Banned Nvidia Chips Power China's Best AI, Report Says
This could affect US tech leadership and your wallet as chip smuggling grows.
A new report claims that China's most advanced open AI models, from companies like DeepSeek and Moonshot, were trained on Nvidia's powerful Blackwell chips. These chips are banned from export to China by the US government, which wants to keep AI superiority. The evidence comes from analyzing the models' internal structures, which show fingerprints of the banned chips.
The problem is bigger than just smuggling. Nvidia CEO Jensen Huang recently left the company after admitting an affair with an executive at Megaspeed, a firm accused of smuggling chips into China. While that's juicy gossip, the real issue is that Nvidia might face legal trouble for export violations. The US bans Blackwell chips to China, but allows weaker H200 chips with a 25% tariff if shipped via a roundabout route.
However, it's not certain that laws were broken. The chips could have been in offshore data centers, like in Malaysia, where 275,000 Nvidia AI chips were imported in 2024-2025. That's a loophole: the ban aims to stop Chinese labs from training on top chips, but if they can just go to Malaysia, it's not very effective. Experts say the US should control who uses the chips, not just where they are.
This matters because if China develops its own chips, Nvidia could lose billions in sales. Already, 600,000 H200 chips are stockpiled in Hong Kong, waiting to enter China. The US wants to keep its tech edge, but the current rules may not be enough. For you, it means the AI race is heating up, and the outcome could affect which AI tools you use and how much they cost.
- Chinese AI models like DeepSeek and Moonshot were trained on Nvidia's banned Blackwell chips, according to new research.
- Nvidia CEO Jensen Huang left the company amid a scandal involving chip smuggling and an affair with a Megaspeed executive.
- The US export ban may be ineffective because chips can be used in offshore data centers like in Malaysia, which doesn't violate the rules.
Why It Matters
This could weaken US tech leadership and lead to more regulations that affect AI prices and availability.