The money finally figured out that you can't train frontier models on vibes and venture capital—you need actual electrons.
The Summary
- Guggenheim Partners Executive Chair Alan Schwartz warned at Milken that US electrical grid limitations could cost America its AI leadership
- Infrastructure constraints aren't a future problem—they're already slowing buildout of the data centers that underpin the agent economy
- Wall Street's starting to price grid capacity as the real bottleneck, not compute or capital
The Signal
Alan Schwartz isn't a tech founder chasing hype. He's the executive chair of Guggenheim Partners, a firm managing north of $300 billion. When he tells the Milken Institute that America's electrical grid is the constraint on AI development, he's reading a balance sheet, not a pitch deck. The implication is stark: the US could lose the AI race not because we lack talent or money, but because we lack watts.
The timing matters. We're past the ChatGPT moment where AI was a curiosity. We're into the phase where companies are building agents that work while humans sleep, where inference costs money at scale, where training runs measure in megawatt-hours. Data centers for frontier models aren't office parks. They're industrial facilities with power requirements that dwarf small cities.
"The US risks falling behind in development of artificial intelligence because of the need to upgrade the electricity grid."
Here's what Schwartz is really saying: capital follows infrastructure. If you can't plug in, you can't build. And if you can't build in the US, you build somewhere else. That's not a hypothetical. Ireland, Singapore, and parts of the Middle East have already started positioning themselves as AI-friendly jurisdictions partly on the strength of their power infrastructure and grid planning.
The grid problem cascades:
- Training clusters need reliable baseload power, not intermittent renewables alone
- Data center buildouts face multi-year permitting delays for electrical upgrades
- Utilities weren't designed for the sudden, massive loads AI infrastructure demands
This isn't about Elon tweeting that we need more nuclear plants. It's about transmission lines, transformer capacity, and the boring, expensive work of rebuilding electrical infrastructure that hasn't seen meaningful investment since the 1970s. The kind of work that takes a decade and doesn't photograph well for LinkedIn.
The Implication
If you're building anything in the agent economy, start thinking about power like you think about talent. It's finite, it's regional, and it's about to get expensive. The smart money is already handicapping which US states and which countries have the grid capacity to support AI at scale. Watch where the hyperscalers are actually breaking ground, not where they're announcing partnerships.
For policymakers, this is the test. China doesn't have better AI researchers than the US. But if they can deliver industrial power to data centers faster than we can navigate environmental reviews and utility commission hearings, they will win by default. Infrastructure is strategy. The boring stuff is suddenly the only stuff that matters.