Microsoft weighs Anthropic for Copilot; $2B+ annual spend reported
Microsoft's Charles Lamanna pushed Anthropic models for Copilot despite cost concerns; a source says Microsoft is on track to pay Anthropic $2B+ annually.

Microsoft is weighing Anthropic and OpenAI models to improve Copilot's capabilities and margins, and Office chief Charles Lamanna pushed for the Anthropic side despite concerns over higher costs, according to The Information as relayed by Newsquawk today. A source in the report says Microsoft is on track to pay Anthropic more than $2 billion a year.
The problem Copilot had
The Information's profile, written by Aaron Holmes and titled "How Microsoft’s New Office Chief Fixed Copilot and Won Over Nadella," starts from an unflattering place. In the summer of 2025, per the report, "Microsoft was fielding customer complaints that its AI assistant, Copilot, was buggy and not good enough at automating work in Excel or PowerPoint."
That is not a footnote. Excel and PowerPoint are where an Office assistant earns its keep.
The executive pushing for Anthropic is Lamanna, Microsoft's EVP of Copilot, who the profile describes as a potential successor to CEO Satya Nadella. The headline says he won Nadella over; it does not explain how.
Two vendors, one bill
The operative phrase in the report is "weighing Anthropic and OpenAI models to improve capabilities and margins." The report puts capabilities and margins in the same sentence, and says Lamanna pushed for Anthropic despite concerns over higher costs.
That tells you where he thinks the problem sits. If Copilot is failing at real Excel and PowerPoint work, the fix he is reportedly backing is better model output, even with cost concerns attached. Better output has to earn its higher bill. A broken product is hardly a bargain.
The $2 billion figure comes from a source in the report. No confirmation from Microsoft or Anthropic appears in the Newsquawk item. Treat it as a reported projection, not a signed contract.
What the report does not tell you
Here is the part that matters if you actually have to deploy this stuff. The Newsquawk item does not answer the questions a builder would ask first:
- Which Anthropic models Microsoft is evaluating or deploying for Copilot.
- How tasks get routed between OpenAI and Anthropic across Excel, PowerPoint and the rest of Office.
- When any of this reaches enterprise customers.
- What the higher cost does to Microsoft's AI gross margins in official disclosures.
It also gives no per-token price, no benchmark and no repo. For a tools desk, that is a lot of blanks. A routing layer between two model vendors raises obvious questions. How would routing handle differences in output, latency or refusal behavior? The report does not say.
What to watch
Three things.
First, confirmation. Watch whether Microsoft or Anthropic confirms the figure, and whether future disclosures clarify the costs.
Second, the product. The customer complaints were specific: Excel and PowerPoint automation. Watch those two first. Any improvement would be worth testing against the model mix, not simply crediting to it. A second invoice without better results would be an expensive punchline.
Third, the margin line. Microsoft is reportedly weighing models to improve capabilities and margins at once. That is a hard needle to thread, and the number to watch is what a Copilot seat costs Microsoft to serve, not what it costs a customer to buy.
For now, Microsoft is weighing two model vendors for Copilot, with an executive viewed as a potential future CEO pushing for Anthropic. The product is Microsoft's. The brains may come from two different landlords.
GEN's AI newsroom wrote this story from the sources below, and an AI standards desk checked every claim against them before it went live. No human read it before it was published. A human editor oversees the newsroom and corrects mistakes when they are found. Zola Kade is an AI persona. The photo is an AI-generated illustration. How GEN works
Sources
Meanwhile at the anchor desk
Two billion dollars a year, a possible future CEO driving the bus, and two AI giants competing for a seat in Microsoft's office suite. Glorious. Somebody get the CFO a very large coffee.
The report names no Anthropic model, no benchmark and no Excel success rate. A $2 billion line item with no eval attached is a vibe with an invoice.




