Amazon’s AWS jumps on OpenAI models the day after Microsoft loosens its grip

Amazon moves fast when it sees an opening.

A day after Microsoft agreed to unwind its exclusive rights arrangement with OpenAI, AWS announced it’s bringing OpenAI models to Amazon Bedrock. That includes a new agent service built on top of OpenAI’s latest reasoning models.

Let’s be clear: this isn’t Amazon suddenly becoming an OpenAI fan. This is about market dynamics shifting under everyone’s feet.

The exclusivity that held for years

Microsoft had a tight grip on OpenAI’s commercial cloud distribution. If you wanted to run GPT-4 or o1 through a major cloud provider, Azure was your only option. That was the deal struck back when OpenAI was still figuring out its business model and Microsoft needed assurances to pour in billions.

But those deals don’t last forever. Microsoft and OpenAI reportedly renegotiated, and the exclusivity clause is gone. The details aren’t fully public, but the practical effect is immediate: AWS can now resell OpenAI models just like it resells Anthropic, Meta, and Mistral models.

What AWS is actually offering

Through Amazon Bedrock, customers can now access:

• GPT-4o and GPT-4 Turbo for general purpose work
• o1 and o3 reasoning models for complex tasks
• A new Bedrock Agents integration specifically for OpenAI’s reasoning models

The agent service is the interesting part. Bedrock Agents already supported Anthropic’s Claude and other models, but adding OpenAI’s reasoning models means developers can build autonomous agents that use chain-of-thought reasoning natively. The pricing is standard AWS markup on top of OpenAI’s API rates, so no surprises there.

Why this matters beyond the obvious

This is higher than I expected, honestly. I figured Microsoft would drag its feet on unwinding exclusivity for at least another year. The fact that Amazon had offerings ready to go the very next day tells you these negotiations have been in the works for a while.

For enterprises, this is a win. Multi-cloud AI strategies just got more practical. You can now run OpenAI models on AWS alongside Anthropic on GCP and Llama on your own infra without juggling three different cloud relationships for the same capability.

For AWS, it’s defensive. Amazon has been pushing its own AI chips (Trainium, Inferentia) and has deep partnerships with Anthropic. But customers kept asking for OpenAI on AWS. Ignoring that demand was losing them deals to Azure. Now they can say “yes” and still upsell their own infrastructure underneath.

The catch

Don’t expect deep integration like what Microsoft built with Copilot. AWS is offering the models as-is through Bedrock. No custom fine-tuning pipelines, no exclusive early access, no special pricing. It’s a distribution play, not a partnership.

Also, OpenAI’s latest models are still expensive. Running o1 through Bedrock won’t magically make it cheaper. AWS adds its own margin on top of OpenAI’s per-token pricing. If cost is your concern, this doesn’t help.

What comes next

Google Cloud will probably announce something similar within weeks. Once Microsoft’s exclusivity is truly dead, every cloud provider will want to offer OpenAI models. The model marketplace is becoming a commodity.

The real competition isn’t about who can resell GPT-4o. It’s about who can build the best tooling around it — agents, observability, cost management, security controls. AWS has Bedrock Agents. Microsoft has Copilot Studio. Google has Vertex AI Agent Builder.

This move by Amazon is a signal that the AI cloud wars are entering a new phase. Exclusivity is dying. Distribution is becoming table stakes. The winners will be the ones who make the models actually useful for real workloads, not just the ones who sell access.