Palantir CEO Alex Karp delivered a blistering critique of the AI industry's dominant business model this week, accusing companies like OpenAI and Anthropic of building enterprises on "effing insane" foundations that leave customers paying escalating token costs for questionable value while risking their proprietary data.

In a fiery interview with CNBC's "Squawk Box," Karp pulled no punches in attacking the token-based pricing approach that has become standard among frontier AI labs. His comments reflect growing frustration among enterprise customers who feel trapped in an unfavorable economic arrangement.

"Something Has Gone Completely Wrong"

Karp argued that the current AI business model fundamentally misaligns incentives between AI providers and their enterprise customers.

"Something has gone completely wrong," Karp said. "The basic view among enterprises in this country is 'I'm going to chillax and waste my time with tokens, I'm going to get no value, and they're going to get my IP.'"

The Palantir chief said that enterprise leaders have privately confided their concerns about AI companies gaining access to sensitive corporate data and competitive advantages—what he termed their "alpha." These executives, Karp suggested, are "livid" but unwilling to voice their frustrations publicly.

The "Tokenmaxxing" Problem

Karp's critique centers on what some in the industry have dubbed "tokenmaxxing"—the practice of monetizing AI primarily through per-token charges that scale with usage. This model has generated enormous revenue for companies like OpenAI, but Karp argues it fails to deliver commensurate value to customers.

Palantir recently published a nine-point manifesto on "AI sovereignty" that elaborates on these concerns. The document warns: "Data retention is your treasure. Transfer it at your own peril."

Karp's alternative vision emphasizes enterprise-controlled AI deployments where companies maintain ownership of their data and the insights derived from it. Palantir's business model depends on selling software that runs inside customer environments rather than through cloud-based API calls.

Beyond Business Models: National Security

Karp extended his critique to national security concerns, taking particular aim at Anthropic's reported reluctance to allow its AI models to be used in certain military applications.

"Are we really going to outsource the battlefield of this country to the consensus view in Silicon Valley? That is effing insane," he declared.

The comments came as Palantir announced an expanded partnership with Nvidia to develop open AI models for use in U.S. government agencies—a project that directly competes with frontier labs for defense and intelligence contracts.

Industry Pushback and Validation

Karp's broadside echoes concerns raised by other technology leaders. Microsoft CEO Satya Nadella recently warned that entire industries could "find their knowledge commoditized right out from underneath them" through AI interactions.

For enterprise customers evaluating AI investments, Karp's critique raises important questions about data security, cost predictability, and long-term strategic control. Whether his alternative vision gains traction may depend on whether enterprises decide the current model's convenience outweighs its drawbacks.

The AI industry's response to these critiques will shape how the technology evolves from experimental tool to enterprise infrastructure in the years ahead.