The United States is closing a route that may have let Chinese-linked companies buy advanced Nvidia chips through overseas units. That turns AI infrastructure into an even sharper test of compliance, supply chains and investor patience.
Washington has moved from warning about AI chip diversion to spelling out who can buy the most powerful processors, even when the buyer is not sitting inside China. The Commerce Department’s Bureau of Industry and Security issued guidance on May 31 clarifying that advanced computing items require export licenses when a customer is headquartered in Country Group D:5, which includes China, or Macau, or when its ultimate parent company is based there, even if the ordering unit sits somewhere else.
That sounds technical. It is not. It goes directly to the question of whether Chinese technology groups can use subsidiaries, cloud partners or data center projects in places such as Malaysia, Thailand and the broader Southeast Asian market to gain access to Nvidia’s Blackwell and Rubin-class systems, along with AMD’s MI350x chips. The shipping address matters less now. The real test is who ultimately controls the buyer.
According to Reuters, the Commerce Department move was aimed at closing a year-old potential loophole that may have allowed the world’s most advanced chips to reach Chinese entities located outside China. One industry source with deep supply-chain knowledge estimated the number of chips that may have moved through the gap in the hundreds of thousands. That figure has not been formally confirmed by the government, but it explains why the guidance arrived with unusual urgency.
Export controls are supposed to stop the sale of restricted chips into China. The weakness was that global AI infrastructure does not respect clean national borders. A Chinese internet company can form or use an overseas subsidiary. A cloud provider can build a cluster in Southeast Asia. A data center can serve customers across borders without every chip physically entering mainland China.
That is why the latest clarification matters. It treats ownership and control as the key risk, not just geography. If a unit outside China is effectively part of a China-headquartered company, the transaction now sits inside the license requirement. For suppliers, distributors and data center operators, that means customer due diligence becomes much harder than checking a delivery destination on a form.
The rule also lands at a difficult moment for Nvidia. The company remains the central supplier for the AI buildout, and Blackwell systems are among the most valuable products in the global technology market. Nvidia has repeatedly argued that American companies should remain competitive in large overseas markets, while Washington has argued that the most advanced chips cannot be allowed to strengthen China’s AI and military capabilities.
Both points can be true. Nvidia’s growth depends on selling into a world where demand for compute is enormous. U.S. policy depends on keeping the highest-end training capacity away from strategic rivals. The more valuable the chips become, the more pressure builds on every weak point in the chain.
Investors now have to price policy risk
For investors, the immediate issue is not whether Nvidia suddenly loses its core AI demand. It will not. U.S. hyperscalers, sovereign AI projects and large cloud customers are still spending aggressively. The issue is that every tightening of export policy changes the revenue map around the edges, and those edges matter when expectations are already stretched.
China has long been one of the most complicated pieces of Nvidia’s story. The company has sold modified chips into the market when allowed, watched Washington restrict some products, and faced pressure from Beijing for domestic firms to use Chinese alternatives. The new guidance adds another layer because it can affect not only direct China sales, but also overseas data center projects tied to Chinese companies.
That is important for hyperscaler customers outside the United States as well. A cloud provider in Malaysia or Singapore may not think of itself as part of the U.S.-China chip fight, but if its ownership, tenants or financing connect back to a Chinese parent, it could find itself under far closer review. Compliance is becoming a commercial feature of AI infrastructure, not a back-office legal issue.
There is also a practical enforcement problem. Chips are small relative to their value, server supply chains are layered, and AI compute can be accessed remotely. A Blackwell system does not have to sit in Shanghai to help a Chinese model developer. That is why regulators are looking beyond physical shipments and into control, access and end use.
The latest step will not end diversion on its own. Companies determined to obtain restricted hardware will keep looking for brokers, shell buyers and friendly jurisdictions. But it raises the cost of doing so and gives U.S. authorities a clearer basis to scrutinize deals that previously sat in a gray area.
The larger implication is that AI infrastructure is becoming geopolitical infrastructure. Data centers, chips, cloud contracts and export licenses now move together. For Nvidia, that means the market will keep rewarding demand, but it will also watch Washington more closely. The next thing to watch is whether the Commerce Department turns this guidance into tougher enforcement actions, because that is when a policy clarification starts showing up in order books.
Also read: Apple gives Poke a rare lane into iMessage for AI agents • OpenAI makes ChatGPT memory more useful across conversations
