It begins quietly, as many stories do, in a small rural town where the horizon seems impossibly broad. The town planning commission gathers in a modest room, the air thick with the scent of burnt coffee and aged carpet, to hear that their town will soon win the modern economy: 10 new data centers within the town’s boundaries. Not just one or two, but 10. The PowerPoint presentations shine with promises: construction jobs, some permanent positions, “community investment,” and a new tax base that will “transform the region.”

Sure, there will be jobs. But not the jobs that rebuild a town’s soul. Data centers don’t employ thousands once they’re up; they employ dozens, sometimes fewer, depending on how automated the operation is. The real impact isn’t people—it’s power, land, transmission capacity, and water. When you drop 10 massive facilities into a small grid, demand spikes don’t just happen inside the fence line. They ripple outward. Utilities must upgrade substations, reinforce transmission lines, procure new-generation equipment, and finance these investments. Guess who ends up paying a meaningful portion of that over time? Local ratepayers, in one form or another, will face higher bills or the quiet deferral of other infrastructure work.

Water is often the second shoe to drop. Even when operators insist they’re “water efficient,” cooling is cooling, and cooling at scale is never free. Some facilities will use evaporative systems; some will use closed-loop systems; some will promise innovation that appears impressive in a press release. Meanwhile, the town’s farmers now watch the aquifer levels and the weather forecast with equal anxiety, except now they’re competing with an industry whose thirst is measured in engineering diagrams, not drought stories.

This is what the data center boom looks like on the ground: a glossy promise wrapped around very physical constraints.

The new religion of capital spending

Here’s the part we don’t say out loud often enough: High-tech companies are spending massive amounts of money on data centers because the market rewards them for doing so. Capital expenditures have become a kind of corporate signaling mechanism. On earnings calls, “We’re investing aggressively” has become synonymous with “We’re winning,” even when the investment is built on forecasts that are, at best, optimistic and, at worst, indistinguishable from wishful thinking.

Cloud providers are leading this push because they can. They have scale, cash flow, and a narrative that Wall Street loves: Demand is exploding, supply is scarce, and only the giants can build fast enough to capture the upside. Once that story takes hold, the spending itself becomes a proof point. The bigger the number, the more serious the company appears. The more serious the company seems, the more investors assume the company must know something they don’t. It’s a self-reinforcing loop that looks like confidence and behaves like momentum.

Then layer on the multi-billion-dollar tech-to-tech deals we’ve watched pile up during the past few years: hardware commitments, long-term GPU supply arrangements, “strategic partnerships,” capacity reservations, and entire ecosystems of vendors jockeying to be inside the tent. These deals aren’t just operational decisions; they’re market theater. They give executives something concrete to point to, something that sounds like inevitability. If Company A is signing a 10-figure commitment with Company B, then surely the future is already here, and you’d better buy a ticket.

Tech giants building power plants

This isn’t merely about constructing a few more buildings filled with servers. The demand profile of AI changes the equation. Training and inference at scale require dense compute, specialized hardware, and increasingly exotic networking. This results in more power per square foot, more heat per rack, and greater stress on the grid. The hyperscalers and their peers are responding the way rational industrial players respond: by trying to control the bottleneck.

If the bottleneck is electricity, the solution—at least on a spreadsheet—is to secure it. That’s why we’re seeing moves that echo vertical integration: long-term power purchase agreements, partnerships with utilities, investments in transmission, and, yes, a growing appetite for building or enabling dedicated generation. Call them “power plants,” “energy campuses,” or “co-located generation assets.” The name is less important than the intent. The hyperscalers want predictable, scalable power because their revenue story depends on predictable, scalable compute power.

The bet is straightforward: When demand spikes, prices and utilization rise, and those who built first make bank. Build the capacity, fill the capacity, charge a premium for the scarce resource, and ride the next decade of digital expansion. It’s the same playbook we’ve seen before in other infrastructure booms, except this time the infrastructure is made of silicon and electrons, and the pitch is wrapped in the language of transformation.

Predicting demand

Now for the part that makes people uncomfortable at conferences. In my opinion, a lot of these high-tech players—including the cloud providers—are in way over their heads. Not because they can’t build. They can build. Not because they can’t raise money. They can raise money. The risk is that they are treating a forecast as if it were a law of physics.

Nobody truly knows what AI demand will be in three or five years. We observe trends, adoption curves, and product launches, and we have considerable hope. But demand isn’t just interest; it’s budgets, governance, and integration. Demand generates security reviews, procurement cycles, data readiness, and countless small frictions that add up to one big constraint: Enterprises are slow to adopt any technology, and AI is no exception.  

Enterprise adoption doesn’t happen because a vendor says it will. It happens when the technology fits the operating model, the risk is understood, the data is accessible, the legal team is satisfied, the business case withstands scrutiny, and someone is willing to be accountable for the outcome. That always takes time. If you’re building infrastructure as if every enterprise will sprint, you’re ignoring decades of evidence showing that most organizations jog, and still more will walk.

Then there’s the cost reality. AI systems, especially those that deliver meaningful, production-grade outcomes, often cost five to ten times as much as traditional systems once you account for compute, data movement, storage, tools, and the people required to run them responsibly. That cost multiplier is not a rounding error. It’s a different category of spending. Most enterprises do not have unlimited funds. They will experiment, pilot, and selectively deploy. The idea that they will all suddenly bankroll massive, continuous AI consumption at hyperscale levels? That’s a leap.

So, what happens if demand doesn’t land where the builders expect it? Overcapacity. Price pressure. Write-downs. A quiet shift from “we’re investing aggressively” to “we’re optimizing efficiency.” And back in that rural town, a community is left with the long-term consequences of a short-term hype cycle: a grid strained to the limit, water debates turned political, and a landscape transformed by buildings that may never operate anywhere near their promised peak.