The AI bubble won't burst; it will deflate—and here's why
Mikhail T. (Sh0ny)
21 июня 20262 min read
In short
The widespread belief that the AI industry is bound to collapse is based on a false metaphor. In reality, the bubble isn’t bursting; it’s slowly deflating—the financial architecture surrounding the models begins to crack when subsidies run out and businesses demand a real return on investment.
The Bubble Won’t Burst—It Will Deflate
Many expect that one morning the AI market will crash, just like the dot-coms did in the early 2000s. But this view is fundamentally wrong. It’s not “AI in general” that’s collapsing—the financial architecture built around it is deflating. And this is happening slowly, at a different level than the one everyone is focused on.
As long as subscriptions to AI services were generously subsidized by providers, there was no need to ask, “What’s in it for us?” But in the first quarter of 2026, frontier companies switched their corporate clients to token-based payments—the subsidy ran out. The question was asked, and the answer came back empty.
Specific Examples
- Uber burned through its annual token budget in a single quarter. Its COO candidly admitted: it’s impossible to draw a line from impressive metrics to actual delivered value.
- SemiAnalysis laid out the math: with a $200/month subscription, a customer can burn through $8,000–14,000 worth of tokens—meaning the provider is effectively subsidizing your use of the service.
- Meta introduced limits just a couple of weeks after it had encouraged its employees to burn more tokens.
- According to the WSJ, both companies are discussing a sharp price cut for services that are already operating at a loss.
The Funding Scheme Falls Apart
Perhaps the most telling sign is when the “shovel” providers start to have doubts. On June 9, the official AWS account (the very same Amazon that makes money off every token you use) posted: “More AI code doesn’t make a team faster—it might even slow them down.” Six million views. When an infrastructure giant publicly abandons the central tenet of its own marketing—it’s not a slip of the tongue; it’s a demand to “prove the ROI,” coming straight from the top.
The Productivity Myth Is Crumbling
A recent NBER study confirms the math: the number of lines of code has increased, but the number of applications actually shipped has not. The myth of explosive productivity growth is being debunked even among those who sell shovels.
Source: Habr