A company best known for defense-and-data software just became the loudest voice telling businesses to stop pouring their secrets into AI. In a heated CNBC interview, Palantir CEO Alex Karp accused the big AI labs of running a "wealth tax" on business, and David Sacks and Chamath Palihapitiya quickly picked up the thread. Pat's read: it's a good reason to look harder at open source.
What Karp actually said
Karp's core charge is that the big AI labs are running a "wealth tax" on business: they charge companies high fees for their tools while collecting the data that flows through them, data that then improves the labs' own models. "This is the voice of American business that is being channeled through me," he said, claiming the CEOs he speaks with privately are "livid."
The reason it stings is structural. The company you pay for AI is also building products, and everything you run through it is a map of where the valuable work is.
Sacks and Chamath sharpen it
Reacting to the interview, White House AI czar David Sacks put the risk in plain terms: enterprises are "transferring their knowledge, their know-how, their trade secrets or customer data to these model providers who might eventually decide to compete with them." Chamath Palihapitiya backed it without hedging, calling it "insanely dumb" to keep feeding a provider your edge when a company "could find an independent way to access this intelligence."
The evidence
This isn't only theory. The Information reported that Anthropic has blindsided some of its own business partners, and Claude has been expanding into Design, Science, Security, Legal, and Code, the same kinds of work many of Anthropic's customers sell. A model provider with millions of users can see which tasks people run most, and those tasks are a shortlist of products worth building.
The proposed fix from Karp's camp is sovereignty: keep your data, the model, and the computing under your own roof instead of renting all three through an API. Palantir's own pitch, unsurprisingly, is open-weight models running on hardware you control.
Worth keeping honest: the labs do sign agreements promising not to train on customer data, and Karp is talking his book, since Palantir sells the alternative. That doesn't make the underlying point wrong, but it's why the loudest voice here also has the most to gain.
Strip away the billionaire feud and this lands on anyone who types into an AI. Every idea, client note, and rough draft you feed ChatGPT or Claude goes to a company that is also shipping its own products.
For a one-person business, that is usually a fair trade. The help you get back outweighs the risk, and the promises not to train on your data are real.
But the bigger the edge you're guarding, the harder it gets to justify handing it to a company that might turn around and compete with you. An open-source model you download and run yourself keeps that data in the building.
Running your own AI is still a real lift. It gets lighter every month, and the reason to bother is finally about more than saving money.
Sources
- Forbes: Palantir billionaire Karp blasts AI industry as 'effing insane' - David Sacks on Karp's CNBC interview (clip) - @TheChiefNerd - David Sacks - Chamath Palihapitiya - The Information: Anthropic Blindsides Its Business Partners - The PrimeTime: The World's Evilest Company