Microsoft on track to pay Anthropic $2B a year for Copilot

The Information reported on October 12, citing one source, that Microsoft is on track to pay Anthropic more than $2 billion a year for models used in Copilot. Office chief Charles Lamanna reportedly persuaded Satya Nadella despite internal cost concerns. Neither company confirmed the figure.

Anthropic's newer models reportedly performed better on advanced Excel and PowerPoint tasks. Microsoft now blends Claude with OpenAI models across Word, Excel, Outlook and PowerPoint, a major shift from a Copilot strategy built almost entirely on OpenAI.

In January, Microsoft's Anthropic spending was put at about $500 million a year. A separate October 6 report said Microsoft cut projected internal Claude spending by more than a third, though customer-facing Copilot use reportedly offset much of that. The report does not say whether Microsoft buys API usage or dedicated capacity.

Model routing sends each task to the model that does it best. An API is the connection software uses to call another company's service, usually billed by usage.

Copilot sits inside the world's most widely used office software, so Microsoft's model choice shapes what millions of workers use daily. Even with its OpenAI stake, Microsoft is paying a rival lab billions, a sign that single-vendor AI strategies are fading.

Microsoft has invested tens of billions of dollars in OpenAI since 2019 and long built all of Copilot on its models. The relationship has grown more distant: OpenAI signed deals with other clouds, while Microsoft built its own models and began working with other labs. The size of the Anthropic payments shows that diversification has moved from experiments into the core product.

Do not lock into one AI tool; keep data and workflows portable. Pick models by task and run small comparisons, try AI on spreadsheet and presentation work where models have clearly improved, and track cost by usage rather than seat count.