View / Is revenue the wrong way to value AI?

Last week, I got called out for a column where I said Anthropic was winning in enterprise over OpenAI, at least based purely on revenue.

Jordan Wilson, who hosts the Everyday AI podcast, called that line “a very SV/Twitter bubble take, not yet backed by real data.” He has a point. It feels like AI is everywhere, but we’re still at an early stage in the race. Traditional financial metrics like revenue and profit don’t tell you much about who wins in the long run. Moreover, OpenAI and Anthropic may end up having very different business models. Anthropic may be more of an enterprise software company and OpenAI a hybrid enterprise-consumer company now, but even that could change. Many of the most important metrics aren’t public.

What I’d like to know is: How many tokens is each company generating on a daily basis? How many of those tokens are monetized and what is the average price per token? What percentage of compute capacity is being utilized at any given time, and for what purpose? Is it training, inference, internal experiments, research?

Right now, Anthropic appears to be revenue-maxxing. Try building something with Claude code on a $20-per-month subscription and, if you’re like me, you’ll quickly run out of your allotted tokens and have to pay extra to finish the job. I’ve found I can go a lot longer on Codex with an OpenAI subscription without hitting those limits. When I do hit the limits on Codex, it’s not as expensive to keep going.

OpenAI could turn the screws on its users, but it hasn’t yet. And because of that, Codex offers a better and less costly user experience for people like me, who are not software developers and probably are not working as efficiently as possible on their projects. (Software developers are a great target market, with roughly 27 million to 47 million of them in the world, but there are a lot more people like me).

When you give “normal” people the ability to easily make customized software applications, the technology quickly becomes indispensable.

If you’re OpenAI or Anthropic, you want to get as many people using your product as possible today so that you’ll have as many customers as possible down the road. One way to do that is to offer lower prices, forgoing some revenue today for more in the future.