage investigation itself will continue. As more details emerge, they could reshape the debate again, either confirming the most serious allegations or complicating the narrative. The public record is still incomplete, and much remains to be learned.
The Bottom Line
The Trump administration's move to block China's remote access to advanced US AI chips is overdue, significant, and incomplete.
It is overdue because the loophole was well understood for years. While Washington argued over which chips could be sold directly to China, Chinese firms were renting the same computing power from data centres abroad. Smuggling estimates suggest hundreds of thousands of advanced chips reached China anyway. The controls leaked in ways that should have prompted decisive action long before a spy scandal forced the issue.
It is significant because it extends export controls into the cloud, where much of AI development actually happens. Treating computing power as a controlled resource, rather than focusing only on physical hardware, is a necessary update to a framework built for a different era.
And it is incomplete because rules alone will not stop determined actors. Enforcement capacity, allied coordination, technical safeguards, and consistent policy will matter as much as the regulation itself. The shifting approach to chip exports over the past several years has undermined confidence that Washington can sustain a coherent strategy.
The departure of Jensen Huang, the man who built the company at the center of this fight, captures how much has changed. The AI chip trade is no longer just a business. It is a matter of national security, intelligence, and geopolitical competition. Nvidia's chips made the AI revolution possible. Now the question is who gets to use them, and who decides.
For Washington, the lesson is uncomfortable but clear. A wall with a door in it is not a wall. The United States has finally moved to close the door. Whether it can keep it closed is the test that comes next.


