The scale of capital going into AI data centres is unmatched since the early internet boom. But the revenue coming back from these deployments is not matching the hype. The maths doesn't add up for everyone involved.
Shareholders are getting restless. They see the billions flowing to Nvidia and the energy suppliers. They want to see a return. The problem is the cost of inference and training is still high. The use cases that justify this spend are not broad enough yet. A correction is coming.
This doesn't mean AI stops. It means the blank cheque era is over. The hyperscalers will tighten the purse strings. Startups relying on cheap capital for GPU clusters will struggle. The energy deals signed last year will get renegotiated or cancelled. The market is shifting from land grab to efficiency.
The real risk is not a sudden pop. It is a slow bleed of confidence. If the top line revenue growth stalls, the spending cuts will be brutal. The infrastructure buildout was a bet on exponential demand. If demand is merely linear, the whole sector is overbuilt. Paul