Salesforce's blowout quarter has a clear message: data owns the AI economy, not the models
While investors spent a year fearing that AI would kill business software, the companies building the most powerful AI models quietly became Salesforce's biggest customers.
Key points
- Salesforce's fiscal Q1 2026 revenue grew 11% to $11.35 billion, beating analyst expectations by a wide margin.
- Nine of the top ten AI companies now run on Salesforce and Slack, with their combined spending up 435% year over year.
- Salesforce processed 104 trillion customer records this quarter, a 355% increase on the same period last year.
- Agentforce, Salesforce's AI agent product (software that can carry out multi-step tasks on its own), grew from $100 million to over $1.5 billion in annual recurring revenue in 18 months.
- Salesforce announced a $25 billion share buyback, the largest in the company's history.
For about a year, a fear gripped investors: AI agents were going to make business software companies obsolete. Markets wiped out roughly $2 trillion in software company value chasing that idea. Analysts named it the "SaaSpocalypse." Salesforce's latest earnings report suggests the fear was badly misdirected.
So what does Salesforce actually have that AI needs?
Data. Enormous, trusted, proprietary data that AI agents can't do useful work without.
An AI agent trying to close a sales deal still needs somewhere to look up the customer, log the contract, and check what pricing that customer was offered last year. Without reliable data behind it, the agent is guessing. Junk in, junk out.
Salesforce is one of the largest stores of business customer data anywhere in the world. This quarter, its Data Cloud product ingested 104 trillion customer records, up 355% year over year. Every AI agent that completes a task also produces new data, and that data has to live somewhere trustworthy. Agents do more work, more data flows in, and the data makes future agents more useful.
Analysts at Wells Fargo put the underlying economics plainly: when one ingredient becomes cheap and abundant, value moves to whatever's still scarce and hard to copy. Intelligence is getting cheap fast. Frontier large language models, the technology behind chatbots like ChatGPT and Claude, leapfrog each other every few months and are growing increasingly interchangeable. Proprietary business data isn't getting cheaper. It's getting more valuable.
What did the numbers actually show?
The results demolished expectations. Non-GAAP operating margins (a measure of profitability that strips out one-off accounting items) reached 34.1%. Adjusted earnings of $5.90 per share nearly doubled, crushing the analyst consensus of roughly $3.27. Salesforce raised its full-year revenue guidance to as much as $46.4 billion.
The most telling figure wasn't a margin number, though. It was the customer list. Nine of the top ten AI companies, the same firms supposedly building software that would replace Salesforce, now run their own operations on Salesforce and Slack. Their combined spending rose 435% year over year.
We've tracked this partnership since it emerged. Our earlier story from 27 August found that Claudeforce's credit-based pricing model was drawing fresh scrutiny from analysts who warned of hidden cost spikes. The commercial logic is clearer now: Salesforce CEO Marc Benioff, speaking on CNBC, said the leading AI models "depend on CRM... They do not replace them."
Not every software company benefits from this shift. Firms whose only advantage is doing tasks an AI can now replicate cheaply face real pressure. Companies sitting on proprietary data that agents genuinely need are in a different position entirely. That's the judgement Benioff is backing with a $25 billion buyback.
Common questions
Does this mean AI is less powerful than we thought?
No. It means the value is shifting from companies that build AI models to companies that own the data those models need to do real work.
Should ordinary software users be worried about their jobs?
The evidence here suggests AI agents are adding to the workload that flows through business software platforms, not replacing the platforms themselves. What happens to individual roles is a separate and still open question.
