OpenAI has struck a partnership with Amazon Web Services that lets it deliver its AI models to U.S. defense and civilian agencies for both classified and unclassified work. The tie up builds on OpenAI’s recent agreement with the Pentagon, under which its models will run inside classified Defense Department networks and replace Anthropic after the Trump administration labeled that rival a “supply chain risk” over a dispute on AI use in mass surveillance and autonomous weapons.
Because AWS is already deeply embedded across federal systems and operates GovCloud and classified regions, the alliance gives OpenAI rapid access to agencies it could not easily serve on its own without building certified secure infrastructure. OpenAI says it will still choose which models are offered and can insist on extra safeguards for sensitive deployments, while the Pentagon deal explicitly includes technical and contractual limits meant to keep its models away from domestic mass surveillance and fully autonomous lethal weapons, a stance the company argues also reassures large corporate customers watching how its technology is used by governments.
How OpenAI replaced Anthropic
Anthropic won a contract with the Pentagon for up to $200 million last year, supplying the Claude model for classified military and intelligence systems through Palantir $PLTR and AWS $AMZN. But the relationship broke down this February. Anthropic refused to remove restrictions on select deployments - particularly domestic surveillance and potential use for autonomous weapons - and the Pentagon formally designated the firm a "supply chain risk" after a series of warnings and decided to phase out the collaboration over six months.
Shortly thereafter, OpenAI announced that it had won a contract to supply ChatGPT and other models to roughly 3 million DoD employees for both classified and unclassified work. According to The Information, the contract is expected to bring in "only" a few million dollars over 15 months, a fraction of OpenAI's estimated $30 billion in 2026 revenue. But the strategic value is elsewhere: OpenAI has entered the nation's critical infrastructure where a competitor has been a major player.
Why the AWS deal is key for OpenAI
The partnership with AWS moves OpenAI beyond the Pentagon itself. Amazon's cloud has deep roots in many federal agencies, and AWS has agreed to offer OpenAI products across its government clientele, according to sources. So OpenAI is gaining.
Access to a broad base of agencies through existing AWS contracts.
The ability to deliver AI for both classified and unclassified environments without having to build its own certified infrastructure.
a strong signal of confidence to large enterprise customers who often see government deployments as a test of reliability.
The deal is also made possible by an adjustment in the relationship with Microsoft. Following OpenAI's move to a for-profit structure, the contract was updated in 2025 so that Microsoft would not have a pre-emption right on all compute for OpenAI and so that OpenAI could provide APIs to government and security customers regardless of the cloud. So OpenAI's first products continue to run on Azure, but the company is now free to use other providers - including AWS - for specific segments such as national security.
The impact on Amazon and the "cloud war" on government
For Amazon, the partnership with OpenAI is a welcome counterpunch at a time when Microsoft $MSFT had a strong marketing argument for Azure thanks to its exclusive relationship with OpenAI. AWS can now offer OpenAI models to governments alongside its own services and those of other partners, strengthening its position as an AI "neutral marketplace".
Moreover, the cloud is strategically important to government:
It generates long-term stable revenue from multi-year contracts.
Increases reputational capital with corporate customers.
gives providers deeper insight into specific security and regulatory requirements, which they can then monetize in the commercial sphere.
This gives AWS the argument that it can be a platform for a variety of top models - from custom to Anthropic (where allowed) to OpenAI - and that it can respond quickly to changes in Pentagon or other agency policies.
Risks and open questions
On the Anthropic side, the dispute with the Pentagon may spill over into the legal arena. The company has previously indicated that it will challenge the DoD's actions, and the outcome of the dispute may affect future rules for AI purchases across federal agencies. At the same time, President Donald Trump's decision to have authorities phase out other models to replace Claude within six months creates room for OpenAI, xAI and other competitors that have already tapped into classified networks.
For OpenAI and Amazon, the main risk is political and reputational. The national security segment is lucrative but also sensitive - work on classified projects, questions around surveillance and autonomous weapons, and public debate about where AI has limits in defense. If controversial deployments emerge, pressure to regulate and limit cooperation with some contractors could quickly increase.
Financially, the new contracts do not yet represent a major part of OpenAI's revenue - a Pentagon contract of "a few million dollars" over 15 months is more of a pilot project than a major source of profit at the scale of a company with an expected $30 billion in revenue in 2026. More important is how quickly it can expand AI services to more agencies through AWS and other channels, and what volume of contracts it can pick up after Anthropic's final exit.
What to watch next
In the coming months, the key:
What specific agencies other than the Pentagon will begin using OpenAI over AWS.
How Anthropic's dispute with the Pentagon will play out, and whether it will lead to new rules for AI in government.
whether OpenAI will begin to list larger government contracts as a significant revenue segment, or whether national security will remain a strategic rather than financial pillar.
how Microsoft, Google and others in the government AI solutions space will respond.