Research firm SemiAnalysis put Anthropic and OpenAI’s subscription plans through a stress test, and the results are brutal. They found that heavy users — think high-frequency coding and other long-running tasks — can burn through weekly usage caps fast, sending the companies’ gross margins into negative territory. For example, a customer on the $200/month ChatGPT Pro plan (with 20x usage limits) can consume up to $14,000 worth of API tokens. Meanwhile, the equally priced Claude Max subscription can burn through $8,000 worth of tokens.
https://twitter.com/SemiAnalysis_/status/2064815042374074396
On the margin front, if you assume the companies make a 75% gross margin on API list prices, the actual usage rate of subscribers directly determines whether the subscription business is profitable. For ChatGPT Pro at $200/month, once average usage exceeds 5.7%, margins turn negative. At 100% usage, OpenAI would be looking at a gross margin of -1650%. For Claude Max, margins go underwater past 10% usage, and hit -900% at full usage.
Even the cheaper $20/month plans aren’t safe. ChatGPT Plus can consume up to $700 worth of tokens, and Claude Pro up to $400. Their breakeven usage rates are 11.4% and 20%, respectively. Since cutting usage caps outright would spark a PR disaster, SemiAnalysis predicts AI companies will instead start funneling new models or advanced features exclusively through API channels to insulate themselves from the high costs.
This aligns with what Anthropic already did with its latest flagship, Claude Fable 5. Subscribers got a free trial until June 22, but starting June 23, they’ll need to pay extra using consumption credits until computing capacity catches up and the model can be folded back into the regular subscription.
