AI news ·
Nvidia requires cloud clients to sell at least 51% of services to non-Chinese firms
Nvidia now requires cloud clients to sell at least 51% of their services to non-Chinese companies, closing a gap where chips stay outside China but compute is rented to Chinese firms online.
Nvidia now requires its cloud computing clients to sell at least 51% of their services to companies that are not Chinese, according to research findings on the company's compliance program. The threshold is written into contracts between the chipmaker and the providers that rent out its hardware, and it targets a gap in export controls: a chip can stay physically outside China while its computing power is rented to Chinese firms over the internet.
The rule lands during a broader compliance crackdown, driven by pressure from Washington to stop advanced AI chips from reaching Chinese entities through indirect routes.
What the 51% floor does
The requirement sets a minimum, not a ban. Chinese customers can still buy capacity from an approved cloud provider. They just can't dominate that provider's book. Nvidia's threshold caps how much of any single provider's capacity can flow to Chinese firms.
The rule is one of several steps Nvidia has taken to tighten oversight of customers in Asia. In July 2026, the company cut its list of approved Asian customers by more than 50%, according to the research findings.
That cut hit so-called neoclouds hardest - newer providers in Singapore, Malaysia, and Japan that focus on renting out GPU computing power, often for AI workloads. Nvidia is now taking a much closer look at who those landlords rent to.
The vetting has also become far more hands-on. Nvidia's enhanced compliance checks now include:
- Physical inspections of data centers
- Contract verifications to confirm who the end customers are
- Interviews with end users of the hardware
The measures follow guidance from the US Commerce Department issued in May 2026. That guidance aims to curb the diversion of advanced technology to China. The chips most in focus include Nvidia's Blackwell-series processors, with the effort targeting access by Chinese intermediaries operating elsewhere in Asia.
How the workarounds took shape
US export controls on advanced chips began tightening in 2022. Since then, Chinese firms have looked for other routes to Nvidia's computing power.
One of those routes runs through overseas data centers. Chinese tech giants such as Tencent and Alibaba are reportedly obtaining hardware through overseas arrangements described as legally compliant. Tencent, in particular, is reportedly securing AI chips through data centers in Japan and Australia.
Nvidia's 51% threshold puts limits on that arrangement without closing it. The scrutiny has also reached Nvidia's own financial disclosures: the company modified its revenue reporting for fiscal 2026 to better reflect where customers are located.
Who carries the cost
For Nvidia, the research findings note that demand for its advanced chips remains significant despite the cuts to its client list. Cutting approved Asian customers by more than half means turning away some buyers.
For neocloud operators in Singapore, Malaysia, and Japan, any provider whose client base leans heavily Chinese now faces a choice. It can rebalance toward non-Chinese customers or risk losing access to Nvidia hardware. The inspection regime adds friction on top: site visits, contract reviews, and end-user interviews take time and paperwork.
For Chinese tech firms, the research findings suggest a potentially prolonged period of limited access to advanced AI technology. That could weigh on operational capabilities and growth prospects, particularly for companies building large AI models that depend on high-end compute.
For investors, the findings point to possible price swings in semiconductor stocks as markets digest these regulatory moves.
Why this matters for your organization
If your company buys GPU capacity through a cloud or neocloud provider, expect vendor due diligence to reach your own contracts and staff. Providers facing the 51% test will push end-user verification down the chain, so procurement, legal, and compliance teams should be ready to document who ultimately uses the compute and where. Firms in regulated sectors that depend on third-party AI infrastructure should also model the risk that a preferred provider loses Nvidia access and has to be replaced. Teams tracking export-control exposure can build the underlying policy knowledge through AI regulatory compliance courses.