Salesforce cites Claude costs for lower profit margin outlook
Salesforce told investors it could not raise its profit margin guidance because of the cost of running Anthropic's Claude models internally, revealing how heavily AI spending is now weighing on the CRM giant's finances. Deputy CFO Mike Spencer told the Deutsche Bank Technology Conference that the company had "unleashed" Claude across its research and development cycle around six months ago to test how far it could accelerate its product roadmap, and the resulting token spend ate into margin improvements investors had been hoping for.
Salesforce's operating margin came in at 20.5 percent for its Q2 results (ending 31 July), but full-year guidance stands at 20.1 percent, a gap Spencer attributed directly to Claude token costs. He said the firm is now moving into "refinement mode," matching specific AI models to specific tasks rather than defaulting to the newest and most expensive options, and is comparing costs across multiple providers including OpenAI, Cursor and Claude, while experimenting with Grok. In May, CEO Marc Benioff had said Salesforce expected to spend $300 million with Anthropic in 2026; the company joins others such as Cockroach Labs in trying to curb AI "tokenmaxxing" costs, amid Gartner warnings that coding-agent spending could soon exceed developer salaries in some regions.
In short
- Salesforce says heavy Claude token spending held back margin guidance
- Full-year operating margin guidance held at 20.1%, down from Q2's 20.5%
- Firm now picking cheaper, task-appropriate AI models across multiple vendors
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Salesforce’s broad use of Claude can be seen as a deliberate investment in faster product development: testing the model across research and development may reveal where AI improves productivity and helps the company compete. Accepting lower near-term margins could be reasonable if the learning leads to better products or more efficient, targeted use of AI over time.
The case against
Investors can reasonably expect Salesforce to manage costs and protect margins, especially when expensive AI use has not yet produced clearly reported returns. Deploying a costly model widely before matching it to suitable tasks risks spending heavily for uncertain gains, so tighter controls and evidence of productivity improvements are warranted.
Explained simply
Salesforce makes software that businesses use to manage customer relationships, sales and support. It earns revenue mainly by selling subscriptions, and investors watch its profit margin: the share of that revenue left after costs.
Anthropic is an AI company whose Claude models can help with tasks such as writing and checking software code. Salesforce has been using Claude in product development, where each request to an AI model uses computing capacity measured in tokens; frequent or complex use can add up to significant costs.
Salesforce’s spending reflects a broader question for technology companies: whether AI tools can speed up work enough to justify their computing bills. Companies are comparing models and assigning them to different tasks to manage costs while continuing to develop AI features.
Read the full article at the source →
Originally published by The Register as “Salesforce blames its Claude addiction for denting profit margin guidance”.