SAN FRANCISCO, March 19, 2026, 02:23 PDT
OpenAI has signed a deal to sell access to its AI models to U.S. defense and other government agencies through Amazon Web Services for classified and unclassified work, widening the ChatGPT maker’s move into sensitive federal systems. The push has also opened a new line of tension with Microsoft after the Financial Times reported that the software giant is considering legal action over a broader Amazon-OpenAI cloud arrangement. 1
The timing matters. Government contracts are turning into a proving ground for AI firms and a trust signal for big corporate buyers, and Reuters said the AWS channel could help OpenAI land customers that view federal work as proof a vendor can handle sensitive jobs. OpenAI had largely stayed on the unclassified side of government work until now, and Reuters said last fall’s revised Microsoft terms allowed partnerships with other cloud providers for national-security customers, helping clear the way for AWS even as broader commercial questions stay unresolved. 1
The government sale sits inside a much larger Amazon tie-up announced in February. Amazon and OpenAI said AWS would become the exclusive third-party distribution provider for Frontier, OpenAI’s platform for AI agents — software that can take multistep actions across business systems — while the two sides build a “Stateful Runtime Environment,” meaning the system can keep memory, identity and context across longer tasks instead of starting fresh each time. Amazon also said it would invest $50 billion in OpenAI, while OpenAI committed to roughly 2 gigawatts of AWS Trainium capacity over eight years. 2
Publicly, OpenAI and Microsoft had tried to calm those questions. In a joint statement on Feb. 27, the companies said Azure would remain the exclusive cloud for what they called “stateless” OpenAI APIs, while OpenAI’s first-party Frontier product would continue to run on Azure; Reuters, citing the FT, said Microsoft now believes the Amazon structure may violate the spirit, if not the letter, of that pact. 3
Sam Altman cast the Amazon alliance as a way to put OpenAI’s systems to work at “real scale.” Amazon CEO Andy Jassy said AWS had “lots of developers and companies eager” to run OpenAI models on its cloud, a sign that model makers and cloud providers are increasingly selling a full stack, not just raw compute or a standalone chatbot. 2
Anthropic is the clearest competitive backdrop. Reuters reported the Pentagon moved away from Anthropic after a dispute over unrestricted military use, but OpenAI later wrote that it had asked Washington to make the same terms available to other labs and that Anthropic should not be treated as a supply-chain risk. 1
OpenAI says its own Pentagon arrangement comes with tighter guardrails. The company said the contract bars domestic surveillance of U.S. persons, the use of its tools to direct autonomous weapons systems and other high-stakes automated decisions, and that OpenAI will keep its safety systems and cleared personnel involved in the deployment. 4
But this could still get messy. Microsoft’s stance could slow the AWS rollout or force a narrower reading of who can host what, and defense work will keep drawing scrutiny over whether written safeguards hold once the tools move into live federal systems. The question now is less whether OpenAI can sell into government at scale than how far it can stretch beyond Azure without damaging the partnership that helped finance its rise. 5
