The real AI bottleneck is not a chip
For all the attention on processors and model training, the actual constraint on the AI buildout is something older and less exciting: electricity. Data centers are power-hungry, and the grid is not ready to feed them as fast as they want to be built. Adam O’Dell’s “Tech-Opoly” pitch makes this the centerpiece, arguing that the companies owning the grid’s pinch points, the switchgear, the transformers, the distribution channels, will capture the buildout’s spending no matter who wins the chip race. The framing is correct that power, not compute, is the bottleneck today.
Why power is the constraint
The physics is simple and unforgiving. A single large data center can draw as much power as a small city, and hyperscalers want to build dozens of them. Utilities face interconnection queues that stretch years, and even when a project clears the queue, the physical gear needed to connect it, transformers, switchgear, breakers, has its own multi-year backlog. Every layer of the problem points back to the electrical supply chain.
The result is that utilities are reporting record load growth for the first time in decades, and the demand is concentrated in the regions where data centers are clustering. Our data center power demand explainer walks through how that demand has reshaped the market.
Where the money actually flows
When a data center connects to the grid, the spending fans out across a handful of businesses. Electrical equipment makers build the switchgear and transformers. Distributors move the gear from factory to job site. Engineering firms plan the layout and handle the permitting and construction management. O’Dell’s basket spreads across all three, which is why the theme is more about the electrical layer broadly than about any single product. We trace the flow in our data center infrastructure stocks explainer.
The pinch point is real, but entry price matters
The thesis has a genuine foundation. The electrical layer is scarce, backlogs are real, and the demand is durable. What is less certain is whether that foundation is still cheap. Several of the names in the pitch have already run up sharply since they were first teased, and a real bottleneck can still be a bad purchase at the wrong price. The honest way to read the theme is as a confirmed demand story with a late entry point, not as a fresh discovery. The utility load forecasts that underpin the theme have been climbing for a couple of years now, and the market has had ample time to digest them. The demand is real; the question is how much of it is left on the table for anyone buying today rather than a year ago.
How to think about the power stocks
For an investor, the useful question is not whether the AI buildout needs power, it obviously does, but which companies convert that need into pricing power. Makers of scarce equipment can raise prices; resellers on thin margins cannot; engineering firms capture fees rather than upside. Sorting the basket along that line is the difference between owning a chokepoint and merely renting space next to one. Our AI data center power stocks explainer digs into both sides of that question.
The three bottlenecks inside the power story
The phrase “power is the bottleneck” hides three distinct problems, and each one points to a different kind of company. The first is generation: there has to be enough electricity to begin with, which is a question for utilities and independent power producers, not the equipment makers. The second is transmission and distribution: the grid has to move that electricity from where it is generated to where a data center sits, and that requires new lines, substations, and the transformers and switchgear that populate them. The third is connection: a data center has to actually plug into the grid, and the interconnection queue, the waiting list for new connections, now stretches years in the busiest regions.
O’Dell’s basket lives almost entirely in the second and third bottlenecks, the equipment and the engineering around getting connected. That is the right neighborhood, because those are the parts of the story a public investor can actually buy. You cannot easily buy a stake in a utility’s future load growth, but you can buy the company that builds the gear the utility has to order to serve it. The tradeoff is that everyone else has figured out the same thing, and the equipment names have already been bid up. The bottleneck is real; the question for a new buyer is whether it is still cheap.
NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.