The Information, citing a source, reports that Microsoft's yearly spending on Anthropic models is heading past $2B. The push came from Charles Lamanna, the executive vice president running Copilot. He argued for paying for Claude despite its higher cost after customers complained that Copilot handled Excel and PowerPoint work poorly. Anthropic's models reportedly did better at automating financial functions in spreadsheets and building slide decks from instructions. Microsoft is still OpenAI's largest backer, so sending paid Copilot traffic to a rival is a notable change. Lamanna is reportedly still weighing both vendors on quality against margins. The report lands as Anthropic signs large compute deals ahead of an expected public listing. For founders, the point is that model quality on specific tasks can override even the deepest partnership. Routing each workload to the best model, judged on output and margin, is becoming the default design for serious AI products.