A week after Anthropic sold Wall Street on refusal, Alex Karp went on CNBC and sold the opposite — the rage of every enterprise that thinks the frontier labs are taxing them. Same war, opposite trench.
Alex Karp came on CNBC’s Squawk Box on July 1 to talk about a chip partnership. He spent the next twenty minutes detonating.
Ostensibly there to discuss Palantir’s new sovereign-AI deal with Nvidia, the CEO instead unloaded on the frontier AI labs, calling the industry “effing insane” and accusing the biggest model makers of overcharging customers while quietly harvesting their data. When co-anchor Becky Quick noted that he sounded angry, Karp didn’t flinch: “This is the voice of American business that is being channeled through me.” After the segment appeared to wrap, he asked whether they were still on air — and kept going. Palantir stock rose more than 9% that day.
Strip out the theatrics and there’s a real argument underneath — the most coherent attack yet on how the frontier labs actually make money. Karp’s charge is that enterprises are paying premium per-token prices for value that never fully arrives, while the labs ingest their proprietary data and market “alpha” to sharpen models the enterprises will then compete against. He calls it a “wealth tax” on American business, and says the technology has been “irresponsibly oversold.” Whatever you make of the delivery, the pricing critique lands: frontier models are billed far above the marginal cost of inference, and that spread only holds while customers believe the value justifies it.
The pitch hiding inside the rant is Palantir’s answer to exactly that anxiety. Its new deal folds Nvidia’s lower-cost Nemotron models into Palantir’s platform so government and enterprise customers can run AI in secure environments while keeping control of their own weights, data, and compute. Karp’s framing was blunt: customers want to own the means of production rather than rent intelligence from a lab that also sells to their rivals. Sovereignty, not tokens.
Put that next to what Anthropic’s Dario Amodei was selling the week before, and the real shape of the AI fight comes into focus. As we argued here, Amodei has built the most defensible brand in AI on the things his lab refuses to do — a moat made of trust. Karp is selling the mirror image: don’t trust any of them, own your stack so they can’t tax you. Two opposite pitches, one identical customer fear — handing your crown jewels to a company that also does business with your competitor. When the models themselves are converging, nobody’s moat is the model anymore. The war has moved to who controls the layer underneath it.
There’s an irony Karp mostly glided past. Palantir is one of the frontier labs’ most important deployment partners; by his own nod, a great deal of what Anthropic does in the field runs on Palantir’s rails. The revolt he’s channeling is real, but the man channeling it sells the picks and shovels to the same labs he’s calling insane — and just launched a product designed to convert that resentment into Palantir contracts.
Which is the tell worth holding onto. Karp was talking his book. The Nvidia deal was the thing the meltdown was selling, the “livid” CEOs he kept invoking are his prospective customers, and the 9% pop was the market pricing in the pitch. A diagnosis can be accurate and self-serving at the same time, and this one is both.
Still, the soft spot he’s jabbing is genuine — and it’s the same one underneath Anthropic’s story. The frontier labs’ pricing power funds everything, including the safety posture Amodei markets so effectively. The moment cheaper open-weight models and credible sovereignty pitches give enterprises a real alternative, that premium starts to compress, and the whole structure it pays for gets less comfortable. Karp’s performance was unhinged. It may also have been early.
