Palantir CEO Alex Karp Calls Firms ‘Insane’ for AI Token Use

Palantir CEO Alex Karp says AI companies like Anthropic and OpenAI who use the AI token model are “insane,” as costs for both firms continue to rise.

“I’m not throwing shade at them, but something has gone completely wrong,” Alex said in an interview on CNBC, adding that US enterprises are being unproductive and are simply wasting time with the token model.

As AI the cost of AI continues to rise on an industrial scale – and new models are proving pricier than their predecessors – companies are moving on from the trend of “tokenmaxxing” and toward a clearer goal of return on their investments.

Tokenmaxxing refers to the act of employers incentivising employees to use as much AI as possible without considering the consumption impact, a move that can prove financially risky.

As strategies like these become more commonplace and as AI becomes less of a novelty technology, more companies are opting for more open-weight models to perform similar AI tasks at considerably lower costs.

The US-China AI rivalry

Despite the AI push in the US, many Chinese models are also accelerating capabilities, with industry experts suggesting that China could soon catch up with US frontier labs.

Alex says that the AI industry should not underestimate the speed of China’s AI model development.

In a move to support the government’s AI strategy, Palantir announced an expanded partnership with NVIDIA earlier this week, with plans to use the chipmaker’s AI tools to provide US government agencies with bespoke models.

Alex adds that compared to standard, more costly models, open-weight models could be a potential solution for CEOs frustrated by AI labs.

He claims that many AI labs like Anthropic and OpenAI are deliberately misleading corporate partners and the public by trying to oversell the risks of the technology, while at the same time offering powerful models to global companies and governments.

Palantir’s nine-point post on AI sovereignty

Furthering Alex’s views on how companies use AI, at the beginning of this week, Palantir released a point post on X detailing the importance of “AI sovereignty”.

The post urges companies to keep their data in house, rather than outsourcing it to external institutions.

Reflecting Alex’s views on tokenmaxxing, the post also criticises this kind of practice, saying that spending tokens is addictive and only leads to “false progress”.

The nine-point post covers:

• How AI sovereignty dictates a company’s future
• The importance of data retention
• How tokenmaxxing harms value and decreases institutional fortitude
• How companies must control their own weight and knowledge
• How there is no contradiction between sovereignty and alpha
• Why companies shouldn’t politicise technical issues with sovereignty
• The importance of real expertise
• What companies can learn from successful institutions
• Why companies should only listen to institutions with a good track record

Comparing the benefits of AI sovereignty over AI labs in the interview with CNBC, Alex claimed that companies in the latter group “don’t understand how unlikeable they are” and that their products never “actually work the way” customers expect.

In comparison, Alex praised NVIDIA, saying the AI chip maker is effective because they have “control over their compute, their models, their data stack and their alpha”.

He adds: “They want to know they own the means of production. It’s not being transferred to someone else.”