AI Energy Bubble: Why Data Center Forecasts Might Be a Bluff

Energy Bubble: Why Data Center Forecasts May Be a Bluff

For the past six months, we've heard nothing but talk of artificial intelligence consuming all the planet's electricity. Sam Altman is hunting for billions for chips and reactors, Microsoft is reanimating Three Mile Island, and utility companies across America are rubbing their hands in anticipation of a windfall. But what if all this panic about capacity shortages is just another overheated narrative, behind which corporations want to justify budgets for decades to come? Gregg Orrill, an analyst at AEGIS Hedging, decided to set aside marketing presentations and simply do the math. His conclusion sounds like a cold shower for the industry: energy companies are planning to build twice as much capacity as data centers will actually need.

Let's recall how we got here. Just two years ago, the utility sector was considered the most boring place on the market. Stocks grew slowly, dividends were paid reliably, and demand for energy in the US barely changed for decades.

But then came ChatGPT, and suddenly it turned out that training each new model requires exponentially more electricity. Big tech companies literally started fighting for the right to connect to the grid. As a result, forecasts for energy consumption growth skyrocketed.

Utilities, unaccustomed to such attention, quickly revised their investment plans, allocating billions of dollars to building new substations and transmission lines. But therein lies the systemic error that Orrill points out. In chasing the trend, the industry began extrapolating peak demand infinitely, ignoring basic laws of efficiency and market correction.

The problem is that current forecasts are built on the assumption that every announced data center project will be fully implemented and will consume energy at 100% capacity around the clock. In reality, we see a huge gap between connection requests and actual construction. Many startups reserve capacity "just in case," fearing shortages, thereby creating an illusion of resource scarcity.

This is a classic example of the "bullwhip effect" in supply chains: a small fluctuation in demand at the end of the chain (the desire to train a model) causes massive distortions among raw material and infrastructure suppliers. If even a third of these projects are not realized, energy companies will be left with excess assets that consumers will ultimately have to pay for through increased tariffs.

Moreover, the AI industry is currently at the stage of wild pursuit of efficiency. Algorithm developers understand that infinitely inflating model parameters is a dead end. Training methods are emerging that require far less energy, and next-generation chips are becoming increasingly energy-efficient.

If tomorrow OpenAI or Anthropic find a way to train models 30% more efficiently, the entire investment strategy of utility giants will fall apart. We already saw something similar in the early 2000s, when the world actively laid fiber optic cables in anticipation of an internet boom. Back then they built so much infrastructure that most of it remained unused for years, and cable-laying companies went bankrupt en masse.

History tends to repeat itself, only now instead of cables we have giant transformers and nuclear reactors.

The situation is further complicated by the fact that regulators and politicians have enthusiastically embraced the theme of "energy sovereignty for AI." Under this banner, it's much easier to approve large-scale construction projects and subsidize old power plants that were scheduled to be closed. But when the dust settles, it may turn out that we've built infrastructure for a world that doesn't exist.

Investors should take a closer look at how justified the energy companies' appetites are. For now, the market believes in endless growth, but the first quarterly report from a major cloud provider showing a slowdown in capital spending could trigger a chain reaction. And then it will turn out that the lone analyst was right, and the rest of the market was simply chasing a shiny object.

Key takeaway: are we confusing real demand with panic buying in advance, and who will pay the bills for unused megawatts when the hype around AI finally dies down?