Pentagon Completes Seven-Company AI Bloc — Anthropic Is the Only Major Lab Without a Deal
The US Defense Department announced on May 1 that Nvidia, Microsoft, Amazon Web Services, and AI startup Reflection AI have signed agreements for advanced artificial intelligence deployment on classified military networks. The additions bring the total to seven companies with such agreements: Google, OpenAI, SpaceX, Nvidia, Microsoft, AWS, and Reflection AI. Anthropic is the only major AI lab that does not hold one.
Each agreement permits “lawful operational use” — a formulation the Defense Department chose deliberately. It replaces the specific usage restrictions Anthropic had insisted on, including prohibitions on autonomous weapons assistance and limitations on targeting applications. The Pentagon rejected those terms, labelled Anthropic a supply chain risk — a designation previously reserved for Chinese companies like Huawei and ZTE — and voided the company’s $200 million military contract.
The Terms the Others Accepted
“Lawful operational use” is expansive. Defense officials briefed on the agreements have described it as covering targeting assistance, intelligence synthesis, and operational planning on secret and top-secret classification networks. No company that signed has disclosed what specific applications it has agreed to permit.
AWS did not object to the terms. Its spokesperson referred to the Defense Department as “the Department of War” — the agency’s name before 1947 — and said AWS “looks forward to continuing to support” its modernisation efforts. The AWS deal was negotiated late into Thursday evening, suggesting the Pentagon was moving quickly to complete the set.
The seven-company picture consolidates several months of bilateral announcements. Google signed first, agreeing to “any lawful government purpose” with no veto clause. OpenAI, SpaceX, and others followed in sequence. The completion of the group with Nvidia and Microsoft on May 1 effectively renders Anthropic’s absence the defining fact of the Defense Department’s AI posture, rather than a negotiating posture in progress.
Anthropic’s Position: Commercially Irrelevant, Symbolically Sharp
The commercial damage to Anthropic has been negligible. Its revenue run rate stands at approximately $30 billion. Its valuation is approximately $900 billion on secondary markets, up from $380 billion in February. The $200 million Pentagon contract it lost represents less than one percent of its annualised revenue. The gigawatt-scale compute deal with Google and Broadcom, due to come online in 2027, dwarfs it.
The dispute has been costly in a different register: Anthropic is now formally labelled a supply chain risk by the agency that defines what it means to be a trustworthy vendor to the US government. The Trump administration has extended the dispute by blocking Anthropic’s Mythos model from broader federal deployment and directing agencies to limit their use of Claude. Neither measure has had material revenue impact.
The underlying question — whether AI labs have the right to impose usage restrictions on governments as a condition of access — has now been answered in practice. The seven companies that signed accepted the government’s terms. The one that didn’t was replaced.
Reflection AI
Reflection AI, the least prominent of the seven signatories, is a San Francisco lab focused on building AI agents for autonomous work. Its inclusion alongside Nvidia and Microsoft suggests the Pentagon is building a diverse set of model providers rather than concentrating on the two largest labs. Reflection AI’s specific capabilities for classified use are not disclosed.