Palantir CEO Alex Karp leveled direct frontier labs criticism at the biggest names in artificial intelligence Wednesday, telling CNBC that every enterprise customer his company deals with is privately frustrated by how Palantir‘s AI rivals operate.
| Item | Detail |
|---|---|
| Karp’s core claim | Every enterprise Palantir works with is privately unhappy with frontier labs |
| Anthropic valuation | $965 billion (confidential IPO filing) |
| OpenAI valuation | $852 billion (last private round, March 2026) |
| OpenAI S-1 filed | June 8, 2026 (confidential, with SEC) |
| OpenAI compute spend (2028 est.) | ~$85 billion in that year alone |
| Palantir AIP launch | 2023, connecting LLMs to enterprise data and operations |
“It’s not just the man and woman on the street that is unhappy with the frontier labs,” Karp told CNBC’s Sara Eisen. “It’s in private, every single enterprise we deal with.” His specific complaint: that these labs don’t understand customer businesses and are focused on “tokenmaxxing,” or burning through AI tokens to perform productivity rather than deliver it.
The timing is pointed. Both Anthropic and OpenAI are heading toward public markets, and Karp’s comments land directly on the narrative each company needs investors to believe.
Two Massive Valuations, One Frontier Labs Criticism
Anthropic confidentially filed for an IPO after raising $65 billion in private funding, pushing its valuation to $965 billion. The company recently launched Claude Opus 4.8, positioning it as a step forward in coding and professional tasks. OpenAI submitted its own confidential draft S-1 to the SEC on June 8, 2026, a week after Anthropic moved. Its last private round, closed in March 2026, put its valuation at $852 billion on $122 billion raised, with Goldman Sachs and Morgan Stanley tapped to lead a potential offering.
That is a combined paper value near $1.8 trillion for two companies that have yet to demonstrate they can convert token-burning scale into durable enterprise profit. Karp is essentially arguing that enterprises are already skeptical, and that the frontier labs criticism circulating in private boardrooms is more widespread than Wall Street’s IPO enthusiasm reflects.
The cost side of the equation adds weight to his argument. OpenAI projects spending roughly $85 billion on computing power in 2028 alone, even assuming its sales double from the prior year. That is the structural gap the labs have to close: model costs rise with usage, and if enterprise customers are unhappy now, they will push back harder as bills climb.
Where Palantir Positions Itself in the Frontier Labs Debate
Karp’s argument is not that large language models are irrelevant. “It is not that large language models aren’t crucial for the world,” he said. “It’s just the implementation is where the value is, certainly in the next seven years.” That framing maps directly to what Palantir sells.
The company launched its AIP (Artificial Intelligence Platform) in 2023, designed to connect large language models and other AI with customers’ data and operations across both commercial and government sectors. The product sits between the frontier labs and the enterprise, which is precisely the layer Karp says the labs can’t or won’t occupy effectively.
He also told CNBC that most of Anthropic’s publicly discussed projects are “running on Palantir,” a claim that positions the two companies as partners even as he criticizes the broader lab ecosystem. He called Anthropic cofounder Dario Amodei “a very, very important person” while noting his own frequent disagreements with him.
Karp’s broader frontier labs criticism extended to the politics of AI. He called AI “a massive revolution” with dangers and opportunities unique to the U.S., and pushed back against partisan framing of the technology. “You can’t do a blue-red debate,” he said. The comment comes from a CEO who has navigated his own political contradictions, describing himself as a “card-carrying progressive” while aligning with the Trump administration and having previously donated to both Harris and Biden campaigns.
Whether enterprises are as uniformly frustrated as Karp suggests, or whether his comments are strategically timed for maximum contrast ahead of two landmark AI IPOs, the argument he is making is a commercial one: implementation beats raw model capability, and Palantir is the implementation layer. The real test of that thesis comes when Anthropic and OpenAI start trading and investors can price the gap between frontier model hype and what enterprises actually pay for.
