Amory Lovins

Thank you, Scott. It's great to be here. The US grid is at a genuine inflection point — the combination of AI data center load growth, electrification of transportation, and the need to decarbonize is creating a set of pressures we haven't seen since the grid was first built.

The old model was built around slow, predictable load growth — maybe one to two percent a year. Utilities could plan decades ahead with reasonable confidence. What we're seeing now is a step-change. A single hyperscale AI data center can draw 500 megawatts or more. That's the equivalent of adding a mid-sized city to the grid overnight. The interconnection queue is overwhelmed. Transmission is the binding constraint. And the planning cycles that utilities use — five, ten, twenty years — are completely mismatched to the pace of technology deployment.

Absolutely — and in fact the efficiency opportunity in AI compute is enormous and largely untapped. The energy intensity of AI inference has been dropping dramatically as chip architectures improve. But the efficiency gains in the data center itself — cooling, power distribution, server utilization — are still leaving enormous savings on the table. The most efficient hyperscalers are operating at a PUE of 1.1 or below. The industry average is still closer to 1.5. That gap represents hundreds of terawatt-hours of wasted electricity. Efficiency isn't just good for the environment — it's the fastest way to relieve pressure on a grid that can't build transmission fast enough.

Transmission is where the energy transition either succeeds or stalls. We have extraordinary renewable resources — wind in the Great Plains, solar in the Southwest, offshore wind on both coasts — but they're not where the load is. Moving that power requires new high-voltage transmission lines, and building those lines takes ten to fifteen years under current permitting and siting rules. FERC Order 1920 was a significant step — it requires regional transmission planning for the first time in a serious way. But implementation is going to be the hard part. The utilities, the states, the landowners, the interconnection operators — they all have to align. That's a coordination problem as much as a technical one.

DERs are enormously valuable precisely because they're located where the load is. A battery behind the meter at a data center can shave peak demand, provide frequency response, and defer transmission upgrades. Virtual power plants — aggregations of distributed resources managed as a single dispatchable asset — are starting to show real scale. California has run VPP programs that have delivered hundreds of megawatts of peak reduction. The challenge is the regulatory framework. Most utility tariffs and market rules were designed for a world of centralized generation and passive consumers. Reforming those rules to properly value distributed resources is one of the most important policy tasks in front of us.

I'd change the planning horizon and the incentive structure simultaneously. Right now, utilities are rewarded for capital investment — building things — not for outcomes like reliability, efficiency, or carbon reduction. If you restructure the regulatory compact so that utilities earn returns on the outcomes they deliver rather than the assets they deploy, you unlock an enormous amount of innovation. The utility of the future is a platform for distributed intelligence, not a one-way pipe from a central power plant. The companies that figure that out first are going to define the next fifty years of American energy.

Thank you, Scott. Keep pushing on this. The grid is the infrastructure that makes everything else possible.