It’s easy to see the big-headlines version of the AI boom: enormous valuations, models released every couple months, companies promising “agentic AI” that will revolutionize everything. What you don’t always hear about is the backbone supporting it: an overwhelming surge in power needs, data-center buildouts and utility forecasts that strain belief. These are the deep pipes of the hype machine, and they well may be among the first places where cracks show.
Here’s what’s going on. Throughout the United States, dozens of utilities are fielding requests for electrical capacity from data center developers who say they will need hundreds of gigawatts of new power to supply the surging demand brought on by the next wave of artificial intelligence. Some forecasts show utilities anticipating as much as 711 gigawatts of new demand across their systems in the coming years — about equal to the peak load on the entire American summer grid. But there’s a catch: GPU producers, the real horses in AI compute, project that over the next several years they can only accommodate on the order of single-digit gigawatts of new load — possibly something like 9.5 gigawatts by 2028. So the apparent AI load is either massively under-estimated in some places, or there is heavy speculation building into forecasts.
Why does this matter? Utility infrastructure is highly costly, slow to build and paid for over the course of many decades through fees imposed on consumers or industry. When that demand doesn’t materialize, the result is either utilities left with excess capacity (and higher bills) or else they under-invest and threaten reliability. Either way, it will ripple through the economy. The AI hysteria can create “demand” for the data centers and compute — but every few levels of abstraction, the physical world (buildings, lines to the grid, air conditioning capacity, motherboards) doesn’t always live up to that shiny sales pitch at your corporate HQ.
More troubling signs: Even those in the industry are beginning to sound alarms. Former Intel CEO Pat Gelsinger was even more blunt: “Of course we’re in an AI bubble,” he said, although he doesn’t think it’s going to burst just yet. A leading UK tech investor highlighted “disturbing signs” that AI companies are over-inflated in value. And even inside big tech, Google’s Sundar Pichai complained about “irrationality” in AI spending. The pattern: hype accelerates beyond business reality, infrastructure lags behind more than predicted.
So what should you watch if you care about this? A few things:
- Track those utility filings: Are they securing real commitments from data-center developers, or are they merely warehousing speculative “options” for future load? With the latter, risk is higher.
- Keep an eye on watch chip manufacturing vs demand: If there is a shortage of GPUs or accelerators, many of these promised data-centers may never operate at full advertised capacity.
- What about rate-payers: will utility customers be forced to subsidise speculative builds? Legal/regulatory push-back is already beginning in states such as Ohio.
- Watch valuations: Just because a company puts together an Ai model or a data-center doesn’t mean they ensure long-term monetisation. Some of the largest anticipated paydays could also ride on infrastructure that is still speculative.
Absent is the flash of hovering drones or “AGI soon” headlines. But there may be real-world consequences as well. If parts of the A.I. infrastructure gambit don’t pan out, the consequences could show up not only in your tech stocks but also in your utility bill, regional economic growth and how many of all those A.I. dreams actually materialize. The models are sexy. The hardware is messy. And the bubble you’re not hearing about could be it.
In other words: Yes it does © Provided by CNBC Qi Liu programming an industrial robot to perform complex tasks Imprent 7, a software platform for manufacturing that includes AI In short: Yes AI has great promise What can be done about it? Yes — great things can come. But just because the hype is loud doesn’t mean the economics are rock-solid everywhere. Sometimes the most terrifying bubbles are those you can’t see.
