Palantir CEO Alex Karp warns enterprises may lose “their IP” as AI labs push token-centric products

Palantir Technologies CEO Alex Karp said Wednesday that many enterprises are becoming skeptical of how frontier AI companies are packaging their products, arguing that the market is drifting toward “tokens” instead of business value and creating risks for companies’ intellectual property.

Karp made the remarks during an appearance on CNBC’s “Squawk Box,” according to coverage of his comments. He characterized the prevailing enterprise mindset as one in which companies fear they will “waste” time on token consumption while also handing over their proprietary information to AI providers. He added that the situation is so misaligned that “something has gone completely wrong.”

In Karp’s description, the concern is not merely cost or performance. He said executives have privately told him they are worried AI firms could gain access to proprietary data and to “alpha,” which he used to mean the unique competitive edge that differentiates a business.

The “token” concept Karp criticized is common in today’s AI systems. In general terms, a token is a unit of text that large language models process and generate. Charging and usage patterns built around tokens can incentivize higher-volume interactions. Karp’s point was that this framework can push users toward consuming compute and generating outputs without necessarily improving enterprise decision-making or safeguarding sensitive information.

Karp also framed his comments as a broader critique of the AI ecosystem, rather than a targeted attack on a single competitor. While he suggested he could debate individual leaders privately, the thrust of his argument was that the ecosystem’s incentives may be steering customers away from what enterprises say they need.

The remarks land at a time when business buyers are weighing how to integrate AI into operations, including where data boundaries sit and who controls them. Karp’s comments underscore a recurring tension in enterprise AI deployments: companies want the capability of cutting-edge models, but they also want stronger assurances about governance, privacy, and reuse of internal data.

What Palantir did not disclose in the CNBC-linked reporting was additional specificity on timing, model providers, contract terms, or whether the company is seeing these concerns translate into reduced spending versus shifting procurement priorities. The coverage also did not provide new Palantir customer metrics tied directly to the “token” critique.

For market participants, the near-term watch is whether enterprise buyers and AI providers respond to the criticism in product pricing, data-handling policies, or contract language, and whether Palantir continues to position itself around enterprise control rather than frontier model consumption. The bigger question is whether the industry moves toward architectures and commercial terms that enterprises view as safer and more directly linked to measurable outcomes, or whether token-centric incentives keep dominating how capabilities are sold.