What OpenAI’s Massive Losses Reveal About AI Economics
OpenAI’s recent financial leak highlights that AI subscription pricing and AI inference costs are dangerously misaligned, with the company’s current ChatGPT pricing economics failing to cover the true cost of running large-scale models at heavy usage levels. OpenAI reported USD 13.07 billion (approx. RM60.1 billion) in revenue but a USD 20.92 billion (approx. RM96.2 billion) operating loss, underscoring a business growing fast while spending even faster. Research and development alone consumed USD 19.18 billion (approx. RM88.2 billion), while sales and marketing climbed to USD 5.73 billion (approx. RM26.4 billion). Those numbers explain why the company’s subscription unit economics are under scrutiny. AI inference costs remain high, especially for frontier models, yet users have been encouraged to pay flat monthly fees and use the system freely. That mismatch raises a hard question: can OpenAI scale ChatGPT profitably without either raising prices, limiting usage, or downgrading model access?
The $200 vs $14,000 Problem: When Subscriptions Underprice Compute
The most striking example of broken subscription unit economics comes from SemiAnalysis’ testing of ChatGPT’s top tiers. A USD 200 (approx. RM920) ChatGPT Pro 20x subscription could cost OpenAI around USD 14,000 (approx. RM64,400) in API-equivalent pricing if a user hits its theoretical maximum usage. That gap shows how far ChatGPT pricing economics diverge from AI inference costs when customers treat plans as “all you can eat.” According to SemiAnalysis, OpenAI starts losing money on ChatGPT Plus and ChatGPT Pro 5x once usage exceeds 11.4%, and on its highest tiers the company crosses into negative gross margin at roughly 5.7% utilization. Anthropic’s plans display a similar pattern, though with slightly more headroom. This means it does not take extreme usage for subscriptions to become unprofitable; ordinary power users are enough to push margins underwater.
Rising AI Inference Costs and the Limits of Scaling
Behind these numbers is a structural cost problem. AI inference costs climb sharply as users move from simple prompts to long-horizon coding tasks and agentic workflows that can consume up to 1,000 times more tokens than standard queries. SemiAnalysis notes that Anthropic reaches zero gross margin on its top plans at around 10% utilization, while OpenAI’s threshold is even lower. That sensitivity explains why enterprises are rethinking generous internal access. Some large organizations reportedly pulled back on heavy AI deployment after costs escalated, including one case where a company burned through USD 500 million (approx. RM2.3 billion) in a single month on Claude usage due to weak limits. To cope, many teams now route routine tasks to cheaper models while reserving frontier systems for demanding jobs, cutting costs by as much as 95% in some reported setups.
IPO Ambitions, Regulatory Heat, and Investor Risk
The timing of OpenAI’s financial leak makes its unit economics more than a theoretical concern. The company has confidentially filed IPO paperwork, meaning these losses and AI inference costs will soon be under public market scrutiny. Yahoo Finance, citing the Financial Times, reported that OpenAI’s net loss reached roughly USD 39 billion (approx. RM179.4 billion) including restructuring and non-cash items, though excluding those brought the loss closer to USD 8 billion (approx. RM36.8 billion). At the same time, a 42-state coalition led by New York’s attorney general has subpoenaed OpenAI over ChatGPT’s engagement design, chat memory, and “sycophancy.” Any constraints on how ChatGPT keeps users engaged could reduce usage—and revenue—at a time when AI compute remains expensive. Prospective investors will have to weigh high growth, heavy spending, and mounting regulatory risk against a business model whose subscription unit economics are not yet proven sustainable.
