The US Commerce Department has introduced interim rules requiring licences for advanced AI chips exported to data centres in Singapore, Malaysia, Thailand, Indonesia and Brazil. The rules also put overseas data centres serving Chinese and other restricted-country companies under review, potentially widening the impact beyond Nvidia to AMD and other suppliers. For investors, the key question is whether export controls will extend from specific buyers to the global AI infrastructure supply chain.
Rules cover data centres in five countries and overseas computing services
The Commerce Department said the new licensing requirements are intended to address the risk that advanced chips could be rerouted to restricted buyers. They also target overseas data centres that provide substantial computing capacity to companies headquartered in China and other restricted countries. The measure could make it harder for Chinese AI firms to access computing power abroad as a way around restrictions on direct chip exports.
The scope of scrutiny may therefore reach beyond who buys a chip. Where it is installed and which customers ultimately use its computing capacity could also become relevant to licence reviews. Data-centre operators and chip suppliers may face closer examination of customer identities, equipment use and the flow of computing resources. The interim rules’ implementation details, and their effect on specific transactions, remain unclear.
Scrutiny of Nvidia-related sales draws attention
The rules come as Nvidia’s chip sales linked to China face heightened scrutiny. Allegations that intermediaries in Singapore helped resell advanced chips to Chinese customers are still under investigation, with no conclusion announced so far.
Dan Ives of Yorkville Ives & Co. said the dispute highlights differences in Washington over Nvidia’s strategic role, CEO Jensen Huang’s leadership and the effectiveness of existing export controls. For markets, the regulatory consequences could extend beyond Nvidia’s direct customers to other companies supplying AI accelerators, custom chips and networking equipment.
AMD’s licences and sales opportunities are also in focus
In its latest quarterly filing, AMD said US export licences allow it to ship MI325 AI accelerators to some customers in China, subject to US inspections. The company also warned that changes to export restrictions could constrain sales, raise compliance costs and prompt customers to switch to competitors not subject to the same limits.
The policy shift presents both a potential opening and a risk for AMD. If Nvidia has fewer customers it can serve, AMD may have more room to compete for orders. But if controls expand to additional overseas markets, AMD’s own licences and customer reach could also be affected. Its ability to take on any additional demand will depend on the scope of its licences and its capacity to deliver products.
Broadcom faces wider industry effects, not the same direct restrictions
Broadcom is part of global AI infrastructure investment through its custom AI accelerators and networking business. The current rules do not explicitly impose the same direct restrictions on Broadcom as on Nvidia. However, a broader review of advanced chip exports could change how markets assess the AI supply chain as a whole.
The stated policy aim is to protect the US lead in AI, but restrictions could also encourage overseas customers to diversify semiconductor suppliers and reduce their reliance on US technology. Nvidia has previously warned that export controls could leave it at a disadvantage against unrestricted competitors and lead customers to use fewer US chips in their designs.
The next question is whether US measures remain focused on particular regions and customers or place broader limits on global AI infrastructure construction and procurement. For Nvidia, AMD and other suppliers, the rules’ reach, licence decisions and underlying customer demand will all shape the effect on sales and competition.