Inspur shipped about $3B in Nvidia Blackwell servers through a US subsidiary
The New York Times reported the servers reached Southeast Asia via Aivres, working around the parent company's blacklisting. Nothing about it required smuggling.
Inspur has shipped roughly $3 billion in Nvidia Blackwell servers to Southeast Asia through Aivres, its US subsidiary, according to the New York Times. The arrangement works around the blacklisting of the Chinese parent company.
What makes it notable is how ordinary it is. There is no concealed cargo, no falsified route, no third-country transhipment. A restricted entity has a subsidiary that is not restricted, and the subsidiary does the selling.
The design flaw in entity-based controls
US export controls operate largely through lists of named entities. The Entity List names companies; restrictions attach to those names.
Corporate identity, however, is a thing companies create. A subsidiary incorporated in a different jurisdiction, with its own name and legal personality, is a different entity as a matter of law even when ownership and control are not in doubt. Closing that gap requires either naming every affiliate as it appears, which is slower than affiliates can be created, or writing controls around ownership and control rather than identity, which is harder to draft and enforce.
The regime has been on the first approach for years, and this is the predictable result at $3 billion of scale.
Three cases, one pattern
Put this alongside the two Taiwanese cases from the same fortnight and a shape emerges.
On 24 August, prosecutors indicted nine people over 74 B300 servers routed to China through Indonesia, Japan and Hong Kong — restricted goods with falsified paperwork, the crude method, 43 percent interception.
On 31 August, prosecutors raided Unimicron over allegations that China-made circuit boards were relabelled Taiwan-made — contested origin documentation on unrestricted components, with no finding made.
And now a blacklisted parent selling through an unlisted subsidiary. Each is a test of a different assumption the control regime makes: that destination is knowable, that origin is knowable, that corporate identity is stable.
The third method is the cheapest and carries the least legal risk, which is why it moves the most hardware. Ninety-nine percent of the volume in these three cases went through the route that required no concealment at all.
Why Southeast Asia
The destination is doing work here too. Servers delivered to Southeast Asia are not necessarily servers used in Southeast Asia; the region has become the main staging ground for compute whose ultimate operator is difficult to establish, and capacity can be resold, leased or accessed remotely once it is racked.
That is the deeper problem with controlling access to computation by controlling the movement of hardware. A restricted party does not need to own an accelerator to use one. It needs an API endpoint.
The domestic alternative is arriving anyway
The strategic value of all this is time-limited. DeepSeek is planning a cluster of more than 160,000 Huawei accelerators at a one-gigawatt site in Inner Mongolia, among the largest known deployments of domestic Chinese silicon. China's Ministry of Industry and Information Technology has set a target of 9,800 EFLOPS by 2030 backed by 3.8 trillion yuan.
Workarounds matter while the domestic supply is not good enough. The controls are, in effect, buying years — and the $3 billion that went out through a subsidiary is a measure of how much they are not buying.
Nvidia, Inspur and Aivres have not commented, and the Commerce Department has not said whether it is reviewing the arrangement.
Runs the newsroom. Rename this profile in the studio to your own byline.
Related
Every weekday, the AI stories that moved money or shipped code.
No cross-posting, unsubscribe anytime. See all newsletters