AI Subscriptions: Steal or Unsustainable? Analysis

TL;DR: AI subscriptions like ChatGPT Plus and Claude Pro are a steal for intensive users, but the business model relies on most users not maxing out the service. If 25% of subscribers exploited the limits, companies would lose money, potentially leading to price changes or restrictions.

What happened?

Analytics firm SemiAnalysis conducted a revealing experiment: subscribing to Anthropic's (Claude Pro) and OpenAI's (ChatGPT Plus) paid plans to push them to the limit and compare the real cost with the pay-per-use API. The results, shared on X, show that with a $200 monthly subscription, you can consume the equivalent of between $8,000 and $14,000 in API tokens. That's a saving of up to 70 times the price paid.

Why is this important?

This finding questions the sustainability of the flat-rate subscription model for AI services. Like gyms or all-you-can-eat buffets, companies rely on most users not fully exploiting the service. Analyst Ed Zitron warns that if 25% of subscribers decided to intensively use the limits, profit margins would turn negative. This could lead companies to tighten restrictions or raise prices, affecting all users.

Consequences for the market

For light users, the subscription remains a steal: they pay $200 for a service they barely use. But for power users, the risk is that companies will adjust limits or eliminate the flat rate. Additionally, this model encourages inefficient use of computational resources, which could have environmental and operational cost implications. In the long term, we are likely to see clearer segmentation between plans for occasional use and plans for intensive use, with prices more aligned to actual consumption.

What readers should know

If you use AI intensively (programming, data analysis, complex tasks), the flat-rate subscription is very cost-effective.
If you use it for sporadic queries, you might be benefiting from a cross-subsidy from power users.
Companies are incurring losses on intensive users, and they are likely to adjust prices or limits in the future.
Always compare the cost of the subscription with the API if your usage is very high; you could save even more.
This model recalls the early days of SaaS, where flat prices were common and later migrated to usage-based models.

In-depth analysis

SemiAnalysis's experiment not only reveals the magnitude of the subsidy but also exposes an aggressive pricing strategy by OpenAI and Anthropic to gain market share. By offering such high perceived value, they attract developers and businesses who later become pay-per-use API customers. It's a tactic similar to 'loss leaders' in retail: selling below cost to capture customers. However, the scalability of AI makes these losses potentially much larger than in other sectors.

"The flat-rate subscription model for AI is a double-edged sword: it attracts users but can be unsustainable if the risk of overuse is not managed well." — TheVortiq

Moreover, there is a consumer psychology component: users who pay a fixed fee tend to use the service more to 'get their money's worth,' accelerating the wear on limits. Companies respond with restrictions like message limits per hour or day, but as the study shows, these limits can be circumvented with strategic use.

In the future, we are likely to see hybrid models: a low base fee plus a variable cost for intensive use. There are already signs of this with OpenAI's 'Pro' plans offering priority access. The key will be finding a balance that allows companies to be profitable without alienating users.