A SemiAnalysis report performed stress tests on major AI subscription plans and concluded that current pricing often fails to cover the compute costs of heavy use. According to the analysis, a $200-per-month ChatGPT Pro plan can consume roughly $14,000 in compute if stressed to API-equivalent rates, while Anthropic’s $200 Claude Max plan can run to about $8,000.
Key figures behind the issue
- The SemiAnalysis stress test shows that intensive users can generate compute costs far above the $200 monthly fee.
- For OpenAI, the report finds the offering becomes loss-making once utilization exceeds roughly 11.4%.
- API revenue, which was supposed to offset flat-rate losses, is shrinking because cheaper open models and other alternatives are pulling customers away.
How business models fell out of sync with usage
The report highlights two simultaneous revenue leaks. Flat-rate subscriptions were originally priced for chatbot-style interactions — one query, one response. Today’s agent workloads and multi-step pipelines can multiply compute demand by as much as 1,000x per user, turning heavy users into loss leaders. The API meter was meant to be the margin engine, but open models and lower-cost options have hollowed out that market, and buyers move to cheaper tokens or providers.
Market moves and concrete examples
SemiAnalysis also points to industry actions that tighten margins:
- Microsoft, Meta and Amazon are capping internal AI spend, reducing large internal demand for external APIs.
- The report notes that Lindy shifted 100% of its traffic to the DeepSeek V4 model, reportedly saving millions of dollars.
Bottom line
This is not an immediate existential crisis for the labs — many have ample capital and user bases — but they have not yet found a pricing or product configuration that turns the most capable models into a consistently profitable business. The mismatch between rising computational cost per user and existing subscription and API pricing means that raising prices risks losing customers, while keeping prices flat widens per-user losses.
Implications
- Current flat-rate and API pricing looks unsustainable if usage patterns continue to shift toward high-compute agent workflows.
- Protecting API revenue requires competitive pricing and meaningful differentiation, or new monetization strategies.
- Labs will need to redesign subscription tiers, metering, or introduce alternative revenue streams to achieve long-term sustainability.
In short, technological capability has outpaced the established economics: the industry has the models and the users, but not yet a reliable way to turn them into profit.
