Salesforce blames its Claude addiction for denting profit margin guidance

AI AND ML

But investors hear CRM giant is now in 'refinement mode,' picking models more carefully 

Salesforce was unable to improve its margin forecasts because of the dollars it is spending on Claude tokens, an investor conference heard last week.

As an example of how heavily the CRM giant is investing in its own AI capabilities to build products for customers, Mike Spencer, deputy CFO and head of finance Salesforce, told the Deutsche Bank Technology Conference last week that spending with Anthropic — the LLM-builder with which it has launched a new partnership — was sufficient to manage investor expectations in terms of margins.

In its results last week, Salesforce said its operating margin, according to accounting rules, would be 20.5 percent in its Q2 results (ending July 31), but its guidance for the full year is 20.1 percent.

Spencer said the drop comes because it was spending so much on Claude tokens.

“Roughly about six months ago we unleashed Claude in our R&D cycle. It's part of the reason we didn't raise margin guidance on the year because we're covering some of the token spend that we've got going. And the goal of that really was to, let's see what we could break. Let's see what kind of advancement our R&D teams can make on accelerating the product road map. Worst-case scenario, we would pull back.”

But Salesforce is now focused on getting greater value for money from AI spending by reconsidering which models it uses for which jobs, in line with many customers, he said.

“We're going into refinement mode. We're going into the zone of ‘prescription model choice for task at hand,’ and what that really means is you don't need to use the latest and greatest model for every single task you might want to do,” Spencer said.

He said for some tasks in software development or building the wider tech stack, the latest models might be necessary, but for “the large majority,” they were not.

“You're totally fine with the second or third generation model. That also includes, by the way, optimization across different vendors. Internally, we've got OpenAI. We've got Cursor, we've got Claude. So we've got a bunch of different model generators. We're starting to experiment with [X’s] Grok. All of them have different cost structures. We are, I think, a good representation of what we see in our customer base.” he said.

In May this year, Salesforce CEO Marc Benioff told the VC podcast All-In that the company had expected to spend $300 million with Anthropic, maker of the Claude model, in 2026.

Salesforce is not the only company trying to optimize its spending on AI to avoid the money pit of so-called tokenmaxxing.

Cockroach Labs CEO Spencer Kimball told The Register in June that internally, the company was using more open source AI models as it tries to optimize spending.

Gartner has found that spending on AI could mean that the cost of coding agents soon outweighs the salary of the developer, at least in some parts of the world.®