Rohan Paul, a user on social platform X, reposted an interview with venture capital tycoon Chamath Palihapitiya on CNBC's Squawk Box, discussing the current pricing landscape of the AI industry.

Right now, AI companies are both earning and burning massive amounts of money, but most customers still haven't figured out how to convert this technology into actual tangible profits.

In the interview, Chamath put forward a very intuitive analogy: he defines 1 million tokens as 'one barrel of intelligence', and current pricing across different providers is wildly inconsistent:

- OpenAI sells one barrel for $26

- Anthropic's latest model sells one barrel for $56

- Elon Musk's sells for $1 per barrel

- Zuckerberg's will soon sell for $1.5 per barrel

- DeepMind's Demis and Google's Sundar are also selling at $1 per barrel

- Chinese vendors sell for $0.5 per barrel

He stated that this pricing system will eventually return to rationality. For the same basic service, there is currently a 100x difference in price. If you bet early on companies that sell high-priced 'barrels' and forcefully pass the costs onto customers, you will most likely end up in trouble. This correction process will be completed by the market itself.

The full interview is posted in the comment section of the original post; only a clip is shared here. Those interested can find the full recording through the original post.

After this statement was released, netizens have offered varying takes:

Some netizens agree with this logic, saying that after computing costs come down, real-world implementation will instead depend more on a team's business capabilities.

Others push back, arguing that AI tokens are not the same as crude oil. Crude oil comes in different grades, and different large models also have vastly different capabilities. Ultimately, pricing depends on how much value is created for the customer. The large price difference right now is essentially a reflection of the large capability gap.

A third group takes a neutral stance, pointing out that AI hasn't reached the stage of full commoditization yet, but different tasks require different levels of model capability. Going forward, companies will definitely allocate different tasks to different tiers of models to drastically optimize costs.

Some netizens also directly pointed out that Chamath's statement certainly carries his own vested interests. It's not unreasonable for OpenAI and Anthropic to charge higher prices—their models truly do deliver stronger performance.

Many industry practitioners have already started testing tiered model allocation: simple tasks are handled by small models, and only demanding complex tasks are routed to top-tier large models. This is indeed the most straightforward approach to cut costs right now. But the industry changes so fast that no one can say for sure what the price of this 'barrel of intelligence' will look like a year from now.

What's your take? Do you currently allocate different tasks to different models in tiers?