Pentagon strikes deals with 7 firms

The Pentagon just made a very specific bet on AI. Not “we’re studying it,” not “we’re running a pilot,” but actual agreements with seven companies to put AI tools inside classified military systems. That matters because the hard part was never getting a chatbot to demo well on an unclassified laptop. The hard part was getting useful models into secret networks, under military rules, with procurement authority behind them. On Friday, May 1, that line moved. ### Which companies got in? The seven firms are OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, SpaceX, and Reflection AI. That mix tells you what the Pentagon wants: not one giant winner, but a bench of model builders, cloud providers, infrastructure suppliers, and defense-adjacent operators that can plug into secure environments without waiting for a single monolithic platform. ### What did the Pentagon actually buy? Not one consumer-facing app. The Pentagon said these companies’ AI capabilities will be available on classified networks for “lawful operational use,” with the goal of speeding data analysis, improving situational awareness, and helping warfighters make decisions in complex environments. In plain English, this is about moving AI from back-office experimentation into operational workflows where timing and classification rules matter. ### Why do “Impact Levels 6 and 7” matter? Because that is the real moat. Impact Level 6 generally maps to secret workloads, and Level 7 is for the most sensitive classified environments. Plenty of AI companies can say they serve government. Far fewer can say their tools are being made available where classified military planning ### Why is Anthropic missing? Because this is not just a tech story. It is also a trust-and-supply-chain story. Anthropic was left out after the Pentagon designated it a supply-chain risk earlier this year, and the administration had already pushed agencies to unwind use of its products, though that effort has been tied up by an injunction classified adoption. ### Why include SpaceX and Reflection? Because the Pentagon is buying capability stacks, not brand familiarity. SpaceX brings secure government relationships and operational infrastructure. Reflection is the outlier name, but that is part of the point — the department is willing to slot in newer AI firms alongside hyperscalers and chip leaders if they can meet the security and deployment bar. This is a portfolio move, basically. ### What changed from last year? The center of gravity shifted from strategy documents to fielding. The Pentagon has spent years talking about responsible AI adoption and building governance through the Chief Digital and AI Office. But internal strategy work for fiscal 2026 points much more directly at accelerating military AI integration through concrete programs, infrastructure, policy, and talent the rhetoric. ### What is the catch? Visibility. Once multiple vendors’ models are inside classified networks, the key questions stop being “can the model answer?” and become “who can audit it, who can switch it off, who owns the logs, and how do commanders compare outputs across tools?” Buying seven lanes at once may reduce dependence on any one company, but it also makes adjacent utility. The seven-company lineup matters. The classified-network access matters more. And Anthropic’s absence makes clear that, for defense AI now, vendor trust is part of the product.