How Anthropic vs. Pentagon puts billions at ‘risk’ for Nvidia

Representative Image. In pic: Nvidia CEO Jensen Huang

Nvidia has spent the last two years minting money as the arms dealer of the AI boom—selling chips to everyone, picking sides with no one. That carefully maintained neutrality is now under threat. Defence Secretary Pete Hegseth’s order barring any military contractor, supplier, or partner from conducting “commercial activity” with Anthropic directly imperils one of Nvidia’s biggest customers. The chipmaker sells the GPUs that keep Anthropic’s servers running—and without a steady supply of those, Anthropic cannot train or run its models at scale.Former Trump AI advisor Dean Ball flagged the problem immediately, calling the move “attempted corporate murder” and warning that Nvidia would effectively be forced to cut off chip sales to Anthropic if Hegseth’s order held.

Nvidia sells chips to everyone—that’s now a liability

Nvidia’s hardware-agnostic model has always been its superpower: sell to OpenAI, Google, Meta, Amazon, and Anthropic all at once, and let them fight it out in software. That approach minted Nvidia into a $3 trillion company. But Hegseth’s sweeping language—covering any company that does business with the US military—puts Nvidia in a bind it didn’t sign up for. It can’t stop selling chips to the Pentagon. And if it can’t sell to Anthropic either, one of its major AI customers is effectively gone.

A label designed for Huawei, now aimed at a US startup

The “supply chain risk” designation under 10 USC 3252 has a very specific history. It’s the same label applied to China’s Huawei in 2018, over fears Beijing was embedding surveillance backdoors into telecom hardware. Anthropic, by contrast, was the first frontier AI company to deploy models on US classified government networks—back in June 2024—and its Claude AI is actively used by the CIA and NSA for intelligence work. University of Minnesota law professor Alan Rozenshtein said the label “clearly was not designed” for a domestic contract dispute. Anthropic has pledged to fight it in court, and legal experts broadly agree that Hegseth’s commercial ban likely overreaches what the statute actually permits.For now, the risk to Nvidia is theoretical. But the precedent is unsettling—if Washington can weaponise a national security label to disrupt chip supply chains over a policy disagreement, no AI infrastructure company is truly insulated from the politics.