The real news out of the AI industry this month is not a smarter chatbot. It is how OpenAI and Anthropic are choosing to charge for the ones they already have.
In April, Anthropic changed how it bills its biggest business customers. Instead of a flat monthly fee per employee seat, large customers started paying based on how much AI they actually use, similar to an electricity meter.
Revenue followed fast. Anthropic's run rate, the yearly pace its revenue is tracking toward, went from about 9 billion dollars at the end of last year to 65 billion dollars by late July.
OpenAI answered with a price cut instead of a price increase. In July it slashed the price of Luna, its cheapest AI model, by 80 percent, alongside a smaller cut to a second model called Terra.
OpenAI's own run rate is now approaching 70 billion dollars, with business customer revenue more than doubling since July. For one quarter this year, Anthropic actually out earned OpenAI, a first, before OpenAI's price cut helped close the gap.
Neither company is winning this by building a better product. They are winning by finding the right price for each type of buyer: cheap access for smaller companies and developers, metered billing for large enterprise accounts. That is a sales and pricing strategy, not a technology breakthrough.
A leaked copy of Anthropic's paperwork for its upcoming stock market listing adds an important detail most coverage of the revenue numbers leaves out. The company lost about 42 billion dollars last year, though most of that is a paper loss tied to how its investor funding is structured rather than cash leaving the bank.
Anthropic is also asking investors to fund roughly 518 billion dollars in future computing and data center costs. That is an enormous bill that has to be paid before any of this growth turns into real profit.
The filing also shows how fragile the customer base still is. Nearly a quarter of Anthropic's revenue last year came from just two customers, Amazon and Google, and neither is locked into a long term contract.
Separately, research from the spend management firm Ramp found something telling about the whole industry. At both OpenAI and Anthropic, the top 1 percent of customers generate 80 percent of all revenue. Most of the money is still coming from a small number of large, AI native companies, not from traditional businesses buying seats for their staff.
For any company shopping for AI tools right now, there are two practical takeaways. Stop assuming your AI bill will look like a normal software subscription: usage based pricing means your costs can rise sharply if your team's use of the tool grows, so budget for a range, not a fixed number.
Also remember that today's low prices are being subsidized by companies racing for market share while losing money, not by a stable, profitable business. That makes them great deals to lock in now, and a poor basis for a five year financial plan.