TheChek News Flow

The Takeaway

OpenAI on Friday signed a new contract with Amazon Web Services to sell its AI to U.S. government employees for both classified and unclassified work, according to two people with direct knowledge of the agreement.

The contract will allow OpenAI to support the Pentagon through a deal it hastily won late last month after the agency ditched its current AI provider, Anthropic. AWS is already a large cloud provider to numerous U.S. agencies and has agreed to sell OpenAI products to other U.S. government customers, these people said. Landing government contracts could help OpenAI win large corporate customers, which often regard high-profile government work as a sign that a tech provider can be trusted.

Parlaying government contracts into corporate contracts is important to OpenAI because the initial government contracts may not be worth much on their own. For instance, OpenAI’s agreement signed late last month to supply 3 million Defense Department employees with ChatGPT and custom AI products is expected to generate just millions of dollars in revenue over the course of 15 months, said one of the people. That’s a sliver of the $30 billion in revenue the company has projected it will generate this year.

The government-to-corporate contract pipeline has worked well for software firm Palantir Technologies, a longtime military contractor that has increasingly made inroads with businesses and generated about $2 billion in revenue from private-sector customers last year.

OpenAI’s AWS deal comes after months of effort to secure government contracts after the AI firm made a critical change to its agreement requiring it to exclusively host its AI on Microsoft, an early backer. As part of OpenAI’s restructuring into a for-profit corporation last fall, a new agreement with Microsoft enabled OpenAI to work with rival cloud providers like AWS to sell AI to national security customers, such as the Pentagon.

To serve civilian agencies such as the Department of Health and Human Services with unclassified work, OpenAI needs to seek a waiver from Microsoft if it wants to host its AI models on AWS, according to one of the people. For civilian agencies with classified work, such as the Department of State, OpenAI doesn’t need to seek a waiver from Microsoft and can work with AWS directly, said the person.

Upgrade to ask Deep Research to…

Why did both OpenAI and Anthropic miss their recent gross margin forecasts?

What new assets are bankers considering securitizing to fund AI data centers?

Who are the partners in OpenAI’s $500 billion “Stargate” computing consortium?

How much debt could Big Tech firms borrow without losing their credit ratings?

As part of OpenAI’s new deal with AWS, the cloud company’s large sales team will sell OpenAI’s products and services to potential government customers, said the people. OpenAI will share revenue from the government with AWS, they said.

For OpenAI, gaining ground with the government and large businesses has taken on new urgency as Anthropic in recent years has become an important AI provider to the Pentagon, which accesses its technology through Palantir. Anthropic sales to businesses and app developers also have surged in recent months following its launch of tools to automate coding and other white-collar tasks. Anthropic is closing the revenue gap with OpenAI despite the ChatGPT maker’s head start.

The Microsoft Restructuring

Last fall, OpenAI pushed for a carve-out to its exclusivity agreement with Microsoft because many government and security agencies already have extensive contracts with AWS. That makes it difficult for tech vendors to sell to these agencies via cloud providers such as Microsoft Azure, according to three people with knowledge of the negotiations.

Partnering with Amazon could also help OpenAI win contracts with intelligence agencies for classified work more easily than working with other cloud providers such as Microsoft Azure, said John Weiler, CEO of the IT Acquisition Advisory Council, a group funded by Congress that advises federal agencies on how to buy software. AWS can point to existing cybersecurity and data privacy protections that agencies have already approved for some work.

OpenAI’s partnership with AWS for government work is part of a deeper relationship between the two companies. Amazon last month committed to invest up to $50 billion in OpenAI, which committed to spend an additional $100 billion on computing services on top of last fall’s $38 billion multi-year deal.

Since the restructuring, OpenAI staffers have been spending time with government agencies explaining how the company can support their work, according to the people. These staffers include Joseph Larson, vice president of government; Katrina Mulligan, head of national security partnerships; and Felipe Millon, who leads government sales to civilian agencies such as the Department of Health and Human Services.

Separately, OpenAI CEO Sam Altman last week met with leaders in Washington including Arizona Sen. Mark Kelly, Virginia Sen. Mark Warner and Michigan Sen. Elissa Slotkin to discuss the recently signed Pentagon deal and the economic impact of AI, according to one of the people.

Civilian Agency Work

Most of OpenAI’s prior government business came from unclassified work for federal, state and local government agencies and ran on Microsoft Azure, said one of the people. For instance, OpenAI in August made ChatGPT available to all federal workers for just $1 per agency, with Microsoft running those AI queries. Google, Anthropic and xAI made similar deals with the government.

All four companies also struck a one-year contract worth up to $200 million with the Defense Department, announced last summer, to prototype custom AI products for unclassified work.

OpenAI previously did win one contract for classified work: a roughly $20 million deal to use its AI for classified research at national laboratories such as Los Alamos, which planned to run the company’s AI on its local servers rather than through cloud providers, said one of the people. (National labs have long been fans of OpenAI models.)

Then in the fall, the Defense Department sought to renegotiate its contracts with the four major AI model makers, in part to ensure that the agency could use their technologies for “any lawful use.”

While xAI and OpenAI agreed to the change, Anthropic—which unlike the others can run its AI in some classified settings—required that the government would never use its Claude models for mass domestic surveillance or for operating autonomous weapons. Anthropic refused to comply with a February 27 deadline demanding that it drop that requirement.

In response, Hegseth said Anthropic was a “supply chain risk,” meaning contractors working with the Defense Department are required to stop using Anthropic’s tech. That created an opening for OpenAI, which worked on winning a deal to sell its AI models to the Pentagon for classified work.

OpenAI’s deal with the Pentagon has become a flashpoint for the company. Some OpenAI employees voiced their frustrations over the company’s decision to agree to the Pentagon’s terms. In response, Altman took to X to explain the contract. After conceding that the deal “looked opportunistic and sloppy,” OpenAI made modifications to prevent use of its tech in autonomous weapons and domestic surveillance