The Subsidised Era Is Ending
"Everyone has been high on the hog using AI effectively for free. The argument is of course that they're profitable on inference , zero proof. None. There's no proof on that."
That is Ed Zitron, who has been calling the AI bubble for two years while almost everyone in the industry called him wrong. He is the author of Better Offline and runs one of the more consistently accurate AI analysis newsletters in circulation.
His current read: the tokenmaxxing era is ending, and the companies that built their revenue projections on subsidised AI usage are about to face a very different market.
Nobody Knows What Anything Costs
The fundamental problem with AI deployment right now: you genuinely do not know how much any task will cost before you run it.
The cost depends on the model, the prompt, the harness being used , Claude Code specifically, some other integration, or direct API access. The same task costs different amounts depending on all of these variables. And because the variables compound, you cannot set a predictable AI budget. You can only observe what you spent after the fact.
This is the situation Uber's CTO walked into. This is the situation Microsoft's fiscal year-end deadline was designed to escape.
The Uber COO, Andrew McDonald, said it directly: "It's really easy to be impressed by AI when you're not paying the cost. But someone eventually has to pay that bill." He was speaking at an event where he publicly acknowledged difficulty justifying the AI budget because there was no direct way to compare it to some sort of outcome.
The Prediction on Anthropic
Zitron's prediction for Anthropic specifically: "I think we are going to start seeing the true value of AI because everyone has been using the subsidised stuff. And I think what is likely is that Anthropic is going to see their revenue burst and then contract violently."
The mechanism: enterprise customers currently consuming AI at subsidised rates , through deals structured for growth rather than sustainability , will hit their budget limits. The Uber scenario will repeat across companies. Some will restructure their AI deployments. Some will cancel licences. Some will route through cheaper alternatives.
None of this invalidates AI as a technology. It clarifies that AI as a business model has been running on subsidised usage and investor capital rather than genuine unit economics.
What Comes After the Bill
Every major technology transition has a subsidised adoption phase followed by a pricing reality phase. The internet had it. Cloud computing had it. The subsidised phase drives adoption. The pricing reality phase drives consolidation, efficiency, and the winnowing of companies that cannot survive sustainable economics.
The AI pricing reality phase is not a collapse. It is a correction. The companies that built real workflows producing real value at real cost will survive it. The companies that built leaderboards rewarding token consumption will not.
The bill has arrived for Uber. It is arriving everywhere else on a slightly different schedule.