Recent research and industry analysis have cast a spotlight on the sustainability of flat-rate AI subscription models, such as OpenAI’s ChatGPT Pro and Anthropic’s Claude Max, as heavy users extract far more value than their monthly fees would suggest.
A widely circulated experiment by SemiAnalysis involved purchasing every available subscription plan from OpenAI and Anthropic, then running intensive, long-horizon coding tasks until the weekly usage limits were reached. The findings were striking: a $200/month ChatGPT Pro plan could deliver up to $14,000 worth of API-equivalent compute if fully utilized, while Anthropic’s Claude Max 20x plan could reach about $8,000. Even lower-tier plans, like ChatGPT Plus at $20/month, could provide up to $700 in API value. These figures represent the value at retail API pricing, not the actual operational cost to the providers, but they highlight a significant subsidy for power users.
However, subsequent analysis clarified that the actual compute cost for heavy users is closer to $4,000 per month for OpenAI, rather than the viral $14,000 figure. The higher number reflects the API-equivalent value, which includes margins and overhead, not just raw compute expenses. Still, this corrected figure is many times higher than the subscription fee, and the underlying business model relies on the majority of users consuming far less than their allowance—effectively subsidizing the heavy users, much like gym memberships or wholesale clubs.
This subsidy model is under increasing strain. As more users adopt agentic workflows—where AI agents operate at machine speed and consume vast numbers of tokens—the average usage per subscriber is rising. Both OpenAI and Anthropic have reported lower-than-expected gross margins, with Anthropic even operating at a negative margin in recent years. Large enterprise customers have also reported losses on AI tools priced below cost, prompting some to develop their own models or shift to usage-based pricing.
In response, AI companies are beginning to adjust their offerings. Anthropic, for example, has started limiting access to its most advanced Claude Fable 5 model through metered usage credits rather than unlimited flat-rate plans. Industry observers expect a broader shift toward usage-based or tiered pricing, smarter backend routing to cheaper models for simple tasks, and the reservation of the latest models for API access only.
For now, subscribers to these AI services are enjoying exceptional value, but experts warn that these generous terms are unlikely to last. As the economics of AI provision evolve, users and developers are advised to treat current subscription prices as provisional and to avoid locking themselves into a single provider’s flat-rate plan.
Sources
Internal sources
- A $200 ChatGPT subscription could cost OpenAI $14,000 if you actually used it
- The $14,000 AI Subscription Myth, Corrected
- ‘The math doesn’t work’: Why your $200 AI subscription is secretly worth thousands | Tom’s Guide