It seems to be based on faulty math, or ignorance…
About of 80% of Anthropic’s revenue comes from pay-per-token API calls by its enterprise customers—established companies, start-ups, and cloud resellers—not subscription plans. OpenAI is more dependent on consumer subscriptions, but it recently shifted at least some of its focus to enterprise customers as well, because that’s where the money is. Both companies have significantly improved their margins on serving inference tokens, especially for their lower-tier models. They’re still shoveling money into the maw of infrastructure and training, however—the big question is when (if?) their inference revenue will cover their training and infrastructure spend.
Since neither company is public, it’s hard to source reliable information about the composition of their revenue and expenses. For instance, the 80% figure I mentioned above pops up on a number of websites, all of which appear to be getting their information from the same source, a private market research firm named Sacra. How reliable is it? At $50 per month per seat, one hopes they’re plugged into reasonably robust data sources …