The AI buildout is about to hit a wall made of electrons and water, and the backlash is already showing up at the ballot box.
The Summary
- Morgan Stanley warns US AI data centers face a 32 GW power shortfall by 2028, enough to trigger state moratoriums on new facilities
- New cooling tech cuts on-site water use but quadruples demand at power plants, shifting the resource burden rather than solving it
- Voter anger over data center electricity impact is growing ahead of US midterms, threatening to reshape energy policy debates
- Water use at data centers expected to quadruple by 2028 in a country already facing scarcity concerns in key tech hub regions
The Signal
The math is simple and brutal. Morgan Stanley projects a 32 gigawatt power gap between what AI data centers need and what the grid can deliver by 2028. For context, that's roughly the entire power consumption of Australia. The shortfall isn't theoretical. It's already prompting talk of regulatory moratoriums in states that were supposed to be the backbone of AI infrastructure buildout.
The industry thought it had a workaround. Advanced cooling systems reduce water consumption at the facility level, making individual data centers look cleaner on paper. But the water didn't disappear. It moved upstream to the power plants generating the electricity those facilities consume. It's accounting arbitrage, not conservation.
"The shift in cooling tech may prompt regulatory scrutiny due to increased water demands on power plants, especially in water-scarce regions."
Here's where it gets political. Voter discontent over data center electricity use is rising fast enough to matter in midterm elections. Residents in Virginia, Arizona, and parts of the Midwest are watching their power bills climb while tech companies lock in cheap industrial rates. They're seeing water reserves drain in drought-prone areas. And they're connecting the dots between the new data center down the road and the grid strain warnings from their utility companies.
The water situation compounds the power problem. Data center water use is on track to quadruple by 2028, hitting hardest in regions already managing scarcity. When cooling tech shifts water demand to power plants, it concentrates the impact at facilities often located in water-stressed areas. Thermal power plants need massive water volumes for cooling. Adding AI's electricity demand on top of baseline growth means those plants either expand water use or throttle output.
Key constraints converging:
- 32 GW power gap by 2028
- 4x increase in water consumption
- Growing voter backlash in swing regions
- State-level regulatory threats already surfacing
This isn't a supply chain hiccup. It's infrastructure hitting physical limits while the political window for expansion closes. The companies racing to build Web4 agent infrastructure assumed the power would be there when they needed it. That assumption is breaking.
The Implication
Watch for three things. First, a bifurcation in where AI gets built. Companies with early power deals and existing cooling infrastructure will have a structural moat. New entrants will face permitting hell and community opposition. Second, expect serious money flowing into small modular reactors and on-site power generation. When the grid can't deliver, you build your own. Third, the politics of AI are about to get local in a way that matters. Voter anger over electricity bills and water use will shape which states welcome data centers and which ones freeze them out.
If you're building agent infrastructure or tokenizing compute, your location strategy just became your biggest risk factor. The power isn't coming. Plan accordingly.