Too Many Data Centers...

In classic tech history, every major breakthrough—from railroads to the internet—eventually overbuilds, overpromises, and leaves a trail of empty warehouses and bankrupt investors. AI is real, impressive, and already changing the world. But the current race to construct giant glowing data centers the size of small cities, consuming city-scale electricity and lake-scale water, looks a lot like those earlier booms. Wall Street cannot build them fast enough. Today we climb the ladder and break down 10 reasons the AI data-center surge may be expanding faster than sustainable demand.

#10: Everybody Is Building at the Same TimeMicrosoft, Google, Amazon, Meta, Oracle, OpenAI, CoreWeave, and countless private firms are all racing to add the exact same capacity at once. Competition is normally healthy—unless everyone is building for a future that never fully arrives. The railroad boom, telecom boom, fiber-optic boom, and internet itself transformed the world, yet thousands of companies never survived long enough to enjoy the success. Technology can succeed while individual investments fail. People confuse those two things constantly. It’s like every neighbor showing up to the potluck with potato salad; eventually someone realizes dessert was the missing piece.

#9: AI Gets More Efficient Every YearToday’s models demand enormous computing power. Tomorrow’s almost certainly will not. Engineers continually deliver smaller models, better software, more efficient chips, improved compression, and smarter algorithms. Every generation of computers has followed this path: room-sized machines shrank into smartphones, and a USB drive once required an entire room. Moore’s Law has slowed but still points to roughly doubling power every few years. Spending billions on capacity sized for today’s technology risks stranding assets when the same work needs half as many servers three years later. Progress is real—and it usually gets smaller, cheaper, and faster.

#8: Data Centers Could Become CommoditiesInvestors dislike hearing this, but eventually everything becomes ordinary. Cloud storage, internet bandwidth, and server hosting all went from exciting innovations to price-driven commodities. When everybody builds the identical product, margins shrink. The first movers often earn extraordinary returns; the tenth company does not. Competition drives prices down until the nicest barista (or the fanciest cooling system) no longer matters if the quality is the same. At that point, price wins.

#7: Governments May Stop Rolling Out the Red CarpetCities and states currently compete aggressively with tax breaks, free land, and infrastructure upgrades to attract data centers. Everyone wants the jobs and investment—until residents start asking why electric bills are rising, why there is constant noise, why tax incentives exist, and why a giant warehouse consumes enough power for tens or hundreds of thousands of homes while using significant water. Some projects advance; others face growing opposition. Large infrastructure projects often do. Companies are already becoming more tactical as pushback mounts. People generally support technology; they simply do not always want the industrial version in their backyard. Some facilities have even begun construction before locals fully realized the scale.

#6: The Reliability Trap (Concentration Risk)Optimists rarely mention this: concentrating a massive share of global computing power into a few dozen interconnected mega-campuses creates single points of failure. One bad hurricane, regional grid stumble, or cooling-supply disruption no longer produces a minor outage—it risks systemic collapse of AI services. Super-campuses are efficient, yet they trade local redundancy for centralized scale. The classic trade-off holds: the faster and bigger it grows, the harder it can fall. Call it the Egg-Basket Theory.

#5: The Power… and the WaterThese buildings do not merely use “a lot” of electricity; some request hundreds of megawatts or even gigawatts—enough to rival a small city’s consumption. Cooling adds further demand. Depending on design and climate, facilities may consume millions of gallons of water over time (some recycle heavily, others lean on air cooling). Multiply that by twenty campuses in the same region and the strain becomes obvious. Someone must still build the substations, transmission lines, power plants, and pipelines. That infrastructure is not free, and the next power bill has to land somewhere. Plugging an extension cord into the wall does not magically produce artificial intelligence; the supporting grid and generation must exist first.

#4: Wall Street Has Done This BeforeHistory does not say technology fails. History says investors sometimes lose their minds. Railroads changed America; hundreds of railroad companies vanished. The internet transformed the world; thousands of internet companies disappeared. Fiber optics revolutionized communications, yet the late-1990s boom left vast amounts of capacity underused for years after the bust. Housing, solar, canals, and telegraphs followed similar patterns. Every major innovation creates winners—and also speculation. People hear “AI is changing the world” and immediately conclude every company building AI infrastructure must be a great investment. Those are two completely different statements. The invention survives; many investors do not.

#3: Do We Actually Need This Much AI?This may be the simplest unanswered question. How much AI do people actually want in daily life? Search engines, email, word processors, photo editors, cars, refrigerators, and soon perhaps toasters are all gaining AI features. Some prove genuinely useful; many are barely touched after the first week. Nobody knows what everyday demand looks like in five or ten years. Will most people use AI all day, or will they treat it like voice assistants—occasionally, then forget it exists? Building capacity for every possible future is expensive, especially if the future turns out smaller. Asking AI to organize one’s life and receiving “get a daily planner” is not exactly revolutionary demand.

#2: The Business Model Is Still Being InventedAI companies are spending staggering sums on GPUs, electricity, engineers, networking, construction, cooling, power plants, and cranes. Yet the revenue side remains uncertain. Millions use AI, often for free, on the assumption that paid subscriptions will eventually cover costs. A $20-per-month ChatGPT-style plan sounds promising until you examine the token economics: users can consume far more than $20 worth of compute (sometimes hundreds of dollars) while businesses receive deep discounts. If the base model loses money on paying customers, scaling the losses does not create a durable business. Spending hundreds of billions to generate modest monthly subscription revenue works only if efficiency improves dramatically—and that is far from guaranteed.

#1: The AI Might Come HomeThe biggest wildcard is that a growing share of AI may never need the giant buildings. Phones already carry dedicated AI processors; laptops, desktops, and even some cameras have AI chips. Every generation gets faster, smarter, and more efficient. Many routine tasks already run locally. Giant data centers will not disappear—training frontier models and supporting heavy enterprise workloads will still require centralized power. But if millions of everyday requests never leave the user’s device, that is a fundamentally different future than the one many investors currently assume. Asking “my computer” instead of “the cloud” means no internet latency, no distant warehouse, and no billion-dollar server farm for every simple query. Large language models are moving toward the user, and that puts a dent in data-center demand.

Final ThoughtsThis is not an anti-AI argument; it is a hardware and investment debate about where the calculations will actually run. Data centers will remain essential, but they are more likely to become partners than the sole drivers of the industry. History shows that large investment cycles of this kind almost always end badly before the long-term victors emerge from the ashes. Do not confuse a powerful technology with a guaranteed investment. People have been right about the technology and wrong about the timing countless times before.

What do you think? Are we building exactly the right number of AI data centers, or are we heading toward another technology boom-and-bust cycle? Drop your thoughts in the comments.