Alex Karp, chief executive officer of Palantir Technologies, told CNBC that AI labs have “completely, irresponsibly, oversold” their models. Karp said that enterprise customers are increasingly angry because they are being billed on a per-token basis, receiving limited tangible value, while the labs may be absorbing customers’ intellectual property (IP) and market advantage (alpha).
In his remarks Karp presented himself as speaking for other CEOs who feel similarly frustrated in private but are reluctant to voice those criticisms on camera. He highlighted a practical tension between vendors’ sales pitches and their billing practices: marketing promises automation and intelligence, the invoice itemizes token usage, and contractual terms can expose customers’ data to reuse or training that erodes their proprietary know-how.
Karp’s comments feed into a broader conversation around so‑called “AI sovereignty.” In this context the term has shifted from geopolitical connotations to a procurement and control concern: companies are increasingly treating AI supplier selection and contract terms as matters of defensive purchasing. Businesses do not only fear inaccurate or low-quality models; they fear paying external providers in ways that commoditize the knowledge and capabilities that made those businesses competitive.
The criticism also serves Palantir’s position in the market: the company sells its own enterprise AI and data-management solutions, so a sharp public critique of prevailing vendor practices is both a strategic and communicative act. Karp’s argument centers on three linked issues — token-based billing, poor value delivery relative to cost, and the risk that vendors repurpose customer data — which together drive corporate demand for stronger contractual protections and greater control over AI deployments.
In short, Alex Karp’s remarks call attention to a structural friction between the business models of AI service providers and the risk-management needs of enterprise clients. The debate is relevant for procurement teams, legal counsel and technology buyers as they negotiate pricing models, data-use clauses and IP protections with AI vendors.
