Palantir’s Alex Karp likens excessive Artificial Intelligence usage by companies to ‘porn addiction’

Artificial Intelligence (Ai) is rapidly reshaping operations across industries, but Alex Karp, chief executive of Palantir Technologies, believes many organizations are becoming too fixated on AI adoption without proving that it delivers meaningful business results.

During a live conversation with TBPN at Palantir’s AIPCon 10 event, Karp likened the excessive pursuit of AI usage to “porn addiction.” He took aim at what he called “tokenmaxxing” — the growing tendency among companies to prioritize higher AI consumption and computing power while paying less attention to whether those investments generate tangible value.

“People are just sitting there all day, kind of like a porn addiction,” he said while discussing internal AI usage metrics. Karp argued that organisations have prioritised increasing token consumption over assessing tangible value delivered.

In the AI ecosystem, tokens serve as the basic units that large language models use to process and generate content. Since most AI providers charge customers based on token usage, the metric has become an important benchmark for both vendors and enterprises.

Karp’s remarks echoed concerns previously raised by Shyam Sankar, chief technology officer at Palantir, during the company’s earnings call last month. Sankar described Palantir as operating in “a no slop zone,” arguing that lower AI costs and greater token consumption do not automatically lead to stronger business outcomes.

"More tokens means more slop, and the more commodity cognition you consume, the more you need a system that can prevent the economic harm, so you can harness the economic value," Sankar said, as quoted by Business Insider.

According to Karp, AI models are highly capable when it comes to handling specific tasks, such as generating an analysis of China’s economic growth. However, he noted that more complicated challenges involving supply chains, industrial systems, military operations, manufacturing, and the oil and gas sector require ongoing decision-making and carefully designed workflows that extend far beyond chatbot-style interactions.

"They are enhanced by large language models. They are not replaced by large language models," Karp said.

In such environments, Karp said, large language models can enhance human productivity and support decision-making, but they cannot replace the operational frameworks that businesses rely on.

He added that advanced AI capabilities are likely to become increasingly available and commoditized over time. The real challenge for organizations, he argued, will not be gaining access to AI tools but determining which problems are worth solving and where the technology can create the greatest impact.

According to Bloomberg, retail giant Walmart Inc. recently placed caps on how much employees can use its internal AI assistant. Similarly, Uber Technologies Inc. has introduced a $1,500 monthly spending limit per employee for specific AI coding tools. The ride-hailing firm implemented the restriction after entirely exhausting its annual budget for Claude Code, a popular assistant developed by Anthropic PBC.

This trend is playing out across multiple industries. Organizations are consuming "tokens"—the standard metric for AI processing power—at such a rapid pace that initial, unrestricted experimentation is being rolled back. This budget-conscious rationing stands in stark contrast to previous corporate mandates that demanded workers aggressively adopt the technology. Companies like Accenture Plc and Coinbase Global Inc. previously warned employees that resisting AI could derail their career advancement or cost them their jobs. Meanwhile, Starbucks Corp. tied 25% of its tech workers' bonuses to department-wide AI adoption targets.

The sheer scale of this consumption was highlighted by Alphabet Inc. CEO Sundar Pichai at the Google I/O conference on May 19. He revealed that monthly token usage across Google's AI platforms had surged sevenfold over the past year, reaching 3.2 quadrillion tokens. Pichai remarked that many enterprises had already drained their yearly token budgets before the end of May.