AI Investment & Business

Microsoft replaces OpenAI’s image models with its own MAI tech in PowerPoint and Bing

Microsoft is swapping OpenAI's image models for its own MAI tech in PowerPoint and Bing, with model chief Suleyman citing an 85% cost cut.

Microsoft is replacing OpenAI’s image-generating models with its own technology in popular products like PowerPoint and Bing, Bloomberg reported Thursday. “It’s faster, it’s cheaper, it’s higher quality, it drives better retention,” Microsoft AI chief Mustafa Suleyman said in an interview with the outlet.

The swap is the latest step in Microsoft’s methodical campaign to reduce its dependence on its most famous partner. The company has been developing Microsoft AI (MAI) models that can compete with those from OpenAI and Anthropic, in part to drive down the cost of deploying AI across its product line. Suleyman told Bloomberg the MAI models are about 85 percent cheaper to run in PowerPoint, for example.

The economics are striking because Microsoft gets free access to OpenAI models under its longstanding partnership. But Microsoft still must supply the expensive computing infrastructure to run them — meaning more efficient in-house models translate directly into savings, as Bloomberg noted.

MAI Models Spread Across Microsoft’s Product Line

Image generation matters more than it might seem. Customers routinely generate and edit images in Microsoft apps — creating infographics in PowerPoint, for instance — and asking the software to create an image is one of the most common early AI prompts, making a positive first experience commercially important.

The displacement is not limited to images. Microsoft has also begun replacing OpenAI and Anthropic text models inside widely used Office apps, and the company says MAI models in Excel are on par with OpenAI’s GPT-5.6 for common tasks. Meanwhile, companies including T-Mobile US and EasyJet are now using Microsoft’s newest voice-generating models in their call centers, the company announced.

The bottom line, per Suleyman: MAI models now run in more than half of all Microsoft products and are being tested in all of them. “It is a very strong showing, and we’re going to continue to push extremely hard,” he told Bloomberg.

What It Means for the OpenAI Partnership

The shift reframes one of tech’s most consequential alliances. Microsoft has invested tens of billions of dollars in OpenAI and built Copilot on its models — yet it is now systematically demoting those models where its own are cheaper to serve at scale. That mirrors the diversification Microsoft has pursued on infrastructure, including its multi-billion dollar deal with France’s Mistral AI.

For OpenAI, the erosion of default placement inside the world’s most widely deployed productivity suite arrives at an awkward moment, as the company manages fallout from its models’ breach of Hugging Face during a security test and pushes into new enterprise and consumer verticals.

For Microsoft, the calculus is straightforward: every prompt served by a MAI model instead of an OpenAI one cuts inference costs on infrastructure it pays for either way. At the scale of PowerPoint, Excel, and Bing, an 85 percent per-task cost reduction compounds into one of the strongest business cases yet for vertical integration in AI.

FAQ

Why is Microsoft replacing OpenAI’s image models?

Cost and performance. Model chief Mustafa Suleyman says its MAI models are faster, cheaper, and higher quality, and about 85 percent less expensive to run in PowerPoint.

Doesn’t Microsoft get OpenAI’s models for free?

Yes, under the partnership — but Microsoft supplies the costly compute to run them, so more efficient in-house models still produce significant savings.

How widely are MAI models deployed?

Suleyman says MAI models are used in more than half of all Microsoft products and are being tested in all of them, including Excel, where Microsoft says they match GPT-5.6 on common tasks.