A San Francisco clean-power company teaching the electric grid to think in real time - one battery, one megawatt, one AI agent at a time.
The mark sits where the engineers do: somewhere between a power plant and a server rack, depending on the hour.
It is a Tuesday afternoon in San Francisco and somewhere in Texas a 100-megawatt battery is making a decision. Charge now or wait twenty minutes. Sell into the day-ahead, or hold for the imbalance market. The decision is small. The math is not. Behind it sits a software platform called PowerOS, built by a company called Equilibrium Energy, run by people who used to work inside the utilities they now sell to.
Equilibrium is, in its own words, a digital-native power company. That phrase reads strange next to the rest of the industry. Power has historically been the slowest-moving corner of the economy, a place where Excel is considered modernization and a five-year deployment cycle is normal. Equilibrium did not come to argue with that. It came to route around it.
Anyone who has spent time inside a power company will tell you the same thing in roughly the same order. The data lives in twenty places. The models live in another twenty. The traders, the operators, the originators, and the developers all use different tools that do not talk to each other. Decisions get made in spreadsheets, glued together by people who have memorized things they should not have to memorize.
Layer on top of this a grid that is changing faster than at any point in its 140-year history. Renewables come online by the gigawatt. Batteries arrive in shipping containers. Markets clear in five minutes. Storms knock out predictability. The asset is physical. The market is financial. The interface between them is, charitably, a mess.
Ryan Hanley, the founder, had spent his career at the seams of that mess - at PG&E, then SolarCity, then Tesla, then Advanced Microgrid Solutions, then Shell. He kept watching smart people fail to do simple things because the underlying systems would not let them. The Equilibrium thesis grew out of that frustration, and out of a suspicion that the next decade of the grid would be won by whoever could connect what was already there.
The most quietly radical thing about Equilibrium is the resume of its early hires. NRG. Enel. Citadel. Tesla. SolarCity. The kind of CVs that, in most climate startups, are a footnote. Here they are the strategy. The company decided that to build software for power companies it should hire mostly power-company people, then teach them to ship.
That decision has consequences. It means the product roadmap is informed by traders who have actually lost money on bad weather forecasts, by operators who have dispatched real assets at three in the morning, by analysts who know exactly which file format a particular ISO insists upon. It also means the company can talk to its customers without an interpreter, which, in the power sector, is a small competitive advantage that compounds.
Five years, three rounds, one operating system. Mostly in that order.
The flagship is PowerOS, which the company calls a connective intelligence layer for power. Underneath that marketing language sits a fairly specific architecture. An enterprise data ontology that knows the difference between a generator and a contract. Forecasting and optimization services tuned to grid physics, not generic time-series. A risk engine. A trading engine. And on top of all of it, AI copilots for traders, researchers, and developers.
The first commercial application of PowerOS is called EQ Mission Control. It is the product NRG runs in production. It is the product that optimizes battery fleets across ERCOT and CAISO. It is, in spirit, the bridge between a control room and a model registry.
An AI-native operating system for power companies. Data ontology, optimization, forecasting, trading, risk. The plumbing nobody wants to build twice.
The flagship application. Real-time portfolio insight, forecasting, decision support, and AI copilots for the people who actually move megawatts.
Equilibrium takes on the optimization risk of battery assets in ERCOT and CAISO. The developers build steel. Equilibrium runs the brain.
AI agents that understand both grid physics and market structure. They draft, they explain, they argue back when the data says they should.
Numbers per Crunchbase and BusinessWire press releases. The grid is unimpressed by capital, but capital does not hurt.
The first piece of proof is NRG. The Fortune 500 power company is using Equilibrium's enterprise portfolio optimization in production, which in this industry is roughly the equivalent of a major bank running a startup's risk engine on real positions. Not a pilot. Not a sandbox. Real money, real megawatts.
The second piece is the battery business. Equilibrium signs tolling agreements with battery developers, takes over the optimization, and shares in the upside. As of 2024 those tolls expanded into California - a notoriously hard market - and the company reported what it called market-leading performance. The third piece is the 2026 partnership with a global commodities trading house to deploy proprietary AI-native power trading on PowerOS, which is the kind of customer that does not typically agree to be a reference.
None of this proves the long thesis. It does prove that someone is willing to pay for it now.
The receipts: not yet a victory lap, but enough paperwork to suggest the music has started.
Equilibrium's stated mission - reimagine the power industry through agentic AI by solving extreme data and systems fragmentation - is the kind of sentence that sounds like marketing until you have tried to do it. Most of the work of the energy transition is not invention. It is integration. New solar plants must talk to old substations. New batteries must respond to old markets. New AI must reason over old data. Whoever wires that together gets to claim a piece of the next grid.
Equilibrium has decided the prize is worth the plumbing. That is a more sober view of the climate-tech opportunity than is fashionable. It also happens to be correct.
Demand is rising again, after two decades of flatlining, driven by data centers, electrification, and reshored manufacturing. Supply is shifting from coal and gas toward solar, wind, and storage. The market clears faster, the weather matters more, and the operator has fewer levers and more knobs. If software does not catch up - real software, the kind that can reason in real time over physical assets - the transition slows. The lights still come on. They just come on more expensively, and with more carbon.
Equilibrium's bet is that the software does catch up, and that they get to write some of it. Eight figures of revenue, nine figures of capital, and a Fortune 500 customer suggest the bet is not crazy. Whether it scales is the next question, and the one the next five years will answer.
The 100-megawatt battery in Texas is still deciding. Charge now or wait twenty minutes. The forecast has shifted. The day-ahead has cleared. The imbalance market is opening. Somewhere in San Francisco, a model that has read every relevant ISO bulletin since lunch is recommending a course of action. A trader is reading the recommendation, glancing at the reason it was made, and either taking it or not.
The battery dispatches. The grid balances. A small amount of carbon is not emitted, and a small amount of money is made. It does not look like a transformation. It looks like Tuesday afternoon. That is the point.