Every AI chatbot answer, image and video you generate starts life in a building most people never see: a data centre packed with power-hungry chips, industrial cooling and backup generators. Those buildings are being constructed at a staggering pace — and, according to fresh industry analysis this week, the maths behind them is starting to look uncomfortable. Forecast infrastructure spending is set to outstrip the revenue AI is expected to bring in, raising a simple question for everyday users: who eventually pays?
The Numbers Behind the AI Building Boom
Reporting by The Register on the widening AI infrastructure gap highlights forecasts in which the cost of building and running AI data centres grows far faster than the money customers are currently paying for AI services. One widely shared estimate put it in human terms: paying for AI’s data centres at prevailing prices would take the equivalent output of far more workers than the economy actually has. However you slice the arithmetic, the direction is the same — spending first, revenue later, and a gap in between.
Why is the bill so large? Modern AI models are trained and run on specialised graphics processors that cost tens of thousands of dollars each, draw enormous amounts of electricity, and become outdated within a few years. A single large facility can need as much power as a small city, plus land, water for cooling, high-capacity network links and fleets of engineers. Unlike a normal office building, a data centre starts ageing the day it opens.
Why This Matters If You Never Buy a Server
It is tempting to treat this as a problem only for technology giants and their investors. It is not. When the cost of providing a service rises faster than what users pay for it, companies have a familiar set of levers — and most of them reach your pocket or your screen.
The first lever is price. Free tiers shrink, usage limits tighten, and features that were once included move behind subscriptions. The second is quality: a chatbot may quietly use a smaller, cheaper model for free accounts. The third is advertising and data, as companies look for new income inside products that were previously clean. If you have already noticed AI features appearing inside apps you never asked to change, you have seen this pressure at work. Our look at free tools that replace expensive software subscriptions is a useful hedge while prices are in flux.
Energy, Water and Where Data Centres Get Built
Money is not the only constraint. The International Energy Agency’s work on data centres and energy tracks how quickly electricity demand from these facilities is growing, and grid connections have become a bottleneck in many regions. Communities increasingly ask what a new facility brings them locally — jobs are relatively few once construction ends — in exchange for the power and water it consumes. Delays, planning disputes and power prices can all slow projects, which is another reason forecasts keep shifting.
That physical reality also explains the industry’s push for efficiency: better chips, smarter cooling, software that squeezes more work from each server, and models that do more with less computing power. Every efficiency gain lowers the cost of each AI answer — and therefore how much of it companies can afford to give away.
Is This a Bubble, or Just Growing Pains?
Sceptics point to earlier technology booms, when companies laid far more fibre-optic cable than the internet of the day needed. Much of that capacity eventually found users — but many of the firms that paid for it did not survive to benefit. Optimists reply that demand for AI is real, growing across business, science and everyday life, and that revenue routinely lags infrastructure in any network industry, from railways to broadband.
Both can be partly right. The safest reading for ordinary users is not to bet your workflow on any single free service staying free forever. Export your important work, learn the underlying skill rather than only the prompt, and keep a second tool in reserve — a habit we also recommended in our guide to the future of digital marketing in 2026, where AI costs are already reshaping how small teams work.
What to Watch Next
Three signals will show how this story ends. First, earnings reports: are paying customers, especially businesses, spending enough to narrow the gap? Second, prices: do subscriptions and free-tier limits keep tightening through 2026? Third, efficiency: if newer chips and models cut the cost per answer sharply, today’s alarming forecasts may age poorly.
For now, enjoy the tools — this is a golden age of cheap, powerful AI for ordinary users — but enjoy them with open eyes. The buildings, chips and electricity behind every clever answer have to be paid for by someone. Increasingly, the industry’s spreadsheets suggest that someone will, one way or another, be all of us.