A recent analysis by SemiAnalysis reveals that AI subscriptions like the $200 ChatGPT Pro plan offer users compute value far exceeding their cost—up to 70 times more—due to fixed fees subsidizing heavy usage by power users. However, this subsidy model is becoming unsustainable, prompting AI companies to shift toward usage-based pricing and tiered plans, meaning current subscribers should take advantage of the exceptional value while it lasts.
A recent analysis by research firm SemiAnalysis reveals that AI subscription plans, such as the $200 monthly ChatGPT Pro, offer far more value than their sticker price suggests, especially for heavy users. By pushing these plans to their usage limits through intensive tasks like agent-style coding, SemiAnalysis found that the actual compute value accessed could be up to 70 times the subscription cost. For example, a maxed-out $200 ChatGPT Pro plan could equate to around $14,000 worth of API usage, while Anthropic’s Claude top tier could represent about $8,000, highlighting a significant subsidy for power users.
This discrepancy arises because AI usage is measured in tokens—small chunks of text processed by the model—and complex tasks consume vast amounts of these tokens. While casual users might only use a few hundred tokens per query, power users running multi-step, agent-style workflows generate enormous token volumes, driving up the computational cost behind the scenes. Despite this, subscription fees remain fixed, meaning providers often subsidize the heavy usage of these customers, absorbing costs that far exceed the subscription revenue.
AI companies tolerate this imbalance because most subscribers do not approach these high usage thresholds. According to SemiAnalysis, OpenAI starts losing money on cheaper plans once a user exceeds about 11.4% of the allowance, and on the $200 plan, the break-even point is around 5.7%. Anthropic fares slightly better, breaking even at roughly 20% usage on lower tiers and 10% on the highest. This means that only a small fraction of users push the services into unprofitable territory, allowing companies to offer generous flat-rate subscriptions to attract and retain customers.
However, signs indicate that this subsidy model may be unsustainable in the long term. Anthropic’s recent move to limit access to its high-end Claude Fable 5 model through metered usage credits rather than flat-rate plans exemplifies the financial strain of offering premium AI capabilities at a fixed price. The high computational costs of advanced models, which require expensive data and complex reasoning, make it challenging to maintain unlimited access without shifting toward usage-based pricing or tiered offerings.
For subscribers, the current situation means they are enjoying exceptional value, but changes are likely ahead. Users can expect smarter backend routing that directs simpler requests to cheaper models, usage-based pricing for cutting-edge features, and tiered plans that reserve the best AI capabilities for higher-paying customers. While the era of unlimited frontier AI at flat monthly fees isn’t over yet, the economics suggest that locking in current subscription benefits while they last is a prudent approach.