Microsoft to tap Anthropic for Office 365 as enterprises weigh risks of AI lock-in

Microsoft is reportedly preparing to integrate Anthropic’s AI models into Office 365, marking a shift from its longstanding reliance on OpenAI technology.

The addition would bring Anthropic’s capabilities into productivity tools such as Word, Excel, Outlook, and PowerPoint, expanding the range of AI options available to customers, according to a report from The Information.

The move reflects Microsoft’s broader strategy of diversifying its AI stack. Alongside its multibillion-dollar investment in OpenAI, the company is building its own models and incorporating offerings from other providers, including DeepSeek, on the Azure cloud.

The report added that developers testing AI features for Office 365 found that Anthropic’s models outperformed OpenAI in certain areas, including automating financial tasks in Excel and creating presentation-ready slides in PowerPoint from user prompts.

To gain access to Anthropic’s technology, Microsoft will reportedly work through Amazon Web Services, which holds a major stake in Anthropic.

The shift underscores a broader move toward multi-vendor strategies in AI, echoing the path enterprises took with cloud computing. The risk of being locked into one vendor is a concern that could influence how companies evaluate and deploy AI in critical applications.

AI choices, costs, and control

Analysts say Microsoft’s plan is notable for enterprises, as Anthropic’s Claude is emerging as a premium offering, particularly in software development.

“LLM API prices are falling generation on generation across most model families of comparable size, but Anthropic has been able to command substantially higher prices for Claude APIs, and especially for Claude Code,” said Alexander Harrowell, principal analyst for advanced computing at Omdia.

Excel already supports programming through formulas, VBA, and, more recently, Python via an external cloud service. Microsoft’s plan may even involve Claude Code generating Python scripts that are then executed in Excel’s Python environment, Harrowell added.

For enterprises, the larger implication is choice. By blending multiple models, Microsoft reduces reliance on a single provider while giving customers the ability to match tools to specific tasks.

“For enterprises, this confirms they need not depend on a single model,” said Pareekh Jain, CEO at Pareekh Consulting. “Just like they embraced multi-cloud strategies to avoid lock-in, they will increasingly adopt multi-model AI strategies. Competitors like Google and Oracle may also take this route, competing in some layers while cooperating in others.”

AI supply chains across rivals

Microsoft will reportedly pay AWS to access Anthropic’s models, a move that may be invisible to end users but carries implications for cost and infrastructure strategy.

“The downside, of course, is the margin stacking that results,” Harrowell said. “AWS is not the cheapest LLM API provider, and their margin is layered on top of Anthropic’s. Microsoft will want to bring it on-platform as soon as they understand it and have the capacity. They seem to be buying capacity in every direction at the moment, with the deal with Nebius possibly reflecting delays in the Maia AI-ASICs.”

This means that delays in Microsoft’s Maia AI-ASICs, which underpin Azure’s AI capacity, may be forcing the company to rely on AWS to run Anthropic’s models.

However, this kind of cooperation between competitors is not unusual for the tech sector, according to Sharath Srinivasamurthy, research vice president at IDC. For instance, Apple sources display panels from Samsung despite being direct competitors in the smartphone market.

“Cross-license patents, sharing of supply chains, and integrating each other’s services are common in the industry,” Srinivasamurthy said. “These examples [of the likes of Apple and Samsung] highlight how competitive interdependencies have long existed in the tech industry, supported by well-defined commercial and legal frameworks.”

Such arrangements underline a reality that carries direct implications for enterprises. “For CIOs, this signals both a hedge against single-partner risk and an acknowledgment that no one provider can meet the full spectrum of enterprise AI needs,” said Ishi Thakur, analyst at Everest Group.