Salesforce Confronts $300 Million AI Bill With Model-Routing Strategy

Salesforce's aggressive adoption of AI models has put a strain on its profit margins. The company has been using Anthropic's Claude models across its research and development cycle, which has led to a $300 million line item in their spending.

Speaking at the Deutsche Bank Technology Conference, Salesforce deputy CFO Mike Spencer said that they opened the floodgates on Claude six months ago, resulting in token consumption that affects their full-year profitability guidance. He framed it as a deliberate strategic investment, aiming to unlock breakthroughs and compress product roadmaps by saturating the development organization with frontier AI.

The company is now shifting from a 'deploy everywhere' posture to what Spencer calls 'refinement mode', matching models to specific tasks and reserving advanced models for complex work. They are also experimenting with other vendors, including OpenAI, Cursor, and X's Grok.