The numbers coming out of the big tech earnings calls are getting harder to justify. Microsoft, Google, Amazon and Meta collectively spent over $200 billion on AI infrastructure last year. That is more than the entire GDP of Hungary. The question is not whether this pace can continue. It is whether it should.
The problem is simple. Supply of AI compute is outstripping demand. Every hyperscaler is building data centres as if they are the only game in town. But the market for AI inference is not growing at the same rate. Most businesses are still experimenting. They are not deploying at scale. The big consumer AI products are losing money on every query. OpenAI reportedly burns through $5 billion a year.
The bubble narrative is easy to overstate. Infrastructure spending is lumpy. Companies build for peak demand, not average. And the long term thesis for AI is still strong. But the short term reality is that capital allocation has become reckless. The hyperscalers are competing on ego, not economics.
When the music stops, and it will, the correction will be painful. The companies that overspent will face write downs. The startups that depend on cheap compute will vanish. The survivors will be the ones that built for efficiency, not hype.
Paul