Microsoft Ditches OpenAI and Anthropic for Its Own AI Models in Key Office Apps

Microsoft has begun replacing OpenAI and Anthropic with its own in-house AI models in core productivity applications, according to Bloomberg. The company is deploying its homemade MAI models in Excel and Outlook, marking a significant break from its previous strategy of leaning heavily on external AI partners. This shift represents one of the most concrete examples of a major tech giant reversing course on third-party AI dependency.

The decision follows months of escalating AI infrastructure costs that have prompted Silicon Valley's largest companies to reevaluate spending. Microsoft had previously invested more than $13 billion in OpenAI and served as the ChatGPT maker's primary cloud partner. Now, the company appears to be prioritizing cost control over maintaining exclusive relationships with the industry's most prominent AI labs. The move also comes after Microsoft restricted internal access to Anthropic's Claude Code tool in mid-May 2026, redirecting engineers toward GitHub Copilot CLI instead.

Microsoft's new approach involves blending multiple AI providers rather than relying on a single partner. According to AI Magazine, the company will pay to use Anthropic's Claude models for some Office 365 features alongside OpenAI's technology, even as it builds out its own capabilities. This multi-vendor strategy contrasts sharply with the years Microsoft spent primarily using OpenAI models across Word, Excel, Outlook, and PowerPoint.

Ars Technica reports that Anthropic's models won placement in Office after outperforming OpenAI at certain tasks, suggesting the shift is driven by performance considerations as well as cost. The change ends OpenAI's exclusive hold on Microsoft's productivity suite, though the exact scope of the replacement remains unclear. Microsoft has not specified which featuresoit features will run on which models, or how quickly its own MAI models will expand beyond Excel and Outlook.

The cost pressures driving Microsoft's pivot are industry-wide. TechCrunch notes that AI costs continue to rise, pushing companies to find ways to cut back on expensive third-party model access. Microsoft's move follows a broader pattern of tech giants reassessing their AI spending as the initial deployment phase gives way to a focus on sustainable economics.

Forbes frames the development as Microsoft building its own AI stack to reduce dependence on external providers. This vertical integration strategy mirrors moves by other major cloud providers, though Microsoft is unique in having both massive AI investments and the infrastructure to replace partners at scale. The company has spent years developing its AI capabilities through Microsoft Research and acquisitions, and now appears ready to deploy that intellectual property directly in its most widely used products.

Microsoft's internal pullback from Claude Code highlights another front in this shift. According to Memeburn, Microsoft began canceling or restricting Claude Code licenses for engineering teams in mid-May 2026, with the Experiences and Devices division, which oversees Windows, Microsoft 365, Outlook, Teams, and Surface, most affected. These teams had been testing Claude Code at scale earlier in the year.

CNBC reports that coding tools have become an increasingly important battleground, with Anthropic's Claude Code establishing an early lead and OpenAI shifting focus to enterprise offerings like Codex. Microsoft and Google are now racing to compete in this space, making their own coding assistants a strategic priority. By redirecting engineers toward GitHub Copilot CLI, Microsoft is both cutting costs and keeping talent within its own ecosystem, even if that means accepting temporary capability gaps.

Microsoft's evolving supplier relationships will likely reshape how enterprise AI is procured and deployed. The company's ability to switch between OpenAI, Anthropic, and its own models gives it unusual leverage in negotiations and flexibility to respond to rapid advances from any provider. This multi-model architecture may become the standard for large enterprises that cannot afford to be locked into single vendors.

However, the transition carries risks. Microsoft's own models may not yet match OpenAI or Anthropic on all tasks, and customers could experience inconsistent quality as the company fine-tunes its routing between providers. The long-term question is whether Microsoft will eventually eliminate third-party AI spending entirely, or maintain a portfolio approach indefinitely. For now, the company appears to be pursuing both paths simultaneously: building self-sufficiency while keeping external options open.

The shift has immediate implications for how AI features are delivered in software that hundreds of millions of people use daily. If Microsoft's MAI models perform well in Excel and Outlook, the company will likely expand their use across the entire Microsoft 365 suite, potentially changing the competitive dynamics between AI providers for enterprise contracts.

For developers building on Microsoft's platforms, the multi-model approach means designing applications that can work with different underlying AI capabilities. For enterprise buyers, it signals that vendor lock-in in AI may be less inevitable than once assumed, even for companies that have made massive strategic investments in particular partners. Microsoft's $13 billion OpenAI stake once seemed to guarantee a permanent alliance; now it looks more like one option among several.