Palantir Technologies (PLTR) shares edged up less than 1% to $132.66 on Wednesday after CEO Alex Karp criticized frontier AI labs in an interview with CNBC, saying many enterprise clients are privately dissatisfied with their billing practices. Karp told CNBC's Sara Eisen that the discontent extends beyond everyday users and noted that most of the Anthropic projects discussed in public are actually running on Palantir's infrastructure.
Sticker Shock and Token-Based Billing
Corporate AI buyers are beginning to set real budgets rather than just running pilot programs. Business Insider reported Wednesday that OpenAI, Anthropic, and GitHub are steering more customers toward token-based billing, where users pay for the volume of text processed by AI models. Coinbase and Walmart have introduced usage caps, while Amazon has dropped its internal token leaderboard.
The stakes are rising as leading model makers seek to demonstrate that large-scale corporate AI use can deliver returns. Axios noted that Anthropic has filed pre-IPO paperwork while major customers experience an AI "sticker shock" phase. The report cited Bain & Company data showing that 40% of nearly 1,000 companies surveyed saw AI deliver cost savings of less than 10% after their investments.
Palantir's Pitch: Own the Means of Production
Karp is actively leveraging this unease to attract customers. At Palantir's AIPCon 10 conference, he told attendees to first try a large language model company and observe that "they don't care about you at all." After paying heavily in tokens, he urged them to return to Palantir, which he says wants customers to "own the means of production." Karp specifically criticized "token maxing," where companies showcase AI activity without addressing actual business problems. He revealed that Palantir has an internal tool designed to prevent enterprises from engaging in such behavior, arguing that more AI calls do not necessarily translate into operational change.
Palantir CTO Shyam Sankar was equally blunt at the event, stating, "More tokens means more slop." Karp's core argument is that large-scale enterprise work still depends on steady, precise processes, with large language models serving as assistants rather than replacements.
Industry Voices and Cost Concerns
Other AI executives have also raised cost issues, though not always in Palantir's terms. Replit's Michele Catasta, the company's president and head of AI, described token leaderboards as "very dystopian" and cautioned that token usage is not proportional to business impact, while warning that heavy AI demand drives up energy and compute usage.
Anthropic's Response: Claude Fable 5
Anthropic is moving to address cost concerns with its new model, Claude Fable 5. Reuters reported Monday that the model is more powerful and charges $10 per million input tokens and $50 per million output tokens. Anthropic claims lower token usage could make tasks cheaper for some users, and the model includes new safeguards to curb cybersecurity misuse.
Analyst Perspectives and Risks
Wedbush analyst Dan Ives reaffirmed an Outperform rating and $230 price target after AIPCon, highlighting Palantir's forward-deployed engineers as the reason the company is "irreplaceable" for AI rollouts. However, the case is not bulletproof. If Anthropic, OpenAI, or others can rapidly reduce per-task costs, Palantir's role as a check on spending could diminish. Governments may also view Palantir as a risky dependency, potentially delaying contract approvals. UK officials are currently examining a £330 million NHS data deal with Palantir after lawmakers raised concerns about patient confidentiality, trust, and overreliance on a U.S. supplier.
Market Reaction
Palantir traded just under 1% higher at $132.66 late Wednesday morning in New York, valuing the company near $341 billion. Karp's latest comments did not appear to move the shares significantly, with the AI spending story failing to generate a new catalyst for the stock in this session.
