Microsoft is removing OpenAI’s image-generation models from two of its most prominent consumer products, PowerPoint and Bing, and replacing them with its own MAI-Image technology.

Microsoft AI lead, Mustafa Suleyman, said the decision was driven by the in-house model’s performance, describing it as “faster,” “cheaper” and “higher quality,” while also improving user retention, according to Bloomberg.

The cost savings in PowerPoint underscore the strategic rationale. Microsoft’s own image model reportedly reduced costs by 84 per cent compared with OpenAI’s GPT-Image-2 on the platform.

For an enterprise product operating at the scale of PowerPoint, an 84 per cent reduction in unit costs can significantly improve Copilot’s gross margins.

The savings come from shifting a feature previously powered through a partner’s API to Microsoft’s own model, without requiring any change to the price customers pay.

The move also fits a broader pattern. Following Microsoft’s decision in July to route Excel and Outlook prompts through its MAI models instead of OpenAI and Anthropic, the company is now extending the same strategy to image generation.

High-volume, routine workloads are increasingly being moved onto Microsoft’s own AI infrastructure, while more advanced frontier tasks presumably continue to be handled by external models.

That does not signal a breakdown in Microsoft’s relationship with OpenAI. Instead, it shows where Microsoft has the greatest control: over the meter, and therefore the cost.