OpenAI just signed a deal with AWS to sell AI to the US government. Not Azure. AWS. The company that Microsoft poured billions into just routed its government sales pipeline through Microsoft's biggest cloud competitor. That is a wild move and I have been thinking about it all morning.
The news broke on X and the thing that caught me first was not the AWS angle. It was the other half of the story: the government picked OpenAI over Anthropic.
That second part deserves more attention than it is getting.
The AWS thing is interesting but probably tactical
Let me get the Microsoft piece out of the way first because it is the flashier headline.
OpenAI and Microsoft have one of the most important partnerships in recent tech history. Microsoft bet something like $13 billion on OpenAI. Azure is deeply integrated into OpenAI's infrastructure. So why is OpenAI going to AWS to chase government contracts?
My read: this is pure distribution math. AWS owns somewhere around 30-32% of cloud market share. More importantly, AWS has deep, established roots in federal government through GovCloud. They have FedRAMP authorizations, existing agency relationships, and a sales team that has spent years learning how to sell to procurement officers and contracting vehicles. Microsoft has similar infrastructure but the relationships in certain agencies just run deeper on the AWS side.
OpenAI is not betraying Microsoft here. They are being pragmatic. If the goal is to win as many government contracts as possible, you work with whoever has the best existing foot in the door. Microsoft is still a massive partner for commercial. This is a specific go-to-market move for a specific channel.
I have seen this kind of thing play out in tech companies before. Partnerships are not monogamous when the market is big enough. OpenAI is acting like a company that wants to win, not a company that feels sentimental about its cap table.
Why OpenAI over Anthropic though
This is the part I keep coming back to.
Anthropic is not some scrappy nobody. They have serious safety research credibility, they are well funded, and Claude is genuinely competitive with GPT-4 class models on a lot of benchmarks. Anthropic also has Amazon as a major investor, which makes the AWS angle even more interesting because you would think Amazon would push its own portfolio company.
Yet the government went with OpenAI.
I have a few theories and I go back and forth on which one I believe most.
Brand recognition is underrated in procurement. Government buyers are not always the most technically sophisticated evaluators. When a contracting officer or a CIO at a federal agency is making a recommendation up the chain, they want to attach their name to something that everyone has heard of. ChatGPT is a household name. Claude is not, at least not yet. In a procurement process where career risk is real, picking OpenAI feels safer. That matters more than people admit.
OpenAI has been playing the DC game harder. Sam Altman has been making the rounds in Washington for a couple of years now. Testifying, meeting, schmoozing. There is a version of this deal that was won in those meetings long before any RFP was written. Relationships built over time, government affairs teams doing the slow work, getting in the room before the procurement even started. Anthropic has smart people but I am not sure they have been as aggressive about cultivating those relationships at the same pace.
Model performance on specific use cases probably matters too. Government AI use cases are often things like document summarization, translation, intelligence analysis support, and code generation for legacy systems. OpenAI has had more time to optimize for enterprise use cases and probably has more reference customers who can vouch for performance at scale in messy real-world environments.
My honest guess is that it is all three of these things combined. Brand, relationships, and a track record that is easier to point to.
What this means if you are building in this space
I am an engineer and a builder, so I always end up asking what the second-order effects are for people actually shipping products.
The government AI market is enormous. We are talking about a procurement pipeline that runs into the hundreds of billions over the next decade when you factor in defense, intelligence, civilian agencies, and state and local. OpenAI just positioned itself to capture a huge chunk of that.
That has real implications for the model market. When a platform gets entrenched in government, it creates lock-in that is incredibly hard to dislodge. Developers build on it. Procurement vehicles get structured around it. Training happens on it. It becomes the default not because it is objectively the best but because switching costs compound over time.
For Anthropic, this is a meaningful setback. Not fatal, but meaningful. They will need to find a different wedge into government, probably through specific agencies or classified programs where safety-focused differentiation actually matters to the buyer.
For builders outside the big model companies, this is a signal. The consumer AI market has OpenAI, Anthropic, Google, and Meta all fighting hard. The government market is starting to consolidate faster than people expected, and OpenAI is winning that consolidation race right now.
The bigger picture
OpenAI is doing what every great enterprise software company does when it hits a certain scale. It is building distribution moats. The product moat matters but the distribution moat is often what actually wins markets.
AWS was a smart pick because of what AWS already has inside those agency firewalls. Government chose OpenAI over Anthropic because of what OpenAI already has inside those decision-maker relationships.
Distribution won this deal. Not benchmarks.
That is the lesson worth carrying forward.