SI Agent News
DevelopingOpen modelsChipsSecurity

Analysis suggests Chinese AI models were trained on restricted Nvidia chips

Evidence found in the weights of Chinese models like Kimi K3 suggests they may have bypassed US export bans to access high-end hardware.

2 millionadvanced chips estimated to reach China in 2024 and 2025

The short version

Why it matters

This discovery suggests that current US export controls may not be fully preventing China from accessing the hardware needed to compete in AI. It also highlights the growing complexity of monitoring global hardware supply chains as companies use offshore hubs to bypass restrictions.

Traces in the weights

New analysis suggests that Chinese AI labs may have successfully obtained Nvidia Blackwell GPUs despite strict US export bans. These high-end chips are central to the ongoing technological competition between the United States and China. The findings were based on an examination of the internal structures of several prominent Chinese models.

Will DePue, who previously worked at OpenAI and now leads Unemployment Labs, analyzed the weights of Chinese AI models. Weights represent the internal connections between simulated artificial neurons and essentially contain the knowledge of a large language model. Because these weights reflect the specific configurations of the hardware used during training, they can act as a fingerprint for the chips involved.

DePue reported finding convincing evidence that the Kimi K3 model from Moonshot AI and the V4 models from Deepseek show traces of Nvidia training. This discovery is significant because these specific chips were intended to be blocked from entering the Chinese market. The presence of these signatures indicates that Chinese labs may have had access to hardware that the US government sought to restrict.

The export ban landscape

The United States has implemented rigorous export controls to maintain its lead in the artificial intelligence sector. These measures were designed to slow down China's development of advanced models by limiting their access to high-performance computing hardware. However, the gap in AI capabilities between the two nations has narrowed significantly in recent years.

According to the AI Index Report 2026 from Stanford University, the US lead in AI was only 2.7 percent as of March 2026. This narrowing of the gap has prompted concerns among US officials regarding the effectiveness of current trade restrictions. Some analysts have expressed surprise at the speed with which Chinese startups are releasing models that can compete with top US systems.

To address these concerns, the US government established a Super Intelligence Force in October. This taskforce is led by the director of the US intelligence service and aims to balance innovation with necessary regulation. The goal is to ensure the US remains at the forefront of the AI race while preventing the technology from being used in ways that compromise national security.

Smuggling and offshore routes

There are several theories regarding how restricted Nvidia chips may be reaching Chinese laboratories. One possibility is that the hardware is being moved through complex, indirect supply chains. The research organization Epoch AI estimates that roughly 2 million advanced chips reached China through such methods during 2024 and 2025.

Another common method for Chinese companies is the use of offshore data centers. By placing their hardware in a third country, these firms can utilize high-performance computing power without directly violating export laws. This practice allows them to train large models while maintaining a layer of separation from the restricted territory.

Specific allegations have also surfaced regarding illegal activity by certain startups. A technology advisor to the White House has cited intelligence sources claiming that Moonshot AI used Nvidia chips illegally. Additionally, reports have emerged of a company linked to a suspected spy being involved in smuggling chips to a data center in Shanghai.

Corporate and legal fallout

The situation has also created significant turmoil for Nvidia as a corporation. Jensen Huang recently stepped down from his leadership position at the company following allegations of an affair with a woman suspected of spying for China. These allegations have led to a federal investigation by the FBI into the matter.

Nvidia's stock has suffered a notable decline following these events. As reported by Tagessau, the share price fell by approximately one-third from its peak in early October. The company has faced scrutiny for its long-term efforts to open the Chinese market to its products, even when restricted to lower-performance chips.

The legal and economic consequences of these developments remain unclear. While the presence of Nvidia traces in Chinese models does not automatically prove a violation of export laws, it highlights the difficulty of enforcement. The ongoing investigation and the fluctuating stock price reflect the high stakes of the current geopolitical and technological landscape.

What to watch

Try it

Look up the Stanford AI Index Report 2026 to see the current percentage gap between US and Chinese AI capabilities.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by t3n · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news