Microsoft and OpenAI: exclusivity is over, vendor lock-in remains

When two giants agree on a "strategic partnership," an engineer should immediately look for where the lock-in is hidden. The news that Microsoft and OpenAI have revised their relationship — Azure is no longer the exclusive cloud for ChatGPT — at first glance looks like a loosening of the grip. But if you look at the details, it's more about shifting risks than real freedom for OpenAI.

Previously, Microsoft paid OpenAI a revenue share and got exclusive access to the model on Azure. Now the scheme has changed: money flows only one way — from OpenAI to Microsoft for cloud resources. Microsoft retains first-refusal — the right of first choice to provide server capacity, and, of course, access to the model itself. OpenAI can work with other clouds, but only if Microsoft declines. Details of the agreement are disclosed in the original Tom's Hardware article.

Round-trip instead of investments

From a financial perspective, this deal resembles a round-trip: Microsoft "invests" in OpenAI, OpenAI immediately spends that money on Azure. In the reports of both parties, this looks nice — Microsoft shows cloud business growth, OpenAI shows capital raising. But for those building products on top of OpenAI, something else matters more: the model remains tied to Microsoft's infrastructure. Even the formal right to switch to another provider is negated by the fact that moving means weeks of migration, rewriting integrations, and potential performance issues.

What changes for those building AI products

For an engineer choosing a stack, this news is not about "OpenAI becoming more independent." It's a signal: if your product is tied to one model and one cloud, you're in the risk zone. At WIZICO, we've seen more than once how startups that built everything on ChatGPT + Azure ended up in a situation where switching providers costs more than tolerating rising prices. Now OpenAI formally has an option, but in practice Microsoft still controls the key resource — compute.

Our advice: design your system so that the model layer is abstracted from the infrastructure. Use universal formats (ONNX, OpenAPI), plan for fallback to other LLMs (via AWS Bedrock or GCP Vertex AI, for example), and don't hardcode Azure-specific features into the code. This isn't paranoia — it's common sense: if a partnership contract gets rewritten every six months, your architecture should survive such changes without downtime.

Ultimately, the loosening of exclusivity is a step toward a more competitive market, but not today. As long as compute is a scarce resource and models require thousands of GPUs, cloud giants will remain the main beneficiaries of the AI boom. And choosing a provider is not just about price, but also about how easily you can leave.