The Electron Is the New Oil

A Grid Built for a World That No Longer Exists

The American grid was built around a simple bargain. Big plants burning coal and gas, or splitting atoms, could ramp up and down on command, because a grid with almost no storage must match supply to demand continuously, every second of every day.^[1] For a century that worked.

Two things are now breaking it at once. Solar has become the cheapest electricity on Earth, but the sun arrives at noon whether you need it or not and leaves every evening whether you need it or not.^[1] And demand has stopped behaving. The International Energy Agency (IEA) expects U.S. electricity consumption to rise close to 2% a year through 2030, more than twice the pace of the past decade, with data centers supplying about half of that growth.^[2]

Utilities’ own forecasts are more dramatic. Grid Strategies, a power-sector consulting firm, compiled them in late 2025 and found 166 gigawatts of new peak load expected by 2030, roughly 90 GW of it tied to data centers.^[3] The Clean Energy States Alliance, summarizing the same report, translates that into about 5.7% annual growth in demand.^[4] These are forecasts, not measurements, and Grid Strategies itself warns they could prove too high or too low.^[3] I would rather show you the range, roughly 2% to 5.7%, than one confident number. The disagreement is part of the story.

Globally, the IEA’s base case has data-center electricity use roughly doubling to about 945 terawatt-hours by 2030. To keep perspective, that is still just under 3% of world electricity, though it is growing more than four times faster than all other uses combined.^[5] In 2025, data-center demand grew 17% and AI-focused facilities grew 50%.^[6]

The load is also lumpy. Data centers consumed about a quarter of all the electricity used in Virginia in 2023, according to the Electric Power Research Institute (EPRI). Its latest analysis, as summarized by the American Public Power Association, puts that share between 39% and 57% by 2030, with seven more states possibly passing 20%.^[7]

Intelligence at scale is an energy problem before it is anything else.

Put the Battery on the House

The obstacle to adding big batteries to the grid is not chemistry. It is the waiting line. Berkeley Lab counted more than 2,060 gigawatts of generation and storage actively seeking a grid connection at the end of 2025, and most of it will never be built: of the capacity that applied between 2000 and 2020, only 13% had reached commercial operation by the end of 2025.^[8]

This is where a startup called Base Power comes in, profiled in August on a16z’s newsletter by Erin Price-Wright and Michael McGuiness.^[1] Its founders, Zach Dell and Justin Lopas, noticed that a home battery and a transmission line do the same job. One moves power through space; the other moves it through time, storing noon’s cheap sunshine and releasing it at 7pm when everyone is home and prices spike.^[1]

Their answer was to stop waiting. Put the battery on the house. The house is already connected, so there is no queue, and the homeowner draws from the battery directly, sidestepping the congested wires.^[1] Base earns most of its money by charging overnight, when power is cheap, and selling back during the evening peak. So, unlike a company that sells batteries at sticker price, it wants bigger batteries on more roofs.^[1] The article also reports that during Texas’s record demand in July, batteries supplied nearly 12 GW at the peak and wholesale prices touched about $0.30 per kilowatt-hour, against evening spikes above $4 in the summers of 2023 and 2024.^[1] I have not independently verified those figures.

Why now? Cost curves. According to the same account, solar has fallen about 20% in cost with every doubling of cumulative production and batteries about 23%, a pattern economists call Wright’s Law, and battery-pack prices are down more than 90% since 2010.^[1] It is the same compounding logic behind the collapse in AI prices I traced in The Free Fall.

I recognize the move. It is the one every builder eventually makes: stop trying to convince the gatekeepers and build the thing that does not need their approval.

The Scoreboard Is Electrical

For most of history, who has the power was a political and military question. In this century it is becoming an electrical one.

Consider the scoreboard. China’s electricity demand passed 9,500 TWh in 2025, according to the IEA.^[2] U.S. consumption that year was about 4,200 TWh, according to the Energy Information Administration (EIA).^[9] China is also where much of the equipment comes from: it accounted for over 80% of global battery-cell production in 2025,^[10] and the IEA’s earlier supply-chain analysis found its share of every manufacturing stage of solar panels, from polysilicon to finished modules, already exceeded 80%.^[11] Whoever has the cheapest, most reliable electrons will tend to train the biggest models and run the densest clusters.

I tracked part of this scoreboard in the AI Compute Index, and I traced the view from India’s side of it in The Refinery Problem. Satellites turn out to be the same kind of question; see The Sovereign Sky.

There is a deeper pattern. Transformative technologies keep tracing back to jumps in how cheaply we can move energy. The a16z account tells the story of Niagara Falls power reaching Buffalo in 1896: it made aluminum cheap enough to produce at scale, and aluminum gave the Wright brothers an engine light enough to fly.^[1] One improvement in moving electricity; a century of consequences nobody could have predicted from where they stood.

The same article closes on Nikola Tesla, who argued in 1900 that “the great problem of science” is to increase the energy available to humanity, and who died in debt in 1943.^[1] A century and a quarter later we finally have the cheap battery he was asking for. What we do not yet have is a grid, a market, or a set of incentives built to use it.

That gap, between what the technology can now do and what our institutions allow, is the whole story of the Singularity in miniature. It is the mismatch I explored in The Clock Speed Problem, and from the legal side in Dispatch No. 7.

The Uncomfortable Symmetry

There’s a version of this essay that stays purely technical — batteries, kilowatt-hours, interconnection queues. But the part that actually kept me thinking about this piece is more personal, and I think it belongs in the book.

I was totally broke. And I had no time to sit in the wreckage and feel sorry for myself, because the arithmetic was brutally simple: to resurrect myself I needed knowledge, and I had none of the kind that would matter. So I did something a sensible person would call insane. I chose the single most bizarre business idea available to a man in my position — no money, no coding experience, no computer-science degree, no academic pedigree of any kind — which was to teach myself software development from scratch and build enterprise-class software.

I built FedMine alone, in four years, without venture backing, without a team of Anduril and SpaceX veterans, without a billion-dollar Series D. What Zach Dell and Justin Lopas are doing at industrial scale — refusing to wait for permission from an incumbent system that has no incentive to move — is the same instinct that drives every solo founder who ever shipped something the “proper” channels said would take five years. The scale is different. The instinct isn’t.

If there is a thesis I keep circling back to in this book, it might be this: the Singularity will not be a single dramatic morning when a model wakes up. It will be a thousand quiet unbundlings like this one — batteries unbundling from utilities, intelligence unbundling from institutions, capability unbundling from permission — each one small enough to miss, until you add them up and realize the ground has already moved.