Think of AI intelligence like oil. A barrel of crude has a global price, and producers compete ruthlessly on cost. Now apply that logic to AI tokens, and you start to see why Chamath Palihapitiya is raising alarms about the economics underpinning the industry’s most celebrated companies.

On a recent episode of the All-In Podcast, Palihapitiya introduced a framework he calls a “barrel of intelligence,” defined as one million tokens. At OpenAI, that barrel costs about $26. At Anthropic, the price is roughly similar. At Chinese AI providers, it’s $0.50. That’s not a typo.

The price gap nobody wants to talk about

OpenAI and Anthropic sit at the premium end, charging around $26 per million tokens. xAI comes in at roughly $1 per barrel. Google prices similarly at about $1. Meta lands around $1.50. And Chinese providers undercut everyone at approximately $0.50.

In English: you could buy 52 barrels of Chinese AI intelligence for the cost of one barrel from OpenAI.

The 45-day doubling problem

According to internal data he referenced, the cost of tokens is doubling approximately every 45 days. Meanwhile, the productivity gains that enterprises are actually seeing from AI integration hover around just 5%.

This dynamic is especially relevant as both OpenAI and Anthropic are reportedly preparing for IPOs. Investors will want to see sustainable unit economics, and a world where customers can get comparable intelligence at a fraction of the cost makes that story harder to tell.

What this means for the AI market

Palihapitiya didn’t frame this as a prediction of doom for any specific company. His point was more structural: the market for AI intelligence is behaving like a commodity market, and companies that price their product at 52 times the competition need an extraordinarily compelling reason for customers to keep paying the premium. Right now, the 5% productivity gain figure suggests that reason might not exist.

For traders and investors, the watchlist items are straightforward. Monitor how aggressively OpenAI and Anthropic adjust pricing ahead of their respective IPOs. Track whether enterprise customers begin meaningfully shifting workloads to lower-cost providers. And pay attention to the 45-day cost doubling trend, because if that continues without a corresponding jump in productivity, the economics of premium AI become very difficult to defend.