The post is a broad attack on current AI economics. It argues that OpenAI and Anthropic are underpricing access relative to what all-in infrastructure and training costs should imply, that enterprise customers are starting to revolt once token bills become visible, and that the industry has raised so much capital it now needs unrealistic levels of labor replacement to pay for itself. It leans heavily on leaked financials, pricing changes, and recent reporting about companies tightening access to expensive models.
People largely bought the near-term conclusion that AI budgets are getting tighter, especially inside big companies. Several described the same pattern. First came an "AI everywhere" phase with loose access, agent experiments, and little discipline. Then finance stepped in. Access got gated, model choice got centralized, and teams were told to justify use of the top-end models. The practical consensus was not that AI had failed. It was that the experimentation phase is ending and enterprise buyers are now treating tokens like any other metered cloud cost.
The sharper point was that affordability is almost the wrong frame. The real constraint is
ROI. Faster code generation is not the same thing as more profit, and many teams still have no credible way to show that AI-assisted output turns into revenue, margin, or materially lower headcount. A lot of the claimed productivity appears to be extra internal work, nicer tooling, backlog cleanup, and executive theater rather than business impact. That makes demand look price-sensitive. Once CFOs ask what the spend actually bought, usage drops fast.
Commenters also pushed back on the post's accounting. Many thought the article blurred marginal
inference cost, subscription pricing,
API pricing, training expense,
depreciation, and overall company losses. The most common correction was that frontier labs may still have healthy gross margins on API tokens while losing money at the company level because they are spending heavily on training the next models. That does not make the business healthy, but it means "OpenAI is unprofitable" does not automatically prove every token is sold below cost.
Open models and Chinese providers sat underneath much of the conversation as the pressure release valve. Even people who prefer
Claude or top-tier models expect daily work to drift toward cheaper "good enough" models, with expensive frontier APIs reserved for the few tasks that clearly justify them. That points to a likely market shape: quotas on premium models, procurement fights over enterprise access, more local or
open-weight deployments, and a narrowing set of use cases where the best closed models can still command a big premium.