The US considers treating offshore cloud access to advanced AI hardware as an export-control issue, signalling a potential tightening of regulations impacting Chinese AI firms and global data-centre operators.
Washington is moving closer to treating offshore cloud access to advanced AI chips as an export-control problem, not just a compliance loophole. According to Bloomberg reporting cited in the lead article, the Bureau of Industry and Security is reviewing whether legally structured compute-rental arrangements used by Chinese AI companies should fall within the scope of US restrictions, even when no chips cross a border. That would mark a significant shift for data-centre operators in places such as Malaysia, Singapore, Japan and the United Arab Emirates that have built businesses around serving Chinese customers.
The issue is a familiar one for regulators and an increasingly acute one for industry: current export rules were written for physical shipments, not remote access. BIS advisory opinions from 2009, 2011 and 2014 concluded that cloud providers do not become exporters simply because a foreign customer uses computing power hosted on their servers. In practice, that has meant a Chinese company can train a model on Nvidia processors in a third-country data centre without triggering the usual export event, so long as the hardware itself is not shipped into China.
That legal gap has become more consequential as advanced GPUs have turned into strategic assets. The lead article says offshore compute arrangements could be boosting China’s effective access to advanced US compute by at least 60% in 2026. Reuters has also described the broader enforcement environment as unusually fractured, with BIS pausing some actions while continuing to warn that chip access rules for China-linked entities remain in force. At the same time, Nvidia has tightened its own compliance checks in Asia, underscoring how much of the practical policing is now being done by the company that makes the chips.
The latest policy signal from Washington is not a formal ban on remote access, but it is moving in that direction. The House passed the Remote Access Security Act earlier this year, and the bill would amend export-control law so that “remote access” to controlled technology could itself require licensing. Tom’s Hardware reported that the measure is designed to close the offshore rental route used by some Chinese firms to reach restricted Nvidia and AMD hardware. If enacted, it would give BIS a clearer statutory basis to regulate cloud access to GPUs in the same way it regulates physical exports.
BIS is also tightening related controls on direct sales. In May, the agency said advanced computing exports to China-linked entities remain licensable even when the recipient is outside China or Macau if its parent is headquartered there. In a separate June update, BIS shifted some semiconductor export reviews to a case-by-case process, including for Nvidia’s H200 and AMD’s MI325X chips. Tom’s Hardware reported that this has allowed some limited H200 activity to resume, but with third-party verification and revenue-sharing conditions attached.
For data-centre operators, the practical message is clear: the legal distinction between hosting chips and exporting them is under pressure. Contracts, customer screening and location controls are now part of export-risk management, not just cloud operations. The current review may not produce immediate prosecutions, but it appears to be laying the groundwork for a tougher rule set that could reach well beyond the traditional definition of an export.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
