Amazon jumps on OpenAI’s latest, now that Microsoft’s exclusive grip is gone

A day after OpenAI got Microsoft to agree to end exclusive rights, AWS announced a slate of OpenAI model offerings, including a new agent service.

Let me be blunt: this is a big deal. For years, if you wanted to run OpenAI’s models in production on a major cloud provider, your options were essentially Azure or nothing. Microsoft had that exclusivity locked down tight, and it made sense—they invested billions. But that era just ended, and Amazon didn’t waste a second.

The timing is almost comically fast. One day after the Microsoft-OpenAI exclusivity agreement was dissolved, AWS is already listing new OpenAI products on its platform. That’s not something you spin up overnight; Amazon clearly had this in the pipeline, waiting for the legal green light.

So what exactly is AWS offering? The big headline is a new agent service. If you’ve been following the AI agent trend—everyone from Google to Anthropic to Salesforce is pushing their version—you know this is where the industry is heading. These aren’t just chatbots that answer questions; they’re systems that can take actions, run workflows, and integrate with other tools. Amazon’s new service lets you deploy OpenAI-powered agents within the AWS ecosystem, which means you can hook them into S3, Lambda, DynamoDB, and the rest of the AWS stack without jumping through hoops.

They’re also offering the latest OpenAI models directly through Amazon Bedrock, their managed service for foundation models. That includes GPT-4o and the newer reasoning models. For companies already deep in AWS, this removes a major friction point. You no longer need to maintain a separate Azure subscription just to access OpenAI’s best stuff. One cloud, one bill, one set of IAM policies.

Now, I have mixed feelings about this. On one hand, more competition in the cloud AI space is good for everyone. It drives prices down and gives enterprises flexibility. On the other hand, I worry about vendor lock-in shifting from Microsoft to Amazon. If you build your entire agent workflow on AWS-native OpenAI services, migrating later becomes painful. But that’s cloud computing in 2026—you’re always picking your poison.

What I find interesting is that Amazon isn’t just offering the models; they’re offering orchestration around them. The agent service includes built-in guardrails, monitoring, and cost controls. That’s smart. Raw model access is commoditizing fast—the real value is in the surrounding infrastructure that makes it safe and practical to use at scale.

Pricing details are still a bit murky. AWS says the agent service will be pay-per-use, with additional charges for model inference and any downstream AWS resources the agent consumes. Depending on your use case, that could get expensive quickly. An agent that makes frequent API calls to S3 or runs Lambda functions for every step will rack up costs beyond just the model tokens. Keep an eye on that.

One thing that surprises me is how quiet this launch has been. No massive press conference, no splashy blog post from Andy Jassy. Just a quiet update to the Bedrock console and some documentation changes. Maybe Amazon is trying to avoid making Microsoft look bad, or maybe they’re just being pragmatic—get the product out, iterate fast, market later.

For developers, this is mostly good news. If you’re building on AWS and have been eyeing OpenAI’s agent capabilities but didn’t want to deal with cross-cloud complexity, your life just got easier. The agent service supports both OpenAI and Anthropic models, so you’re not forced into one ecosystem. That’s a nice touch.

Still, I’d caution against jumping in blindly. The agent space is evolving rapidly, and we’re still figuring out best practices around reliability, security, and cost. AWS’s offering is polished, but it’s also new. Expect some rough edges, especially around complex multi-step workflows.

Overall, this move signals that the cloud AI wars are heating up in a serious way. Microsoft had a head start with OpenAI exclusivity, but now that’s gone, and everyone is racing to offer the most compelling platform. Amazon’s bet is clear: they don’t need exclusive access to win—they just need to make it easier and cheaper to use the best models on their infrastructure. And with this launch, they’ve done exactly that.