Taiwan has indicted nine people in a case that reveals how restricted AI hardware allegedly moved from a controlled supply chain toward Chinese buyers. The case pulls Nvidia GPUs, Supermicro servers, false paperwork, overseas routing, and a growing battle over remote access into one urgent question: can export controls stop advanced AI compute once the hardware leaves the factory?
On August 24, 2026, Taiwan’s Keelung District Prosecutors’ Office indicted nine people over the alleged illegal resale of Supermicro AI servers equipped with Nvidia’s advanced B300 GPUs to buyers in China. The defendants include Chang, a distribution manager at Nvidia’s Taiwan office; Lin and Wang, sales managers at Supermicro’s Taiwan branch; and Lu, the chief executive of Albatron Technology, a Supermicro distributor.
The indictment offers the most detailed public account yet of how restricted U.S. AI hardware moves through Taiwan’s distribution chain to Chinese end users. It describes a sales channel that exists only because of export controls, with each step built around approved buyers, verification documents, and restrictions on resale.
How the alleged server pipeline worked
Supermicro’s advanced AI servers with Nvidia B300 GPUs face strict controls from both companies. Buyers must sit on Nvidia’s approved whitelist, submit end-user and end-use documentation for verification, and accept a no-resale condition. Any order above eight servers triggers site inspections by sales staff and technicians from both companies.
The alleged scheme began in February 2025, when two men surnamed Chen at server sales company Flying Tiger Technology obtained a whitelist position and lined up a buyer. The defendants allegedly concealed that the data center provider named in the paperwork lacked the electrical capacity and bandwidth to house 130 servers, then had Chang tell Nvidia that the required verification was complete.
That approval unlocked the sale of 130 servers to Flying Tiger. After the company received the first two servers and its original buyer failed to pay, Lu allegedly referred Flying Tiger to a Chinese buyer.
A second transaction involved another 64 servers. The head of trading company Long Wins learned about the hardware through Wang at Supermicro’s Taiwan branch. Wang allegedly acknowledged the illegal sale to Lin and agreed to split the proceeds.
The 64 servers from that deal, along with the two servers from the first transaction, reached Chinese buyers in three batches. Fifty of them were routed through Indonesia. In a third transaction involving another 64 servers, eight were shipped through Japan and Hong Kong, while the remaining 56 were seized in Taiwan.
The Flying Tiger executive still at large received more than $21.2 million in illicit proceeds from the resale of 74 servers. The head of Quintai Electronics issued four false invoices to conceal the money flow, diverting about NT$39.16 million in Albatron assets.
- The first transaction involved 130 servers.
- A second transaction involved 64 servers.
- A third transaction involved another 64 servers.
- Fifty servers were routed through Indonesia.
- Eight servers were shipped through Japan and Hong Kong.
- Fifty-six servers were seized in Taiwan.
Ho, Lin, and Lu face separate charges of aggravated breach of trust under Taiwan’s securities law. The nine principal defendants face charges of breach of trust and forgery.
Controls target chips, but compute can travel
Taiwan’s investigation runs alongside a U.S. prosecution involving a similar structure, with that case charged on March 19, 2026. The parallel cases show how export enforcement now reaches beyond chip manufacturers and into distributors, data center paperwork, trading companies, and shipping routes.
But the hardware itself is only one part of the problem. Chinese AI firms are also accessing Nvidia compute power through overseas data centers despite U.S. export restrictions, creating a gap between controlling physical chips and controlling access to the machines that run them.
“Controls physical AI chips. It does not cover remote access to those chips,” said Cassia King.
Moonshot AI has been accused of using Nvidia chips through overseas data centers, and White House official Michael Kratsios accused the company of using Nvidia’s GB300 chips in Thailand. ByteDance, Alibaba, and Tencent are also accessing Nvidia compute power overseas. Aolani, a cloud provider working with ByteDance in Malaysia, said, “Any permitted access to our services, infrastructure or technology is fully compliant with all applicable regulations.”
The Institute for AI Policy and Strategy provided analysis on export controls, while the U.S. Department of Commerce and Bureau of Industry and Security remain involved in U.S. export regulation. A White House official said, “The Trump administration has implemented the most rigorous export control regime in modern history, and remains committed to safeguarding America’s national and economic security.”
Michelle Nie described the goal in direct terms: “The point of chip export controls is to deny China the ability to train frontier AI using advanced U.S. chips.” The Taiwan case shows why that goal depends on more than blocking direct shipments. It also requires tracking who receives servers, where they operate, how they are financed, and whether users can reach restricted compute from another country.
The data center race widens the stakes
The pressure will grow as Southeast Asia adds more infrastructure for AI computing. Plans call for 31 data centers of at least 100 megawatts across Malaysia, Indonesia, and Thailand, compared with two today. Global data center capacity could roughly double to 200GW by 2030.
Those numbers point to a fast-expanding network where servers can sit in one country, customers can operate from another, and AI workloads can move through cloud access without the chips crossing a national border. The plan for two data centers in Malaysia, Indonesia, and Thailand, compared with two today, underlines how quickly the regional footprint is set to expand.
That expansion also gives investigators a sharper map of the challenge. Taiwan’s indictment follows the movement of specific servers through distributors, buyers, false invoices, and transit countries, while the overseas compute cases focus on access to Nvidia hardware after it reaches data centers outside China.
Meanwhile, Supermicro announced on August 20, 2026, that an independent investigation led by its board cleared CEO and co-founder Charles Liang of any knowledge of the alleged scheme to smuggle $2.5 billion in hardware to China. The investigation found no evidence that current members of Supermicro’s senior management knew about the scheme.
The next phase of AI export enforcement will test whether governments can control an entire compute pathway instead of a single shipment. Taiwan’s case puts that pathway on display, and the outcome could shape how Nvidia hardware reaches customers, cloud providers, and AI developers across the world.
Based on
