Palantir reports second quarter earnings on August 3, and Wall Street already expects revenue to climb about 80% from a year earlier. That would follow a first quarter in which revenue from United States commercial customers more than doubled and the company raised its own full year forecast by close to half a billion dollars in a single quarter. Numbers like that are rare for any software company at this size, and they are part of why investors have pushed Palantir's stock to one of the richest valuations in the industry.
But there is a catch buried in the numbers. Almost all of last year's added commercial revenue came from Palantir's own existing customers spending more, not from new companies signing up, and its 20 biggest clients now pay close to 100 million dollars a year on average, up sharply in just twelve months. That is less a story about widespread AI adoption and more a story about a small number of companies going all in on one vendor.
The reason so much money flows from fewer companies comes down to how Palantir actually delivers its product. Instead of selling software and walking away, Palantir sends its own engineers to live inside a client's offices, sometimes for months, building the systems by hand until artificial intelligence becomes part of daily operations rather than a side project. Palantir calls this staff forward deployed engineers, and CEO Alex Karp has said only a handful of the company's salespeople actually sell in the traditional sense.
This approach used to look like Palantir's odd quirk. Now competitors are copying it outright: Microsoft just launched a new unit called Microsoft Frontier Company, backing it with 2.5 billion dollars and 6,000 staff whose job is to embed with clients the same way Palantir's engineers do, and Amazon is reportedly building something similar. When your biggest rivals start copying your delivery model instead of your software, that tells you the real product was never just the code.
The catch is that embedding this deeply creates dependence, and Palantir's own track record shows what that looks like when it goes wrong. Britain's National Health Service signed Palantir to a data contract worth up to 330 million pounds back in 2023. Since then, a parliamentary committee has pushed the government to consider walking away, more than 100 NHS technology staff have called for the platform to be replaced, and the health service's own chief executive has asked for an independent review of whether the promised benefits were ever real.
Palantir is meanwhile telling other companies not to become this dependent on anyone else. It recently joined Nvidia, Microsoft, Meta and more than twenty other companies asking policymakers not to restrict open weight AI models, systems a company can run on its own computers instead of renting access from an outside provider, and Karp has said publicly that he thinks the leading AI labs are overcharging customers for access to their models. The irony is that Palantir's own biggest relationship, the NHS deal, is the clearest example yet of what happens when an organization leans on one AI vendor for years and then tries to get out.
Whatever Palantir reports on August 3, the more useful number for any business leader watching from outside is not the growth rate. It is how many of Palantir's biggest customers keep paying more every year, because that ratio is the real measure of whether the company is becoming essential infrastructure or an expensive habit that is hard to break.