Taiwan has uncovered a secret plot to send Nvidia's most advanced chips to China, according to an indictment by the Keelung District Prosecutors Office. The case involves nine individuals, including a senior Nvidia manager in Taiwan, and highlights the ongoing challenge of enforcing U.S. export controls on AI technology.
The scheme allegedly involved diverting Nvidia B300 servers—equipped with Blackwell Ultra chips—to Chinese customers via Taiwan, Indonesia, and Japan. Prosecutors say the defendants falsified the final destination of the equipment to bypass restrictions.
What Happened: Indictment Details
The Keelung District Prosecutors Office has indicted nine people for their roles in a network that diverted Nvidia servers to China. According to the Financial Times, the central figure is a senior Nvidia executive in Taiwan surnamed Zhang. Two senior sales executives from distributor Super Micro were also indicted, along with a Taiwanese distributor and five unidentified intermediaries.

The prosecutors allege that the group moved 130 Nvidia B300 servers, of which 74 reached Chinese customers. The remaining 56 were intercepted. The servers are used for training AI models and contain chips subject to U.S. export restrictions.
How the Scheme Worked
The plan relied on misrepresenting the final destination of the servers. The defendants claimed the equipment was destined for a data center in Taiwan, but that facility lacked the network bandwidth and electricity needed to operate them. Instead, the servers were routed to China through Taiwan, Indonesia, and Japan.
This method of diversion is not new, but the involvement of a senior Nvidia employee and executives from a major distributor like Super Micro underscores the sophistication of such operations. It also raises questions about corporate compliance and the effectiveness of export controls.
Why It Matters: The Stakes for AI Chip Export Controls
The U.S. has imposed strict export controls on advanced AI chips to China, aiming to slow Beijing's progress in artificial intelligence and military modernization. Nvidia's Blackwell Ultra chips are among the most powerful available for AI training, and their diversion to China would undermine those efforts.
The case also puts a spotlight on Nvidia's internal controls and the role of distributors. As the world's most valuable chipmaker, Nvidia faces intense scrutiny over how its products reach end users. Any suggestion that its own employees facilitated smuggling could damage its reputation and invite tighter regulations.
Key Takeaways
- Nine people, including a senior Nvidia manager in Taiwan, have been indicted for allegedly diverting Nvidia B300 servers to China.
- The servers contain Blackwell Ultra chips, which are subject to U.S. export controls.
- Prosecutors say the defendants falsified the final destination, claiming the servers were for a Taiwanese data center.
- Of 130 servers moved, 74 reached Chinese customers via Taiwan, Indonesia, and Japan; 56 were intercepted.
- The case highlights the challenges of enforcing export controls and the potential for insider involvement.
Comparison of Nvidia's Advanced AI Chips
Nvidia's AI chips vary in performance and export restrictions. The table below summarizes key models mentioned in the context of export controls. The background to this story is set out in China's AI Export Controls: 40 Key Technologies Face New Rules. A related thread runs through China's NVIDIA Chip Replacement Plan Hits HBM Memory Snag. Earlier coverage in 11 Indicted in Human Smuggling Deaths in Texas Cargo Container shows how the argument developed.
| Chip | Architecture | Primary Use | Export Status to China |
|---|---|---|---|
| Blackwell Ultra | Blackwell | AI training | Restricted |
| H100 | Hopper | AI training | Restricted |
| A100 | Ampere | AI training | Restricted (with exceptions) |
| H20 | Hopper | AI inference | Allowed with license |
What to Watch Next
The indictment is likely to intensify scrutiny of Nvidia's supply chain and distributor networks. U.S. authorities may push for stricter compliance measures, and Taiwan could face pressure to tighten its own export controls. The case also serves as a warning to other companies and individuals involved in similar schemes. A closer look at the same subject appears in 11 Indicted in Human Smuggling Deaths of 6 in Texas Container. Anyone following the wider picture can start with 11 Indicted in Human Smuggling Deaths of 6 Migrants in Texas. The same theme surfaces in Human Smuggling Deaths: 11 Indicted in Texas Cargo Container Trag, which takes a different angle.
Nvidia has not commented publicly on the indictment, but the company has previously stated that it complies with all export regulations. Super Micro has also not issued a statement. The prosecutors' findings, if proven, could lead to criminal penalties and further regulatory action. Context on how this played out before sits in Nvidia Wins China Approval for H200, Tweaks Groq Chips. A companion piece, How a Panic Attack Uncovered a Hidden Childhood Trauma Secret, approaches the same subject from another angle.
FAQ
What is the Nvidia B300 server?
The Nvidia B300 server is a high-performance computing platform designed for AI training. It is equipped with Blackwell Ultra chips, which are among the most advanced AI processors available and are subject to U.S. export controls.
Why are these chips restricted from export to China?
The U.S. government restricts exports of advanced AI chips to China to prevent them from being used in military applications and to slow China's AI development. The Blackwell Ultra chips are capable of training large AI models that could enhance China's military capabilities.
Who is the Nvidia executive named in the indictment?
The indictment names a senior Nvidia executive in Taiwan surnamed Zhang as the central figure in the scheme. His full name has not been released. Nvidia has not commented on his status.
How many servers reached China?
According to prosecutors, 74 of the 130 Nvidia B300 servers moved by the network reached Chinese customers. The remaining 56 were intercepted.