Microsoft and OpenAI tweak their mega-deal through 2030

Despite months of whispers that Microsoft and OpenAI were drifting apart, the two tech giants just locked arms until the end of the decade.

Microsoft officially announced the next evolution of its partnership with OpenAI, confirming their joint work on the Stargate supercomputing initiative and laying out the terms that will govern their relationship through 2030.

The core deal stays locked through 2030

The fundamental pillars established back in 2019 remain intact. Microsoft keeps broad access to OpenAI intellectual property, spanning both frontier models and infrastructure, to power its Copilot lineup.

Azure also retains its exclusive grip on OpenAI APIs. The models run on Azure, and access remains available directly through OpenAI and via Azure OpenAI Service. The two companies will also keep their bidirectional revenue-sharing model, ensuring both balance sheets benefit as usage climbs, while Microsoft maintains its stake as OpenAI primary backer.

On top of that, OpenAI just inked a massive new Azure commitment to cover ongoing operations and future model training.

A crack in the compute exclusivity

The biggest pivot in this update comes down to compute capacity.

Under the refreshed terms, exclusivity on new infrastructure is shifting to a Right of First Refusal framework. Microsoft still gets the first shot at providing fresh capacity, but it is no longer an absolute lock.

More importantly, Microsoft has officially greenlit OpenAI plans to build additional capacity on its own. There is a catch, though: Microsoft noted that this independent capacity is designated primarily for research and training rather than commercial serving.

Why it matters

We think Microsoft played this renegotiation smartly, protecting its core commercial moats while avoiding an unsustainable infrastructure burden.

For enterprise customers and developers, the immediate takeaway is business as usual. You do not have to worry about Copilot losing top-tier OpenAI models or Azure APIs suddenly getting deprecated. The commercial bedrock is secured through 2030.

For the broader industry, the real story is compute flexibility. Training next-generation frontier models is becoming wildly expensive. By moving to a Right of First Refusal and letting OpenAI build its own training clusters, Microsoft avoids having to bankroll every single server OpenAI demands, while still keeping other hyperscalers from swooping in easily.

OpenAI gets the freedom to scale research hardware on its own terms. But make no mistake: because commercial API traffic remains tied to Azure, OpenAI is still deeply anchored to Microsoft infrastructure where monetization matters most.