Permission is now the constraint du jour of the AI buildout. Most of the attention so far has gone to power, and more recently to credit. Those constraints are still very real, but to us, the biggest constraint is that nobody wants gigawatt-scale data centers in their backyard.
The midterms made that impossible to ignore. On Tuesday, 41% of the country's planned data center capacity landed in states whose governor backs a pause. Voters sided against data centers in at least 36 of 41 local ballot measures.
The industry is planning for a buildout that communities haven't agreed to. We think a lot of people underwriting data center growth are in for a rude awakening in 2027. SemiAnalysis expects about 50 GW of new data center capacity to come online next year. The smartest infrastructure investors and developers we know think half of that, at best, actually shows up. The buildout is going to take much longer than people think.
The backlash doesn’t stop at data centers. New power plants are now seen as part of the data center buildout, so they inherit all the opposition data centers draw. That’s a big problem, because we can’t grow our industrial base without more power. We need a different way to develop powered land, one that communities actually want because it serves their own ambitions. Getting there, and making new power politically popular, will take new technology to develop and manage energy production in this country. That’s the opportunity we’re so excited about.
It's easy to see how we got here. Training needs massive scale. Inference may need less, but the economics of building still favor going big. In practice, many of these gigawatt campuses won't get built. Every gigawatt campus has to clear local approvals, state politics, environmental permits, equipment shortages and a construction schedule with no slack in it. Any one of those can kill a project or push it out by years. Project Jupiter in New Mexico had to swap its gas turbines for fuel cells after pushback on its air permit. For all the announcements, Epoch AI expected only five campuses to cross a gigawatt in all of 2026.
None of this means we need less power. Whether the driver is AI or re-industrialization, and whatever happens in the compute market, this country has to produce a lot more electricity than it does today. Getting there means changing how energy gets developed on three fronts.
First, powered land has to serve more than data centers. Building only for compute is politically toxic and, frankly, a bad allocation of resources. A site that powers a factory, a hospital and a data center brings jobs and shared infrastructure the community can actually see. Including a data center as one part of an industrial park or power project can help finance the factories and other things communities actually need. In effect, it's an indirect tax on the data center, and we think that's the right way to do this. The data center helps pay for it, and the headline is jobs, innovation and local prosperity.
Second, the line between behind-the-meter and front-of-the-meter is starting to blur. To add firm power at the pace the country needs, every project will be co-located and grid-connected, with a microgrid that ties the onsite generation and storage to the grid. We think powered land ends up "across the meter," effectively just adding to the grid. Projects in the hundreds of megawatts can find grid capacity that a 2 GW request never will. But managing these microgrids will be a real challenge. They have to serve very different tenants on site and comply with the grid's interconnection, reliability and market rules at the same time, and those rules keep changing.
Third, the energy generation supply chain will keep moving. Turbines may not arrive, because of shortages or because an air permit rules them out. Geothermal, nuclear and new storage are all on the way. Powered land of the future has to be able to add and swap power sources over its life without starting over and be flexible enough to accommodate different supply constraints.
Each of those shifts adds complexity. A factory, a hospital and a data center all draw power differently. Onsite generation, storage and the grid all have to work together, under rules that can change with every election. Running that reliably, safely and efficiently for twenty years takes a sophisticated microgrid and software that can design it, operate it and keep optimizing it. Shortages get competed away as supply catches up, but complexity compounds.
Today's powered land market is the Wild West, mostly land flippers holding options on dirt. Development is unsophisticated, and very few developers are building with these realities in mind. The ones that do will move faster. With the technology to show a community a clean, multi-use design it can get behind, they’ll get projects approved, permitted and entitled ahead of everyone else. This is increasingly important in the time-to-power economy.
Managing the energy assets of a flexible, across-the-meter powered land project will be really hard. Someone has to balance immensely different power needs and load profiles of a factory, a data center, a hospital, onsite generation, storage and the grid interconnection every hour of every day, for decades. That will require new technology we don’t have yet. We think that’s one of the biggest opportunities in energy.
The gigawatt campus had its moment. The next decade belongs to powered land people actually want.