Nvidia’s Cloud Partner: A Backdoor for Banned Chips?

$92 million worth of AI servers built around U.S.-banned Nvidia chips reportedly ended up in China anyway—raising fresh questions about whether export controls are real barriers or just paperwork.

Quick Take

Chinese firm Sharetronic disclosed invoices for about $92 million in AI servers that appear to include Nvidia H100/H200-class GPUs restricted for export to China since 2022.
The invoices list 276 Super Micro SYS-821GE-TNHR servers and 32 Dell PowerEdge XE9680 servers, dated May–June 2025, and filed with Chinese agencies.
Super Micro and Dell have said they have no record of selling those servers to Sharetronic, while Sharetronic insists it sourced equipment through “legal channels.”
The disclosure surfaced as U.S. authorities charged Super Micro co-founder Yih-Shyan “Wally” Liaw in a separate alleged $2.5 billion server-smuggling case; he pleaded not guilty.

Invoices Put a Price Tag on a Suspected End-Run Around the 2022 Chip Ban

Sharetronic Data Technology, a Shenzhen-based company that has pivoted into AI data-center infrastructure, disclosed invoices showing purchases totaling 632 million yuan—about $92 million—for high-end servers in May and June 2025. The filings list 276 Super Micro SYS-821GE-TNHR servers and 32 Dell PowerEdge XE9680 servers, models typically configured with Nvidia’s top-tier AI accelerators. Those H100/H200-class chips have faced U.S. export restrictions to China without approval since 2022.

On their face, the invoices don’t identify the upstream seller, which matters because the legal line is not simply “China has them,” but how they were exported and by whom. The reporting also notes why “old stock” explanations are hard to square with the paperwork: the models listed are specific post-ban server configurations rather than vague references to generic GPU boxes. That specificity is what makes the documents a political and enforcement headache, not just a market curiosity.

Company Denials Collide With a Paper Trail and a Missing Middleman

Sharetronic has maintained that it procured the equipment through compliant, legal channels and has said it has no relationship with Super Micro. Super Micro has denied selling to Sharetronic, and Dell has said it has no record of sales to the company either, while emphasizing that it can take action when diversion is detected. That combination—detailed invoices on one side and “no record” statements on the other—keeps the key fact unresolved: who actually moved restricted hardware across borders.

Market reaction suggests investors saw more than a routine procurement story. Reporting indicates Sharetronic’s shares plunged after the disclosures, reflecting the risk that its financing plans and customer contracts could be disrupted if regulators or lenders question the collateral behind server-backed loans. Sharetronic had reportedly pursued bank lending with servers used as collateral, which means any uncertainty about provenance can quickly become a balance-sheet problem. In a constrained economy, that kind of compliance cloud can freeze capital as effectively as a legal injunction.

Nvidia’s “Cloud Partner” Link Highlights How Policy Can Get Blurry in Practice

A major complicating factor is that Sharetronic’s expansion into AI infrastructure has not been purely informal. In 2024, Sharetronic established a joint venture, Guangzhou Fcloud Technology, and that venture obtained Nvidia Cloud Partner status—one of a small group in China. That credential is not the same as permission to import restricted H100/H200-class chips, but it does show how commercial relationships and branding can coexist with tightening export policy. For Americans watching Washington, it feeds a familiar frustration: rules may be strict on paper while global supply chains remain porous.

The broader policy objective of export controls is to slow China’s access to leading-edge compute that could support military or intelligence applications. The problem is that enforcement depends on identifying transshipment routes, resellers, shell entities, and falsified end-use declarations—precisely the types of “in-between” actors that rarely appear on an invoice filed for a domestic loan or corporate disclosure. When the middle of the chain is opaque, accountability becomes diffuse, and the deterrent effect weakens even if the law itself is sound.

Smuggling Charges Add Pressure on U.S. Enforcement—and on U.S. Firms’ Compliance Systems

The timing of the Sharetronic disclosure drew attention because it coincided with U.S. charges against Super Micro co-founder Yih-Shyan “Wally” Liaw in an alleged scheme involving $2.5 billion in Nvidia-powered servers shipped to China; he has pleaded not guilty. The reporting does not establish that Sharetronic is tied to that case, and U.S. prosecutors have not publicly confirmed a connection. Still, the overlap underscores why voters across the spectrum increasingly doubt that the federal government can execute complex strategies competently.

$92 Million In Banned AI Chips Went From Super Micro To Little Known Chinese Tech Company https://t.co/nrGiziw1bj

— zerohedge (@zerohedge) April 13, 2026

For conservatives who backed tougher China policy and clearer “America First” boundaries, the central issue is straightforward: export controls are only as credible as enforcement. For liberals who worry about corporate accountability and national-security risk, the unanswered sourcing question lands in the same place: compliance programs that can’t trace where restricted hardware went are not functioning as advertised. Until investigators can identify the seller, the shipping pathway, and the approvals (if any), the public is left with circumstantial evidence, competing denials, and yet another example of institutions struggling to keep up with reality.