The grid’s biggest new customer
For decades, electricity demand in the United States was basically flat. Efficiency gains offset population growth, and utilities planned for a future of mild decline. Artificial intelligence broke that assumption. Data centers, the facilities that train and run AI models, have become the fastest-growing source of new electricity demand in the country, and their appetite is measured in gigawatts rather than megawatts.
A single large data center campus can draw as much power as a mid-sized city, and the industry wants to build many of them at once. That collision between demand and a grid that cannot expand quickly is the single most important constraint in the AI story, and it is the exact pressure point the Homestacks promotion is built around.
Why the numbers are so large
The scale is easy to understate. A traditional server rack might draw 5 to 15 kilowatts. A rack packed with modern AI servers can draw 100 kilowatts or more, and a single campus might hold thousands of those racks. Multiply that across the handful of companies building frontier models, and the total demand rivals the output of entire regional power systems.
What makes it harder for the grid is timing. Data centers want power immediately, but new generation and transmission take years to permit and build. Interconnection queues, the waiting lists for new projects to connect to the grid, stretch out for years in the busiest regions, which is why power availability, not chip supply, is now the binding limit on AI growth. Our explainer on the 212x AI energy boom digs into how that demand translates into investment.
The Homestacks answer to the bottleneck
The Angel Publishing Homestacks promotion argues that this grid pressure will push some AI compute out of the giant campus and into a distributed appliance installed on new homes. The real product behind the idea is SPAN’s XFRA, an outdoor unit about the size of an air-conditioning compressor that pairs NVIDIA Blackwell GPUs with a smart electrical panel and a home battery. SPAN covers the host homeowner’s power and internet bills, and the first pilot is roughly 100 units in build-to-rent communities in Arizona and Nevada with PulteGroup.
The concept has a real logic to it. Spreading compute across many homes could sidestep the transmission bottleneck by drawing power where it already exists. But the scale today is tiny, and SPAN’s own press release says the appliance is meant to augment centralized data centers, not replace them.
The names that actually sell the power
The promotion’s three picks map cleanly onto the power story. Vistra Corp. (VST) is the most direct, a Texas power producer with purchase agreements with Amazon and Meta and a long-term deal for half of its Comanche Peak nuclear plant. nVent Electric (NVT) supplies the enclosures that carry power inside data centers, with that business near 40% of revenue. Generac Holdings (GNRC) is the home-energy angle, selling batteries and grid controls with a roughly $700 million data-center backup-power backlog.
Each is a real business tied to electricity demand, but none of them lives or dies on a 100-unit housing pilot. The honest read is that these are data-center and grid bets first. Our look at America’s new power grid covers the broader infrastructure that has to get built to make the demand real.
What actually slows a data center down
The constraint on data center growth is rarely the will to build. It is the wait. Before a new campus can draw power, the developer has to secure an interconnection, the agreement that lets the facility connect to the grid and pull a specific amount of electricity. In the busiest regions, the queue for new interconnections stretches years, because the grid was not designed for facilities that consume the power of a mid-sized city.
Permitting and transmission are the second bottleneck. A power line that carries electricity to a new data center can take years to plan, approve, and build, and it faces land-use and environmental reviews along the way. Utilities are expanding, but they are expanding on a timeline that does not match the speed at which the hyperscalers want to deploy new capacity.
That gap is the entire reason the Homestacks idea exists. If the grid cannot serve the big campus fast enough, the pitch goes, then move some compute to where the power already is, in homes that are already connected. It is a clever reframing of a real problem, and it is why power, more than chips or land, is the binding constraint in the AI story.
Ready to see the research? Click here to access the Angel Publishing report.
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.