The AI gold rush looked unstoppable - until investors started asking where the actual revenue is. Now Bloomberg reports a cascading sell-off hitting AI companies, chip makers, and cloud providers as the market questions whether the massive spending on AI infrastructure will ever generate returns proportional to the hype.
This is what a bubble looks like when it starts deflating.
For the past two years, anything with "AI" in the pitch deck went up. Nvidia became one of the most valuable companies on Earth selling chips to train models. Cloud providers like Microsoft, Amazon, and Google spent billions building out AI infrastructure. Startups with barely a product raised at billion-dollar valuations.
The thesis was simple: AI is the next internet. It will transform every industry. The companies building the infrastructure will capture enormous value. Missing out on AI is like missing out on the internet in 1995.
But then reality started asking uncomfortable questions. Where are the profitable use cases? Who's actually paying for AI at scale? Is the revenue real, or are we just watching companies subsidize AI features to claim market share?
Look at the numbers. OpenAI is reportedly losing billions despite charging for ChatGPT. Microsoft is spending enormously on Copilot integrations without clear signs that customers are willing to pay premium prices for AI features. Enterprises are experimenting with AI but not committing to the kind of spending that would justify current valuations.
And the infrastructure spending is staggering. Building out the compute capacity for frontier AI models costs tens of billions. OpenAI announced partnerships for over 30 gigawatts of capacity. That's not a typo - gigawatts. We're talking about power consumption on the scale of small countries.
Investors are now asking: what if this doesn't pay off? What if AI is genuinely useful but not profitable enough to justify the investment? What if we're in a classic infrastructure over-build cycle where everyone spends billions to capture a market that turns out to be much smaller than projected?

