Taiwan prosecutes the staff, not the vendors, in a $2.5bn AI server route into China

The mechanism Taiwanese prosecutors describe is clerical. Export documents and shipping records were falsified to hide where Super Micro machines running Nvidia processors actually landed ^[2][3]. Restricted-goods screening is built to interrogate the counterparty, and a Taiwan-based sales manager routing orders through normal channels is not a counterparty. The people alleged to have broken the rule were the same people who filled in the forms.

The pace is the part worth pricing. US prosecutors say roughly $510 million moved through a Southeast Asian intermediary that supplied fake records, between late April and mid-May 2025 ^[5][10]. That is about a fifth of the $2.5 billion attributed to the whole scheme, inside roughly three weeks ^[13]. A route running at that rate leaves a reconciliation trail between stated end-use and actual destination, and nobody inside the chain stopped it.

Super Micro's public position is that it is not a defendant in the US case ^[8]. That is accurate and narrower than it sounds. The March charges from the US Attorney's Office for the Southern District of New York named co-founder and board member Yih-Shyan "Wally" Liaw as one of three people behind the operation, alongside a Taiwan-based sales manager and a contractor ^[7]. Liaw holds a $464 million stake in the company ^[18]. Corporate non-liability does not keep an indictment off the board or out of the register of large holders.

Two prosecutions now cover the same shipments, and neither depends on the other. Taipei's motive is domestic: in July its prosecutors charged a former TSMC employee over the alleged theft of chip know-how that Taiwan classifies among its national core assets, a case investigators say was aimed at China ^[9]. An authority willing to charge its own foundry alumni has no particular difficulty charging a foreign vendor's local hires, and it works on evidence Washington does not control.

Jensen Huang's formulation in Taipei, that Nvidia insists its partners are compliant ^[16], is both the right thing to say and a statement of the limit. The leverage a chip designer has over a distributor's paperwork is contractual. Nvidia is meanwhile still pitching its licensed H200 as the legal way into China, without approval from Beijing ^[12]. That request is now read against $2.5 billion of hardware that arrived without one.