Revenue outgrowing customer count does not prove deeper AI adoption

Palantir's growth makes me want to inspect account depth. It does not let me infer the depth of every customer's AI deployment from a pair of aggregate numbers.

The Q2 2026 release reports $1.935 billion of quarterly revenue, up 93 percent year over year. Its Form 10-Q counts 1,049 customers for the trailing twelve-month period, compared with 849 a year earlier. That is roughly 24 percent growth in the reported customer count.

Revenue clearly grew faster than that count. The observation is worth investigating. It does not, on its own, establish that enterprise AI spending across the market is concentrating into fewer deployments or that rebuilding workflows caused the difference.

The customer measure also covers a different period from the quarterly revenue figure. An average made by dividing one by the other needs to retain that mismatch rather than masquerade as the economics of a typical account.

An average cannot describe the deployment

Imagine a hypothetical vendor whose revenue rises while its customer count barely changes. Existing customers may have expanded, new customers may be much larger, pricing may have changed or the mix of contracts may have shifted. Several mechanisms can coexist.

An aggregate average cannot identify which one dominates. Nor does a higher average establish that the median customer is spending more. A few large relationships can move the mean while many others remain unchanged.

This matters when someone borrows the vendor result to argue that an internal AI programme should abandon small pilots and fund a few large projects. That may be a good strategy for a particular organisation. The decision still needs evidence about its own workflows, constraints and adoption path.

I would use the financial result to ask better questions about depth, not to skip them. Which existing accounts expanded? What work changed? Did additional spending follow accepted outcomes, broader access or a different commercial arrangement?

Those questions require evidence beyond a top-line earnings figure. A vendor's growth and a customer's return are connected hypotheses, not interchangeable measurements.

Define depth in terms the operator can inspect

For an internal programme, I would describe depth as sustained use inside a specific workflow, with an owner and an outcome measure. Count the eligible work that reaches the system, the work accepted without correction and the exceptions that return to people.

Then compare the same workflow over time. A rising licence bill alongside unchanged throughput is different from a stable bill supporting more accepted work. Adding departments is different from making an existing deployment more useful.

A financial run rate needs similar care. Multiplying a quarter's revenue by four produces an annualised rate. It does not by itself establish annual recurring revenue or tell us how much of the business will recur under a consistent subscription definition.

I would keep that distinction in any comparison with a software vendor reporting ARR. Otherwise the label imports assumptions that the calculation has not earned.

Palantir's quarter supplies a strong reason to examine expansion economics. It supplies no shortcut around understanding what expanded and who benefited.

Treat revenue outgrowing customer count as a prompt for cohort and workflow analysis, not as proof that deeper AI adoption caused the growth.