Salesforce and Anthropic team up as customers get billed by the sip

Key points

• Salesforce reported Q2 revenue of $11.3 billion, up 11 percent year-on-year, beating analyst forecasts.
• The company announced Claudeforce, a joint product with AI model maker Anthropic that combines Claude's capabilities with Salesforce business data and workflows.
• Fifty percent of Salesforce's AI bookings came from existing customers buying more Flex Credits, the company's pay-as-you-go usage tokens.
• Gartner warned in October 2024 that credit-based AI pricing models can expose buyers to unexpected cost increases.
• Salesforce's all-you-can-eat licence deal, announced October 2024, may eventually convert to fixed-quantity contracts, a Gartner analyst believes.

Salesforce had a good quarter on paper. The company, best known for its customer-relationship software used by sales teams worldwide, posted $11.3 billion in revenue for the three months ending 31 July. That beat Wall Street's expectations and sent shares up 12 percent on the day.

The results arrived alongside a product announcement that helped fuel the optimism.

What is Claudeforce?

Claudeforce is a joint product from Salesforce and Anthropic, the AI company behind the Claude family of chatbots and assistants. It connects Claude's conversational abilities directly to the data and workflows a business already keeps inside Salesforce. We first covered Claudeforce on 27 August 2026, before Thursday's earnings announcement fleshed out the commercial picture.

The product includes a plug-in with 37 pre-built "sales skills", small automated tasks designed to handle repetitive parts of a sales process. It also works inside existing Salesforce tools including Slack and Tableau, its data-visualisation product.

Salesforce co-CEO Marc Benioff described a growing family of connected products: Claudeforce alongside a reworked Slack, an AI assistant called Coworker, and a feature called Headless 360 that lets workers pull Salesforce data through apps they already use. That product breadth, as our 26 August story on AI agents failing in enterprise software noted, is exactly where implementation complexity tends to bite.

For ordinary users, this means the software their company uses for sales and customer tracking could start responding to plain-English requests rather than requiring menu clicks. Whether that works well in practice depends entirely on how each company sets it up.

How does the pricing work, and what should customers watch?

This is where independent analysts are raising flags.

Salesforce is pushing a model called Flex Credits: tokens you buy upfront, then spend each time the AI does something. Salesforce President Robin Washington told investors that 50 percent of the company's AI bookings came from customers who'd run through their credits and bought more. That stat tells you the model is sticky. It doesn't tell you whether it's affordable.

Benioff framed multiple payment options as customer-friendly: pay by consumption, by outcome such as a completed transaction, or by basic usage.

Gartner, first reported by The Register AI, warned buyers in October 2024 that credit-based models from Salesforce, Microsoft and ServiceNow carry real risk. Gartner's note said these schemes "often require upfront usage commitments" with "rates that may change unilaterally by the vendor, exposing buyers to unexpected cost increases." Gartner added it had seen no evidence of vendors lowering prices as usage grows, and told buyers to negotiate volume discounts before signing.

Separately, Gartner director analyst Hannah Decker said in January 2025, in comments to The Register AI, that Salesforce's all-you-can-eat licence deal, called the Agentic Enterprise Licence Agreement, would likely convert to fixed-quantity contracts at renewal. Salesforce disputes that. EVP Bill Patterson said: "The claim that we are moving away from capped agreements is inaccurate. Renewals remain flexible."

Should you worry if your company uses Salesforce?

Four things are worth checking before anyone signs an AI upgrade contract.

First, ask whether billing is credit-based and what happens when credits run out mid-month. Second, ask whether the credit rate can change during the contract term. Third, get any volume discounts written into the agreement upfront, not promised verbally. Fourth, read the exit terms carefully before committing to a multi-year deal.

The product announcements are real, the revenue growth is real, the Anthropic partnership is live. The pricing model's long-term cost profile isn't. That's the part worth scrutinising as hard as the technology.