Anthropic Beat the Pentagon. What the Ruling Settles — and Doesn't

A federal judge ruled Thursday that the Pentagon's blacklisting of Anthropic was unlawful — permanently retiring the "supply chain risk" label and blocking the order that told every federal agency, and every company doing business with the military, to cut ties with the maker of Claude, the AI models that have become a default working tool across corporate America. The 59-page decision makes permanent the temporary block from March, and it lands at the most consequential moment in Anthropic's short life: a revenue run rate that climbed from about $9 billion at the end of 2025 to more than $65 billion by the end of July — a more than sevenfold increase in roughly seven months — an S-1 filed with the SEC in June, and a private valuation of roughly $965 billion after its last raise.

The run rate is the number to keep, because it explains why this fight was never only about the government's shopping list. The "supply chain risk" designation is an obscure procurement tool, reserved until now for companies tied to foreign adversaries — never before deployed against an American firm. On February 27, President Trump ordered agencies to stop using Claude, and Defense Secretary Pete Hegseth announced the designation on top of it, warning that no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic. Read that clause slowly. It is a wall not around federal procurement but around the whole of Anthropic's customer base, because in this economy "anyone who sells to the military" covers a large share of the Fortune 500 — and, critically, the cloud providers that carry Anthropic's compute.

The direct federal contract at issue ran into the hundreds of millions of dollars, and Anthropic told the court the label jeopardized that much on its face. But the exposure behind the label was larger by an order of magnitude. In the parallel case, a federal appeals court acknowledged billions in lost revenue and contract terminations. That is the world the ruling protects: Anthropic has committed more than $100 billion to Amazon over ten years, securing up to five gigawatts of Amazon's Trainium capacity, with Amazon investing up to $25 billion more on top of earlier rounds. A national-security label that forced military vendors to cut every commercial tie reached straight into that entire structure. The judge took it down.

It matters that the court did not rule on a technicality. The Pentagon's stated grievance was that Anthropic refused to strip out guardrails keeping Claude from fully autonomous lethal weapons and mass domestic surveillance — and that the company said so in public rather than quietly complying. "The empty invocation of national security is not a blank check to punish and retaliate against government critics," Judge Rita Lin wrote, finding retaliation in violation of the First Amendment, a denial of the process the Fifth Amendment requires, and a designation she called illegal and baseless. It was the same finding as March, when she wrote that nothing in the governing statute supports "the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." The guardrails stayed in; the label came off.

Now the part a headline buries. Thursday's ruling restores Anthropic's right to compete for government and contractor business. It does not require anyone to buy from the company. A separate, government-wide designation is still pending in the D.C. Circuit, where Anthropic already lost an emergency bid to pause the label in April, and the government can appeal. Some federal agencies said they planned to move to rival AI providers, and reports of a Pentagon pact with OpenAI surfaced the same week as the ban. The same pressure ran through other valves during the fight: export controls hit Anthropic's newest systems in June before being lifted weeks later. A resolved chapter in one courtroom is not a signed peace across the federal buyer.

There is also a competitive thread no judge can settle, worth holding onto as the IPO approaches. Anthropic's guardrails are a product choice, and it is the choice that started this entire fight. The court decided the government cannot punish the company for drawing that line. It did not decide the government will buy from Anthropic anyway. That tension is durable — a self-imposed ethical limit sitting between the company and a meaningful public customer — while the revenue stays overwhelmingly enterprise, driven in large part by the Claude coding agent's traction among developers.

For an investor, the honest measure of Thursday's ruling is that it changed the risk around the economics, not the economics themselves. The revenue that took Anthropic from $9 billion to beyond $65 billion in run rate was never going to be won or lost in the federal budget; the thesis was always whether the company can keep converting developer and enterprise adoption faster than its compute bill grows. What Thursday changed is the premise beneath that thesis: the single largest buyer in the world tried to exile the fastest-growing seller in the AI trade over a policy dispute, and a federal judge said no. For the AI trade — and, until the ticker exists, for Amazon's stake as the closest public stand-in — that is regime risk, partially retired.

Keep the two facts separate. A court just protected Anthropic from being declared a security threat to the market it sells into; it did not hand the company a single federal customer, and a second case is still live. The judge restored Anthropic's right to exist in the market. The business — winning users, pricing compute, staying ahead of a compute bill now measured in the hundreds of billions over a decade — was always fought somewhere else, and that is the fight that decides what this company is worth the day it lists.