The Awkwardness on the Books
The core arithmetic of the article isn't complicated: the leading labs' revenue growth is indeed impressive, but the capital expenditure to train the next-generation model and the marginal cost of inference are soaring in lockstep, and the burn rate keeps running ahead of revenue. Subscription prices have been raised again and again, yet heavy users remain a money-losing proposition; the API is one of the few positive-gross-margin businesses, and it's under continuous price pressure from open source models. The conclusion isn't new, but recompute it every few months with the latest figures and still no one can offer a clear path to profitability.
A Few Possible Exits
The disagreement in the discussion is over how it ends. Optimists analogize to the early Amazon and cloud: burn your way to a monopoly first, and economies of scale will eventually push costs down. Pessimists argue AI has no such lock-in effect—switching models costs users nearly nothing, so burning money buys no moat. The middle camp bets on engineering optimization of inference and new revenue like advertising. The outcome of this debate largely determines how many labs are left in a few years, and whether the service you rely on now will raise prices or shut down.
via: Hacker News