Pentagon picks seven AI firms

The Pentagon just moved commercial AI one layer deeper into the national-security stack. It signed agreements with seven companies — OpenAI, Google, Nvidia, Microsoft, Amazon Web Services, SpaceX, and the startup Reflection — to deploy AI tools inside classified Defense Department networks. That matters because this is not ordinary cloud work. These are the environments used for some of the military’s most sensitive data and operations. The real news is not just that the Pentagon wants more AI. It is that it now wants multiple outside vendors running inside its highest-trust systems. ### What did the Pentagon actually do? It reached formal agreements with those seven firms to make their AI capabilities available on classified Defense Department networks for what it called lawful operational use. The list is notable on its own — hyperscalers, model labs, a chip company, SpaceX, and a small startup to support warfighters. ### Why do IL6 and IL7 matter? Because those labels are the gate. The agreements cover Impact Level 6 and Impact Level 7 environments — the Defense Department’s classified trust tiers. IL6 is used for classified national-security information, and IL7 sits even deeper in special, highly restricted environments — the compute, the security controls, and the access rules all have to survive classified deployment standards. ### Why pick seven vendors instead of one? Basically, the Pentagon does not want vendor lock. Defense One said officials explicitly described the spread as a way to avoid depending on a single provider. That makes sense. Different firms bring different pieces — cloud hosting, frontier models, inference hardware, secure networking, or custom agents. In practice, the Pentagon is building a portfolio, not crowning one winner. ### Why is SpaceX and Reflection in this group? That is part of what makes the list interesting. This is not just the usual cloud trio. SpaceX suggests the Pentagon wants AI tied to secure communications and operational systems, not just back-office analysis. Reflection, a much smaller company, shows the door is open to newer model builders in the supply chain.” That last point is an inference from the vendor mix, but it fits the structure of the deals. ### What changed from a few months ago? The backdrop is the Anthropic fight. Multiple outlets said Anthropic had been blacklisted from classified Pentagon work earlier this year after a dispute over safety guardrails, though officials recently hinted the company could still repair that relationship. So the trust and compliance terms break down. ### What does this mean for the AI market? It pushes the AI stack into clearer trust tiers. One set of tools runs on public cloud. Another runs in enterprise settings. A smaller set gets adapted for sovereign government environments. And now a select group is being wired into classified military networks. That fragmentation is expensive and messy, but it is probably the durable shape of the market for serious government AI. ### So what is the bottom line? The Pentagon is no longer treating commercial AI as a side experiment. It is turning outside models and infrastructure into part of the classified operating environment itself. That is a big shift — and the winners are not just the firms with the best models, but the ones that can survive the security, compliance, and deployment burden of classified work.