What Microsoft replacing OpenAI with its MAI models (up to 89% cheaper) means — July 24, 2026
· CompaniesAutomation
Microsoft has begun replacing OpenAI within its own products —Bing, PowerPoint, Dynamics 365— with its MAI model family, achieving cost reductions of up to 89%. What this means for your business and what to do about it. Flash edition for July 24, 2026.
Flash edition of the Enterprise AI Radar: building on yesterday's edition, today one single story commands the headlines. Microsoft has begun replacing OpenAI within its own products with models it has built in-house. It introduced its MAI family — MAI-Image-2.5 for images, MAI-Voice-2-Flash for voice, and MAI-Transcribe-1.5 for transcription — and claims GPU cost reductions of up to 89% compared to OpenAI in the Dynamics 365 contact center, and up to 84% in PowerPoint compared to GPT-Image-2. Bing Image Creator already runs entirely on its own model, and its AI chief, Mustafa Suleyman, says MAI models are already running in more than half of Microsoft's products. Source
Why it matters
Microsoft is OpenAI's largest investor, and yet, it is building its own alternative to avoid dependency for high-volume tasks — voice, image, and transcription — where inference costs eat into margins. Clients like T-Mobile and EasyJet are already operating on this proprietary voice for customer service. The underlying reading is not "Microsoft is breaking up with OpenAI," but something more useful for you: the leading model is no longer the differentiator. When there are several models capable of doing the same job, the price plummets and the decision shifts from "which is the best?" to "which is the cheapest for this specific task?".
For your company
Three actionable items. First: do not marry your automation to a single model provider; set up an abstraction layer — a router or a gateway — that lets you switch models without rewriting the flow. Second: separate tasks. For simple, high-volume work — classifying tickets, extracting data, transcribing, generating placeholder images — get quotes for cheap or proprietary models and reserve the expensive model only for what truly needs it. Third: if you already pay for AI within Microsoft 365, Copilot, or Dynamics, this change reaches you without lifting a finger — cheaper models underneath — but it is also your leverage to renegotiate: ask your providers which model is underneath and what happens to your bill now that inference costs a fraction.
Frequently Asked Questions
Is Microsoft stopping the use of OpenAI?
Not entirely. It remains its partner and investor, and GPT is kept for the most demanding tasks; but Microsoft is already replacing OpenAI in high-volume jobs — image, voice, transcription — with its MAI models to lower costs and reduce dependency on a third party.
Does this make the AI my company pays for cheaper?
It is the signal that inference is becoming cheaper fast. If you buy AI as a service, you have room to re-quote or renegotiate; if you build it yourself, testing cheaper models for simple tasks can cut the bill without losing quality.
Do I need to change anything today?
Nothing urgent. But it is advisable to review where you are tied to a single model and prepare a layer that lets you change providers when price or quality demands it.