China’s tech giant Tencent has found a way to access Nvidia’s banned Blackwell AI chips, without ever importing them into China.
According to reporting by Barron’s, Tencent is using cloud computing infrastructure based in Japan and Australia, operated by Tokyo-based data center firm Datasection, which is deploying Nvidia’s latest Blackwell GPUs. While Chinese companies are barred from owning or operating these chips domestically, current U.S. export rules allow approved foreign cloud providers to run them outside restricted countries.
In short: Tencent doesn’t own the chips, it rents the computing power.
Importance of Blackwell
Nvidia’s Blackwell architecture is its most advanced AI platform to date and is central to training and running next-generation AI models. While the U.S. recently allowed Nvidia to sell its H200 chip to Chinese customers, Blackwell and the upcoming Rubin generation remain off-limits.
That makes cloud access extremely valuable for Chinese firms whose AI workloads are growing rapidly.
Nvidia Says It’s Legal
Nvidia confirmed that the setup complies with existing regulations. A spokesperson said export rules are designed to allow cloud infrastructure to operate outside restricted countries, arguing that overseas demand helps maintain U.S. technology leadership.
From Nvidia’s perspective, the chips stay in allied nations, and American firms still get paid.
Regulators See A Loophole
Not everyone agrees. Some U.S. lawmakers have flagged cloud-based access as a regulatory blind spot, warning that it undermines the spirit of export controls meant to limit China’s access to cutting-edge AI.
“This is a dangerous loophole,” said Rep. Mike Lawler, who has backed legislation to give regulators stronger tools to address indirect access to restricted technology.
Analysts note that this case is especially sensitive because Japan is one of Washington’s most trusted allies, unlike earlier cases involving Southeast Asian data centers.
AI Borders Are Dissolving
The Tencent–Datasection arrangement highlights a broader reality: the AI chip battle is increasingly being fought in the cloud, where physical chip location matters more than corporate nationality.
As long as compute can be rented remotely, controlling who uses advanced AI hardware becomes far harder than controlling who owns it. Whether regulators tighten the rules, or quietly accept this new reality, could shape the next phase of the global AI race.
