Pressure on the control of advanced GPUs towards China has escalated again following the emergence of tax and financial documents indicating the purchase of nearly 300 AI servers by a Chinese company linked to the data center ecosystem. The case alone does not prove a violation of U.S. export restrictions, but it raises questions about how effective that technological embargo truly is in practice.
The company at the center of the story is Sharetronic Data Technology, based in Shenzhen. According to Bloomberg, after reviewing tax records from China’s State Administration of Taxation, the company sold 276 Super Micro SYS-821GE-TNHR servers and 32 Dell PowerEdge XE9680 to a Shenzhen subsidiary between May and June 2025 for a total value of approximately $92 million. The issue is that these models are compatible with AI accelerators that were then under U.S. restrictions for China.
For the Super Micro system, the manufacturer’s specs indicate support for NVIDIA HGX H100 and H200 with 8 GPUs. Regarding the Dell PowerEdge XE9680, official technical documentation shows compatibility with NVIDIA HGX H100, H200, H20, and H800, as well as other options like AMD Instinct MI300X and Intel Gaudi 3. Additionally, U.S. restrictions on advanced computing for China, initiated in October 2022 and expanded since then, make any such transactions particularly sensitive.
However, a key nuance exists. The existence of these invoices does not definitively confirm the specific accelerators installed in all those servers, nor does it, by itself, confirm that Sharetronic violated U.S. export regulations. The core of the story is that these documents suggest access to highly sensitive hardware, but the exact origin of the equipment and the final configuration of each system have not been publicly clarified.
The case comes amidst a turbulent time for Super Micro
The release of these documents occurs in a particularly tense context. In late March, the U.S. Department of Justice charged Yih-Shyan “Wally” Liaw, co-founder of Super Micro, along with two others, for allegedly conspiring to illegally divert AI servers—manufactured in the U.S.—to China via Taiwan and Southeast Asia. The indictment involves billions of dollars in technology and was followed by an independent internal investigation within the company.
Amid this climate, markets reacted strongly to any mention of related entities. Bloomberg reported that Sharetronic’s shares dropped to the 20% daily limit just hours after the scandal involving Super Micro broke. The company responded by stating that all its assets and equipment come from “legal and compliant channels,” and that it does not have any “business cooperation or relationship” with Super Micro.
Dell also distanced itself from the issue. In a statement reported by Bloomberg, the company said it had no record of the alleged sales and that, if any diversion of products or transfers to unauthorized clients or locations are detected, they act swiftly, including ending commercial relationships.
Nvidia’s role further complicates the case
One of the most uncomfortable aspects of this episode is that Sharetronic was not an unknown player in the AI infrastructure world. In 2024, it launched a joint venture called Guangzhou Fcloud Technology, and corporate documentation filed in Hong Kong indicates that this subsidiary obtained the NVIDIA Cloud Partner certification, which identifies cloud providers with infrastructure prepared for modern AI workloads.
This detail does not automatically imply that Nvidia directly supplied the questioned servers, but it adds a significant reputational layer to the case. According to a statement collected by Tom’s Hardware, an Nvidia spokesperson said “FCloud would need licenses and approvals from the relevant governments before acquiring H200 servers,” and that customers are given explicit instructions not to supply servers, support, or services without U.S. government approval. Nvidia also stated that its due diligence processes have supported investigations and prosecutions of alleged contrabandists.
In essence, Nvidia is emphasizing that, if there were diversions, they do not invalidate the company’s compliance framework. Yet the political issue remains: if a Chinese company associated with a Nvidia-recognized network gains access to hardware compatible with restricted GPUs, it reinforces the notion that sanctions do not always stop hardware flow, but may only increase costs, shift locations, or obscure the supply chain.
A symptom of the larger hole in the tech war
The core issue is not Sharetronic alone, nor Super Micro individually. The real debate revolves around whether the U.S. export control system is successfully preventing China from accessing the most coveted AI infrastructure—or if it’s pushing part of that market into increasingly opaque channels. The restrictions introduced in 2022 aimed precisely to limit Chinese access to advanced computing with military and strategic potential. Yet, emerging cases suggest that the problem lies not just in the regulations but in the entire chain of intermediaries, assemblers, integrators, distributors, and end-users.
For this reason, this episode is significant beyond the headlines. It’s not just about a legal verdict, but about how the global AI race also unfolds on the margins: through invoices, server configurations, licenses, re-exports, and suppliers who, on paper, should be out of reach for illicit operations. Meanwhile, each new document, accusation, and market plunge reminds us that AI geopolitics are increasingly decided not only in labs or data centers but also at customs, courts, and supply chains.
Frequently Asked Questions
Is it proven that Sharetronic violated U.S. sanctions?
Not definitively. The published invoices and documents point to purchases of servers compatible with restricted GPUs, but the exact configuration of each system and their final origin have not been publicly confirmed.
What servers are listed in the documents?
The records cited by Bloomberg mention 276 Super Micro SYS-821GE-TNHR and 32 Dell PowerEdge XE9680, sold in May and June 2025 to a Shenzhen subsidiary for approximately $92 million.
Why are those models sensitive?
Because the official models from Super Micro and Dell support accelerators like NVIDIA H100 and H200, and Dell’s also H20 and H800, all within the regulatory discussion on advanced exports to China during that period.
What’s the connection to Super Micro?
The case gained momentum after U.S. authorities accused Super Micro’s co-founder, Wally Liaw, along with two others, of plotting to divert AI servers to China, intensifying scrutiny across the entire supply chain.
via: tomshardware