Microsoft stops pretending. It offers companies its own models against OpenAI

The end of exclusivity, the start of a tough fight

Microsoft sent a clear signal while announcing its financial results for the past quarter. Until now, the company has acted primarily as the biggest reseller of models from OpenAI, in which it has invested billions. But now, CEO Satya Nadella has explicitly warned corporate customers against relying on a single AI model supplier (the so-called frontier labs) and offered them a solution: models Microsoft developed itself.

The company introduced more than a dozen new proprietary models from the MAI family. This includes the first dedicated reasoning model, MAI thinking one, and the small but powerful MAI Cyber One Flash, which competes directly against Anthropic's Mythos. Microsoft has optimized all these models for its own AI chips (Maya 200), promising a significantly better price-to-performance ratio.

Corporations want control, and Microsoft promises it

For enterprise IT departments, the way they should approach the architecture of their AI systems is changing. Nadella is pushing the concept of separating the application layer (harness) from the model itself. Companies shouldn't build applications tightly coupled to OpenAI's API, but rather use Microsoft's tools (like Copilot), under which they can swap models at any time.

Microsoft has hit exactly on corporations' sensitive spots here: fear of vendor lock-in and data leaks. By offering a huge catalog of models (over 11,000), including open-source ones, and advising customers to alternate them, Microsoft weakens the negotiating position of the labs its own AI business has relied on so far.

Playing both sides meets reality

This change of course comes at a moment when OpenAI and Anthropic are beginning to actively bypass cloud middlemen and sell their products and agents directly to companies. Microsoft realized it's in danger of losing the direct relationship with the customer. It's the classic partnership trap: you sell someone else's product for so long until the customer realizes they don't actually need you to buy it.

Nadella also backed his warning against relying on one model with the recent Hugging Face incident, where, according to him, one of the top models failed to defend the infrastructure, while an open-source model worked. That's strong argumentative ammunition against blind trust in systems like GPT-5.

Earnings will show whose models companies really want

The proof of whether this offensive is working will be the breakdown of revenue in Azure. We'll need to watch how much of the reported AI growth will actually be attributed to its own MAI models and how much is still just reselling OpenAI. If Microsoft can convince clients to switch to the cheaper MAI, it will keep higher margins and limit Sam Altman's influence on its cloud business.

Lilith's verdict

Nadella took the gloves off. He is openly telling corporations not to trust his own partners, offering them a table full of his own cheaper copies. Microsoft is returning to what it does best: commoditizing technology before the partner gains power over it.