{"slug": "microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own", "title": "Microsoft is replacing OpenAI's image models in PowerPoint and Bing with its own MAI-Image-2", "summary": "Microsoft is replacing OpenAI's image generation models in PowerPoint and Bing with its own MAI-Image-2, the clearest sign yet that its $13 billion investment doesn't guarantee a permanent product placement. The MAI-Image-2, trained on Azure with thousands of NVIDIA H100 GPUs and approximately 100 billion parameters, ranks third on the Arena.ai image leaderboard, while its successor MAI-Image-2.5 sits at number two on the image editing benchmark. Microsoft has already begun routing prompts in Excel and Outlook away from OpenAI and Anthropic models toward its MAI lineup, with tens of thousands of weekly prompts now handled by its own models.", "body_md": "*Bloomberg confirmed on July 23 that Microsoft is swapping OpenAI's image generation models out of PowerPoint and Bing, deploying its own MAI-Image-2 instead , the clearest sign yet that a $13 billion investment doesn't buy you a permanent seat at the product table.*\n\nThe rollout is real and it's moving fast. MAI-Image-2, which Microsoft trained on Azure infrastructure using thousands of NVIDIA H100 GPUs and approximately 100 billion parameters, now ranks third on the Arena.ai image leaderboard. Its successor, MAI-Image-2.5, launched in May and sits at number two on Arena's image editing benchmark. Both are being pushed into Bing, PowerPoint, and Copilot , products where OpenAI's models had been doing the work. Microsoft declined to specify what share of image generation traffic has already migrated.\n\nThis isn't the first swap. Bloomberg reported on July 7 that Microsoft had already begun routing prompts in Excel and Outlook away from OpenAI and Anthropic models toward its own MAI lineup. Tens of thousands of weekly prompts in those apps are now being completed by Microsoft-built models. The full MAI family, unveiled at Build 2026, spans seven models: MAI-Thinking-1 for reasoning, MAI-Code-1-Flash for coding, MAI-Transcribe-1.5 for transcription, MAI-Voice-2 for speech, and the image pair MAI-Image-2.5 and MAI-Image-2.5-Flash. That's a complete in-house stack covering nearly every major Copilot use case.\n\nEvery token Microsoft routes to its own models is a token it doesn't pay OpenAI for. That's the direct cost structure of the business. When Microsoft sells a Microsoft 365 Copilot subscription and the underlying inference runs on OpenAI's GPT-5 or Anthropic's Claude, a portion of that revenue flows out as third-party inference cost. Swap in a MAI model running on Azure, and that cost stays inside the house. Microsoft has said one of its MAI models, tuned for McKinsey, beat GPT-5.5 on cost efficiency by a factor of ten. You don't need to believe that figure precisely to understand the direction it points.\n\nThe partnership itself has been quietly restructured. CNBC reported in late April that OpenAI and Microsoft renegotiated their agreement, with OpenAI gaining the right to serve customers across any cloud provider and Microsoft no longer paying a revenue share to OpenAI. That's a meaningful change in the commercial relationship , both companies stepping toward independence while maintaining the public posture of a unified platform. OpenAI's response, per TechCrunch, was to promote GPT-5.6 as the \"preferred model\" for Microsoft 365 Copilot. That framing is doing a lot of work for a company that just watched its partner announce seven competing models at its biggest annual developer conference.\n\n## What this means for startups built on a single model provider\n\nOpenAI still handles the majority of Copilot production traffic. Satya Nadella and CTO Kevin Scott have both said GPT models remain central to the Azure AI platform, and GPT-5 fine-tuning and inference features are still coming to Azure. So this isn't a breakup announcement. But the direction is clear, and the lesson for anyone building on top of a single AI provider is uncomfortable.\n\nMicrosoft is the most obvious case because its OpenAI relationship is the most visible and the investment was the largest, but the dynamic isn't unique to them. When a company big enough to build its own models also happens to be paying per-token to a partner, the incentive to eventually in-house is almost impossible to resist. The inflection point is always the same: once in-house quality reaches competitive parity , which the Arena leaderboard suggests MAI has , the margin argument becomes overwhelming. Speed helps too. Microsoft says MAI-Image-2.5-Flash is designed for fast, scalable production workloads, which in plain terms means it's cheaper and quicker per generation than the alternatives.\n\nFor AI startups that sell model access, the riskier version of this story isn't Microsoft replacing OpenAI. It's any enterprise customer that reaches sufficient scale and technical maturity to run the same calculation. The build-vs-buy reckoning doesn't require a $13 billion relationship to matter. A company processing enough AI queries eventually asks whether the unit economics of API dependency still make sense. Some will build. Most won't have to, because the hyperscalers , Microsoft, Google, Amazon , are busy commoditizing the layer beneath them regardless.\n\nOpenAI isn't going away from Copilot tomorrow. But the product bets Microsoft is making now, in PowerPoint slides and Bing image searches and Outlook drafts, are accumulating into something real. The $13 billion bought access and a head start. It didn't buy permanence.\n\n**Also read:** [Nvidia's Jetson chips are becoming the default AI layer for Moon missions](https://startupfortune.com/nvidias-jetson-chips-are-becoming-the-default-ai-layer-for-moon-missions/) • [Black Forest Labs collapses image, video, and robot control into a single model with FLUX 3](https://startupfortune.com/black-forest-labs-collapses-image-video-and-robot-control-into-a-single-model-with-flux-3/) • [Etched is raising at $10 billion and $20 billion simultaneously and investors are lined up](https://startupfortune.com/etched-is-raising-at-10-billion-and-20-billion-simultaneously-and-investors-are-lined-up/)", "url": "https://wpnews.pro/news/microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own", "canonical_source": "https://startupfortune.com/microsoft-is-replacing-openais-image-models-in-powerpoint-and-bing-with-its-own-mai-image-2/", "published_at": "2026-07-23 17:09:30+00:00", "updated_at": "2026-07-23 17:12:41.975128+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-infrastructure", "ai-startups"], "entities": ["Microsoft", "OpenAI", "MAI-Image-2", "NVIDIA H100", "Arena.ai", "MAI-Image-2.5", "Azure", "Copilot"], "alternates": {"html": "https://wpnews.pro/news/microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own", "markdown": "https://wpnews.pro/news/microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own.md", "text": "https://wpnews.pro/news/microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own.txt", "jsonld": "https://wpnews.pro/news/microsoft-is-replacing-openai-s-image-models-in-powerpoint-and-bing-with-its-own.jsonld"}}