Salesforce blames its Claude addiction for denting profit margin guidance

Salesforce has struggled to improve its margin forecasts due to the significant investment in Claude tokens, according to an investor conference held last week. Mike Spencer, the company's deputy CFO and head of finance, revealed during the Deutsche Bank Technology Conference that spending with Anthropic, the provider of the Claude model, was necessary to manage investor expectations regarding margins.

The company's operating margin for Q2, ending July 31, was expected to be 20.5 percent, but their guidance for the full year stands at 20.1 percent. Spencer explained that the drop in guidance is attributed to the heavy spending on Claude tokens, as they unleashed Claude in their R&D cycle six months ago. The goal was to explore new advancements in the product road map; however, if it didn't work out, they were prepared to scale back.

Now, Salesforce is focusing on optimizing AI spending by reconsidering which models to use for specific tasks, aiming for a "prescription model choice for the task at hand." They recognize that not every task requires the latest and greatest model; in many cases, second or third-generation models suffice. The company internally uses models from various vendors, including OpenAI, Cursor, Claude, and X's Grok, each with its own cost structure.

Salesforce is now experimenting with these models to find the optimal balance between cost and performance. This approach mirrors what many customers, including Cockroach Labs, are doing to optimize their spending on AI.