A Practical Plan to Power the AI Boom – Watts Up With That?

By Theodor Engøy

The AI surge is exposing an old truth: electricity is the master resource. Data‑center power demand is projected to more than double globally by 2030. In the United States, the interconnection queue is now so large that both new generation and large loads face multi‑year delays. Unless utilities, grid operators, and hyperscalers strike the right deals, AI capacity will keep outrunning electrons—and public patience.

Start with a simple rule: buy electrons before bytes. Match each gigawatt of new data‑center load to contracted, firm low‑carbon generation (nuclear, hydro, geothermal, gas with CCS where credible) plus storage and specific transmission upgrades. Put these commitments in public, milestone‑based contracts. If the power doesn’t show up on schedule, the load waits. Negotiate water honestly: avoid evaporative cooling in arid regions; everywhere, publish water‑use metrics and site accordingly.

Cities can turn “waste” into an asset. Data‑center heat is already warming homes at scale in Finland, where Fortum and Microsoft are connecting new facilities into the Helsinki‑region district‑heating network.

The hardware is straightforward—industrial heat pumps and large‑diameter pipes. The hard part is governance: contracts, tariffs, and interconnection timelines that align incentives across utilities, municipalities, and cloud buyers.

Next, design internal efficiency gates that executives and regulators can verify. Measure energy per unit of useful work: megawatt‑hours per billion tokens for training; megawatt‑hours per million inferences for operations. Publish PUE and WUE, plus siting rationales and heat‑reuse metrics. Route workloads to the cheapest model that meets quality and risk thresholds. New accelerator generations deliver large gains in performance per watt; lock those savings in rather than chasing square footage.

Finally, widen access to compute without widening the footprint. Public‑interest compute programs can steer demand to the most efficient capacity while seeding skilled talent. The U.S. is standing up a National AI Research Resource Operations Center to coordinate shared resources. Pair efforts like that with “compute passes” redeemable across clusters and clouds, clear eligibility, and quarterly dashboards that report energy, water, and outcomes. That transparency earns the public’s trust.

For utilities, this is a growth story if done right. Long‑term virtual PPAs, storage adders, and targeted transmission upgrades can underwrite new clean capacity with predictable offtake. For operators, it’s a risk‑reduction story: fewer permitting fights, fewer headlines about strained substations, more projects that quietly work. And for communities, it’s tangible benefits—lower heating emissions, industrial jobs, and reliable power.

Electrons are the rate‑limiting step for AI. If we align contracts, metrics, and siting with that fact, we can scale compute without crashing the grid—and bring the public along.

Theodor Engøy is an independent writer based in Ås, Norway, focused on the intersection of AI, energy, and infrastructure. He has no financial relationships with entities cited and is not being compensated for this submission.

This article was originally published by RealClearEnergy and made available via RealClearWire.

Related