AI Pulse · Business Models & Winners

Salesforce has put a real number on the cost of using frontier AI at scale. On its latest earnings call, Marc Benioff said the company expects to spend about $300 million on Anthropic tokens in 2026, mostly for coding work. Salesforce also raised its FY27 revenue guidance by $200 million, to $46.1 billion to $46.4 billion.

The more important detail for customers is how Salesforce plans to charge for “Salesforce in Claude”. It will be priced by API consumption, not by the number of user seats, with an open beta targeted for September. That changes the buying conversation. A seat-based AI add-on is easy to budget but often gets underused. Consumption pricing can better match value, but it can also produce surprise bills if teams start embedding AI into frequent workflows without controls.

For business owners, this is another sign that AI software costs will increasingly look like cloud infrastructure costs. You won’t just choose which tools employees can access. You’ll need to decide which processes are allowed to call an AI model, how often, with what data, and who watches the spend. Coding is Salesforce’s biggest stated use case today, but the same model will reach customer service, sales support, reporting, and internal operations.

This is the kind of thing we build into an AI command centre, tracking both where AI is being used and whether the cost is justified by the work it replaces or improves.