Alex Karp has issued one of the sharpest public attacks yet on the frontier AI industry, accusing OpenAI and Anthropic of overselling artificial intelligence to corporate customers while charging premium prices for tools that often fail to deliver measurable business value.
- Palantir CEO Alex Karp accuses OpenAI and Anthropic of irresponsibly overselling artificial intelligence capabilities to global corporate customers and governments.
- Corporate boards demand proof of ROI after two years of heavy AI spending as token-based costs outpace actual business productivity gains.
- Palantir expands its Nvidia partnership to provide secure AI systems, challenging the data governance and pricing models of frontier laboratories.
Speaking on CNBC’s Squawk Box on Wednesday, the Palantir chief executive said many companies are paying for AI tokens that “create no value” while handing over proprietary data and intellectual property to model providers. He suggested the arrangement is wearing thin inside large enterprises.
Karp’s comments come as the AI sector shifts from hype-driven experimentation to hard questions about return on investment. After more than two years of heavy spending on foundation models, boards and chief executives are scrutinising productivity gains, deployment costs and data governance. The remarks also coincide with Palantir’s expanded partnership with Nvidia to deliver secure AI systems for government customers, putting the company in sharper competition with frontier labs for enterprise budgets.
“This is the voice of American business that is being channelled through me,” Karp said after a CNBC host remarked that he appeared unusually animated during the discussion.
A Growing Enterprise Divide
The first wave of enterprise AI adoption focused on experimentation. Companies rushed to integrate large language models into customer service, coding and internal tools, often accepting high costs for access to fast-improving technology.
That approach is changing. As spending moves from pilot projects to core operating budgets, executives are demanding clear productivity gains rather than larger cloud bills. Karp argued that the dominant token-based pricing model misaligns incentives.
Businesses, he said, are “paying for tokens that create no value” while feeding data that strengthens the very models they are buying access to. His critique is one of the most direct public challenges to the usage-based commercial model used by leading frontier AI companies.
More Than a Pricing Dispute
Karp’s criticism goes beyond cost. He questioned whether governments and major enterprises should rely heavily on a handful of Silicon Valley labs for critical capabilities, especially in national security contexts.
This stance aligns with Palantir’s strategy. While OpenAI and Anthropic focus on building ever-more-powerful foundation models, Palantir positions itself as the enterprise and government layer that deploys AI securely, integrates it into operations and maintains strong governance.
The interview followed Palantir’s recent expansion of its partnership with Nvidia to deliver secure AI infrastructure for US government agencies, highlighting the company’s push to own the deployment layer rather than compete solely on model performance.
The Economics of Enterprise AI Are Being Tested
Karp’s remarks reflect broader pressures in the market. Open-source models have grown more capable, new entrants are challenging premium pricing, and finance chiefs are demanding proof that AI investments improve productivity or create new revenue.
Enterprises are shifting focus from benchmark performance to practical issues: deployment costs, data ownership, vendor dependence and whether usage-based pricing truly reflects delivered value. These questions gain urgency as organisations move AI from isolated pilots to deployment across thousands of employees.
Grey Terminal Note
Karp’s remarks are unlikely to derail OpenAI or Anthropic. They do, however, signal a shift in the conversation surrounding enterprise AI.
For the past two years, the industry has competed largely on capability, larger models, faster benchmarks and more impressive demonstrations. The next phase is likely to be decided by economics.
Enterprises aren’t buying intelligence for its own sake; they’re buying measurable outcomes. The companies that can consistently prove value, rather than simply generate more tokens, may ultimately define the commercial winners of the AI era.
Activate Terminal Layer
Structural analysis of the systems, pressures, and stakeholders behind this story.
