ChatGPT is offered as a free service across multiple platforms, with usage limits that eventually prompt users toward paid subscriptions. It responds with language similar to how queries are submitted, something humans rarely do. It is a small but telling sign of how seamless the technology has become. I use it for quick factual questions and longer processes, such as planning a garden season. Because it is free at the entry level, I feel free to use it. I suspect many people do.
Ease of access matters. When a tool is free and always available, people experiment. A person awake at 2 a.m. might reach for a phone and ask how to sleep through the night. If a solution seems helpful, word spreads. Usage grows not because of marketing campaigns, but because of social diffusion. This is how habits form.
The question is whether such growth — multiplied across millions — materially stresses infrastructure, including the electrical grid.
The U.S. Department of Energy reported in late 2024 that data centers consumed about 4.4 percent of total U.S. electricity in 2023 and could rise to between 6.7 and 12 percent by 2028, depending on growth scenarios. That range is significant. It reflects assumptions about adoption rates, model size, efficiency gains, and capital deployment. These are projections, not certainties.
In public discussion, however, projections often harden into inevitabilities. Upper-bound scenarios become planning baselines. Large numbers circulate with little context. Some usage statistics are widely repeated without clear sourcing. Investor forecasts about billions of weekly uses and massive subscription growth are forward-looking, not present realities.
This is where a larger question emerges:
Is enthusiasm for artificial intelligence and data centers outrunning prudence in financial investment? In other words, do investors have fear of missing out and therefore accept speculative arguments about market capacity more than they should?
Comparisons are sometimes made to the 2008 financial crisis. That collapse was driven by mortgage-backed securities embedded throughout the banking system, amplified by leverage and mispriced risk. Institutions such as Lehman Brothers and insurers like AIG were deeply exposed. When housing prices faltered, the system unraveled because debt was layered upon debt.
AI investment today differs in important ways. Much of it is equity-funded venture capital or corporate capital expenditure rather than highly leveraged household debt. Data centers, chips, and transmission lines are tangible assets, not synthetic securities. Losses, if they occur, are more likely to be concentrated among investors rather than embedded in consumer balance sheets.
Yet there are echoes worth noting. In both periods, capital flowed rapidly toward a dominant narrative. In both, optimistic forecasts shaped infrastructure decisions. In both, participants understood risk existed — but incentives encouraged staying in the game.
The concern is not that investors seek profit. We know that. The concern is whether optimistic projections become assumed outcomes. If infrastructure is built on the expectation of maximum adoption, and adoption plateaus or efficiency improves faster than expected, overcapacity can result. That is not necessarily a systemic crisis. It may be a costly misallocation of capital.
Critics such as Bill McKibben, citing technology writer Ed Zitron, argue that the economics of large AI firms may resemble a bubble: vast capital expenditures today justified by revenue expectations that may or may not materialize. That critique is itself an interpretation, but it highlights the degree to which AI investment rests on assumptions about future returns.
My own daily queries consume negligible electricity. The grid impact, if any, arises from aggregate industrial-scale deployment and the assumptions embedded in those decisions. Casual consumer use is a marginal contributor. Large enterprise integration and model training cycles are the dominant drivers.
So the core issue may not be whether AI will use more electricity — it almost certainly will — but whether forecasts are being treated as destiny. Markets routinely oscillate between overconfidence and retrenchment. The challenge is distinguishing durable growth from narrative momentum.
It is possible that artificial intelligence becomes foundational infrastructure, like electrification or broadband. It is also possible that investment temporarily overshoots practical demand. Both can be true at different stages of a technology cycle.
The prudent stance is neither inevitability nor collapse, but clarity: separate measured data from modeled projections, and projections from belief. When enthusiasm begins to substitute for disciplined evaluation, that is when risk accumulates unawares.
~This essay was developed with the assistance of ChatGPT, an AI tool created by OpenAI, which I used to test arguments, fact check, clarify projections, and stress-test comparisons. The ideas and conclusions are my own.
Journey Home 