The data center craze has all the fingerprints of a classic bubble. Not a healthy industrial expansion. Not the birth of some carefully reasoned technological backbone for the next century. A bubble. One of those moments in history when money becomes detached from gravity and begins floating through conference rooms in search of fashionable nouns.
Artificial intelligence happens to be the fashionable noun of the hour.
Most of these projects are not data centers in any meaningful sense yet. They are land acquisitions wrapped in PowerPoint presentations. A few artist renderings showing glowing server halls under a dramatic sunset. A stack of pre-purchased chips slowly ageing in warehouses because everybody feared supply bottlenecks more than they feared a collapse in demand. And somewhere in the middle of all this theater sits a business model that does not even trust itself enough to demand long-term contractual commitments from customers.
That part should ring alarm bells loud enough to wake the dead.
If the future demand was truly as inevitable and colossal as advertised, hyperscalers and customers would lock themselves into decade-long agreements without hesitation. Infrastructure businesses built on genuine necessity usually do exactly that. Pipelines do it. LNG projects do it. Rail systems do it. Power stations do it. When something is truly indispensable, customers commit because they fear not having access later.
But here? Capacity reservations often come with escape hatches large enough to fly a cargo plane through. Flexible opt-outs. Shorter horizons. Carefully preserved room to retreat if the numbers stop making sense. Which tells you everything you need to know about the confidence level beneath the marketing varnish.
Looks to me like the market itself suspects that it does not actually need this biblical flood of server farms.
But bubbles are not really about necessity. They are about capital rotation. Project developers always require a fresh narrative to justify raising new money. Yesterday it was green hydrogen. Before that it was metaverse nonsense. Before that it was blockchain salvation. Now it is AI infrastructure. The slogans change. The machinery underneath remains the same tired carnival ride.
Find an investor willing to throw a few million at a “next generation AI campus” and what do you actually receive in return? Usually not much more than a glorified business plan, perhaps an option on future land acquisition, maybe a permitting process somewhere in the early larval stages, and a collection of consultants billing frightening hourly rates while talking about “transformative compute ecosystems.”
Meanwhile the promoter gets something very real immediately.
A high-status position.
Press coverage.
Conference invitations.
Investor dinners.
Executive salaries funded by angel capital.
A social-media profile inflated into orbit.
And above all, very little personal downside.
If the thing implodes later, there will always be explanations ready for deployment. Regulatory uncertainty. Energy constraints. Supply chain disruptions. Unexpected market softness. Insufficient policy support. Geopolitical instability. Pick your favorite excuse from the corporate bingo machine.
The actual financial risk sits elsewhere — usually with investors intoxicated by fear of missing out.
And many of those investors are not exactly behaving like people convinced of a glorious future either. They often spread comparatively small stakes across dozens of speculative ventures, hoping one survives long enough to become the next Nvidia-sized miracle. They feed off hype while carefully avoiding the kind of hard, irreversible multi-billion-dollar commitments that real industrial conviction would require.
That should tell you something too.
Yes, we need data centers. Of course we do. The modern world runs on digital infrastructure and computing demand is growing. But there is a vast gulf between “growing demand” and carpet-bombing continents with speculative server parks as if humanity is about to upload itself into the cloud next Thursday afternoon.
Demand simply is not there for this scale of mania.
And then there is the ugly little issue nobody likes discussing in public: energy.
These facilities are not magical digital clouds floating in cyberspace. They are industrial heat engines. Gigantic electricity consumers that require absurdly stable power flows and equally absurd cooling capabilities. Ideally you build them where power is abundant, reliable, cheap, and politically secure. Which immediately narrows the list of viable locations dramatically.
As a developer, the last thing I would want is dependency on unstable grids, activist-driven energy markets, or governments that cannot even guarantee electricity prices twelve months ahead. Real operators know this. Which is why more and more projects quietly drift toward vertical integration fantasies involving dedicated power generation, private substations, direct gas supply, even nuclear discussions creeping back in through the side door.
Because when your servers stop, your money stops.
And if you already spent billions on chips that depreciate faster than luxury sedans, delays become lethal. Time is not neutral in this business. Time burns capital alive.
Which explains why the narrative machine works overtime. Every bubble eventually reaches the phase where promoters need increasingly grandiose visions to keep confidence elevated while reality struggles to keep pace. We are entering that phase now.
The world does need computing infrastructure. But it does not need every empty field transformed into an “AI innovation corridor” financed by speculative capital and powered by economic hallucinations.
Most gold rushes end the same way. A handful become rich. A handful build something real. And the overwhelming majority leave behind abandoned dreams, worthless paper, and expensive holes in the ground.
This one will not be different.
https://robertbryce.substack.com/p/the-data-center-backlash-is-global