OpenAI and Anthropic’s AI Subscriptions May Be Too Cheap for Heavy Users

OpenAI and Anthropic may have a pricing problem hiding inside their most popular AI subscriptions. A new analysis from SemiAnalysis, reported by Techspot, suggests that flat monthly plans such as ChatGPT Pro and Claude Max can become expensive for providers when users push them close to their limits.

The headline figure is eye-catching: a $200 ChatGPT Pro 20x subscription could represent as much as $14,000 in API-equivalent usage if a user fully consumed the available allowance. Anthrophic's $200 Claude Max 20x plan was estimated at around $8,000 in token usage under the same API-pricing comparsion.

That does not mean OpenAI is literally spending $14,000 every month on each heavy ChatGPT Pro subscriber. The figure is based on what the same level of usage might cost through standard API pricing. Still, it highlights the difficult economics behind AI subscriptions, where users pay a fixed fee but the provider's compute cost rides with every prompt, file, code task, image request, or agentic workflow.

Why AI subscription margins are under pressure

SemiAnalysis reportedly tested subscription tiers from OpenAI and Anthropic by running long coding and agent-style tasks until weekly usage limits were exhausted. The findings suggest that utilisation matters more than the advertised monthly fee.

According to the analysis, Anthropic can break even on Claude Pro and Claude Max 5x at around 20% utilisation. OpenAI’s margin appears thinner, with ChatGPT Plus and ChatGPT Pro 5x reportedly becoming unprofitable once usage goes above 11.4%. At the highest tiers, the break-even point is even lower, with Anthropic reaching zero gross margin at roughly 10% utilisation and OpenAI crossing into negative territory at around 5.7%.

OpenAI currently offers a $100 Pro tier with 5x higher usage than Plus and a $200 Pro tier with 20x higher usage, according to its own help documentation. Anthropic’s Claude Max plan similarly starts from $100 per month and lets users choose 5x or 20x more usage than Claude Pro.

Agentic AI is making the problem bigger

The issue is not just that more people are using AI. It is that the type of usage is changing.

Simple chatbot queries are relatively contained. Agentic workflows, coding assistants, long research tasks, and multi-hour autonomous sessions can consume far more tokens. TechSpot notes that agentic systems can require significantly more token usage than a standard prompt, which makes flat-rate subscription pricing harder to sustain at scale.

Business users are already reacting to this shift. Business Insider reports that companies have started placing internal caps on AI usage, moving more work to cheaper models, and rethinking whether every task needs a frontier model from OpenAI or Anthropic. Coinbase, for example, has introduced weekly AI spending caps for employees, while other companies are reassessing projects based on return on investment rather than simply encouraging unlimited experimentation.

Companies are turning to cheaper AI models

One likely outcome is more model routing. Instead of sending every task to the most powerful model available, companies can send simpler jobs to cheaper models and reserve advanced models for harder work.

That approach is already gaining traction. Reports cited by TechSpot and Tom’s Hardware suggest that routing tasks between models can cut AI costs by as much as 95% in some cases. Some companies are also looking at open-source or lower-cost models, including DeepSeek, for routine workloads.

This does not mean frontier models are going away. They remain important for complex coding, reasoning, research, and high-value enterprise work. But the “one expensive model for everything” approach is becoming harder to justify, especially when many businesses are still trying to prove the actual productivity gains from AI.

AI prices may fall, but top models will stay expensive

There is some relief on the horizon. Newer AI infrastructure, including Nvidia’s Blackwell systems, is expected to make token generation cheaper over time. Business Insider reported that SemiAnalysis found Blackwell systems could generate tokens far more efficiently than previous-generation Hopper systems, potentially lowering the cost of running many AI models.

However, cheaper infrastructure does not fully solve the problem. As models become more capable, users also ask them to do more. If token prices fall but agentic workloads grow at the same time, providers may still need stricter limits, usage-based pricing, or separate API access for their most expensive capabilities.

That is the bigger takeaway here. AI subscriptions helped make tools like ChatGPT and Claude mainstream because they were simple and predictable. But as heavy users start treating them like full-time coding, research, and automation platforms, the economics become far less comfortable for the companies running them.

For users, the current flat-rate plans may still offer strong value. For OpenAI, Anthropic, and the wider AI industry, the challenge is working out how long that value can be subsidised before pricing changes again.

FAQ

Why could a $200 ChatGPT Pro subscription be worth $14,000 in usage?

SemiAnalysis reportedly calculated that if a user fully consumed the allowance of ChatGPT Pro 20x, the same level of usage could cost around $14,000 if priced through standard API token rates. This is an API-equivalent estimate, not necessarily OpenAI’s direct cost per subscriber.

Is OpenAI losing money on every ChatGPT Pro user?

Not necessarily. The analysis suggests profitability depends on utilisation. Light users are unlikely to be a problem, but very heavy users can consume enough compute to make flat-rate subscriptions less profitable or unprofitable.

How much does Claude Max cost?

Anthropic’s Claude Max plan starts from $100 per month and offers 5x or 20x more usage than Claude Pro. The 20x tier is commonly compared with OpenAI’s $200 ChatGPT Pro tier.

Will AI subscriptions become more expensive?

It is possible, especially for heavy usage and advanced frontier models. Providers may keep consumer subscriptions simple, but shift the most expensive capabilities towards usage-based pricing, stricter limits, or API access.

Written by

Share

Copied