A blacklisted server giant shipped $5.6B of US AI gear through its Valley subsidiary

Share
A blacklisted server giant shipped $5.6B of US AI gear through its Valley subsidiary

The New York Times spent months on shipping manifests and came back with the cleanest map yet of how export controls actually fail: not at the border, but at the name on the door.


Inspur, blacklisted by the US since 2023, moved at least $5.6 billion of advanced technology out of the United States between April 2024 and February 2026 — including more than $3 billion of Nvidia Blackwell servers — through a Silicon Valley subsidiary the government never listed. The company put on the Entity List was Inspur Group. The company shipping is Aivres, its US subsidiary, which kept the same offices, staff and business after the sign outside was changed. From there the hardware went to data centers and firms in Southeast Asia that serve Alibaba and ByteDance, and to a Malaysian assembler that ships unlabeled, unusually expensive servers into China under a new state-owned brand called Maginfra — more than $700 million of them in six months, over 1,500 priced like high-end AI machines.

Two details do the real work here. The first is that nothing about this needed a smuggling boat. Remote access to foreign data centers is barely restricted, and the Trump administration rolled back the Biden-era rules that touched it; both the administration and Congress are still writing replacements. The second is that enforcement has stopped. Six current and former Commerce officials described to the Times an agency that has lost its appetite for China or Nvidia cases, and ten months have now passed without a single firm added to the Entity List — the longest stretch in 18 years. Read that next to the numbers and the question stops being whether controls work and becomes who is still checking.

The companies are not accused of breaking the law as written. That is the point. Inspur rerouted through a subsidiary the order did not name, and in China it re-registered its main business at a building down the road days after the listing. Nvidia says it does not support diverted products and sells to partners who certify compliance; in May it named Aivres among the companies in full-scale production of its next-generation servers. William George of ImportGenius, who analyzed the trade records, put it flatly: it is hard to look at this and not see a state-sponsored network for bypassing export controls. The Trump administration approved Maginfra for a license to buy Nvidia's less powerful H200 chips this year anyway.

Export controls have always rested on the assumption that a blacklist names a company. This reporting shows the list names an address, and the address moved. Congress and the administration are both drafting remote-access rules right now, with Trump and Xi meeting on September 24 — which makes the drafting window the story, not the deadline.

If a sanctioned firm can keep buying the best American chips by renaming the subsidiary, does the Entity List still mean anything? Tell us in the comments.

Read more

Byteification turns Qwen3 and Llama 3 into byte-level models

Byteification turns Qwen3 and Llama 3 into byte-level models

The tokenizer is the last piece of the language-model pipeline that was never learned end to end — and a Nature paper out this week shows it can be swapped out for less than one percent of a pretraining budget. Researchers have retrofitted four open subword language models into true byte-level models using less than 1% of a typical pretraining budget, and the result — published 7 October in Nature — closes the performance gap that has kept byte-level models a niche curiosity for nearly a decad

OpenAI's dots go mobile and can start your Codex work

OpenAI's dots go mobile and can start your Codex work

One product update and two benchmark papers this hour, all circling the same question: how much of the work can you actually hand over, and what happens to the part you didn't? OpenAI's dots — the always-on agents that get their own cloud computer — can now be created and configured entirely from the ChatGPT app on iOS and Android, and a dot can start Codex work or pick up an existing Codex thread on your behalf. The October 9 release, billed as the first of a fresh run of updates for dots, mo

Coding agents as a hospital: Cockroach's five-month results

Coding agents as a hospital: Cockroach's five-month results

The agentic-coding debate keeps being framed as speed versus trust. Cockroach Labs just published the most detailed trust-first ledger yet — five months of numbers, including the parts that didn't work. Cockroach Labs ran its codebase like a teaching hospital for five months and says the pipeline merged well over a million lines of code with seven total reverts. In the setup the company calls MOLT Sinai, GitHub issues are patients, merging is discharge, and the humans in charge are Chiefs of M

Today in AI — October 10, 2026

Today in AI — October 10, 2026

A day where the boring machinery took center stage: proof checkers, order books, warehouses and the humble text message — plus fresh arguments about who gets to declare any of it safe. Models & Research * Mathematicians get a reliability primer for the Lean Theorem Prover. A guest post by Thomas Hales on Terence Tao's blog walks through what Lean actually guarantees — and devotes itself to the "Summer of Soundness Bugs," the string of kernel bugs found this summer that let false proofs thr