At 4 p.m. on a punishing Texas afternoon, electricity stops behaving like an ordinary commodity. Demand climbs with every compressor groaning against the heat. Wind output may fade. A transmission line can clog. A megawatt that looked routine at breakfast can become painfully expensive by dinner. Somewhere inside that commotion, a battery owner must decide whether to sell now, hold back, or promise capacity to a different market entirely. Gridmatic has built a company around making that decision before the crowd does.
The Cupertino company calls itself an AI-first power company, which is a more useful description than the familiar label of climate software. Gridmatic does build software: weather models, demand forecasts, price forecasts, optimization systems and automated bidding tools. But it also signs power contracts, supplies electricity to commercial customers, schedules batteries and assumes financial risk. A forecast is not the finished product. The finished product is a battery dispatched, a clean-energy contract priced, or a bid accepted in a live wholesale market.
That distinction gives the business its edge and its exposure. Many enterprise AI products advise a human who remains responsible for the result. Gridmatic's machines operate closer to the money. If its models misunderstand tomorrow's weather or congestion, the error can show up in a trading account. When the forecast is right, the company and its customers can capture the gain.
From keyword auctions to kilowatt-hours
Founder Matt Wytock arrived at energy by way of another automated marketplace. At Google, he worked on core Search and Ads systems and created Dynamic Search Ads, a product that reduced the need for advertisers to select every keyword by hand. He later earned a machine-learning doctorate at Carnegie Mellon, where his research included large-scale convex optimization methods for the grid. Gridmatic says he founded the operating company in 2017 after concluding that electricity-market reform was essential to decarbonization.
The analogy between advertising and electricity is imperfect but revealing. Both markets clear repeatedly. Both digest huge numbers of signals. Both punish slow reactions. Power adds physics: transmission constraints, battery state of charge, weather, ramp rates and an unforgiving requirement that supply equal demand. Search ads can wait for the next user. A power grid cannot.
“The power company for an increasingly complex grid.”Gridmatic's deliberately plain description of an unusually technical business
Gridmatic trains deep-learning models to forecast prices at thousands of locations in ERCOT, CAISO, PJM, MISO, SPP, NYISO and ISO New England. Its own weather models feed predictions of renewable output, demand and price. Optimization software then weighs those forecasts against the physical limits of a battery or flexible load. Automated systems submit bids around the clock.
A battery is a portfolio of moments
A grid battery looks simple from the highway: rows of metal boxes behind a fence. Commercially, it is a shifting portfolio. It can buy energy when prices are low, sell when they rise and reserve capacity for services that stabilize grid frequency. Every commitment consumes some combination of charge, time and battery life. The optimizer must also account for market rules that differ sharply by region.
This is where Gridmatic competes with specialist firms such as Habitat Energy, Fluence Mosaic, Stem and other trading desks. Its difference is breadth. It offers bidding and dispatch, risk management, settlement, reporting and qualified scheduling entity services. It can sign tolling agreements with owners, create a revenue floor, share market upside and bring capital from its $50 million Energy Storage Fund. That stack is harder to buy as a tidy software subscription, but it can give developers the predictable cash flows lenders like.
Results are unusually visible in Texas, where ERCOT publishes participant data. Gridmatic says it produced the highest cumulative day-ahead trading profit in that market from 2018 through 2025. In California, the evidence is murkier because CAISO anonymizes asset-level performance. Gridmatic built a shadow-clearing engine to reconstruct the activity of 30 batteries during 2024 and 2025. Its 2026 report found monthly energy and ancillary-services revenue ranging from below $1 to above $6 per kilowatt, with only five assets beating a theoretical time-based benchmark.
The company's optimized Caballero project, a 100 MW/400 MWh battery owned by Alpha Omega Power and Fengate, led that study at a reported 141 percent of its benchmark. Gridmatic calculated that the whole 30-asset fleet would have earned $98 million more if every battery had matched Caballero's capture rate. It is the kind of claim competitors will inspect closely, but the underlying observation is difficult to dismiss: hardware placed in the same market can produce dramatically different results.
The customer is becoming part of the grid
Gridmatic's second act moves behind the meter. Gridmatic Retail supplies commercial and industrial customers in Texas, Ohio and Pennsylvania. The ordinary service is electricity. The more distinctive products match power consumption with renewable generation by the hour, forecast a customer's load, and use flexible demand as something that can be traded rather than merely billed.
Data centers make the logic vivid. They consume large, steady quantities of power and increasingly face both cost pressure and carbon commitments. With EdgeConneX, Gridmatic ran a multi-year 24/7 carbon-free-energy program for a one-megawatt Houston facility. The companies tracked clean supply close to real time; Gridmatic says the program exceeded 85 percent hourly carbon-free matching. A later 10 MW solar agreement with Sol Systems added Texas supply to Gridmatic's portfolio and supported the renewed EdgeConneX relationship.
Forecast, bid and dispatch batteries; settle with the market; manage downside; share upside.
Supply power, hedge volatility, match renewable output and monetize flexible consumption.
Use tolls and revenue floors to make storage cash flows easier to finance.
Gain exposure to storage operations without developing and managing an entire project.
The 2025 AI Load Optimizer pushes further. Large users such as data centers and bitcoin mines can adjust some operations when market conditions change. Gridmatic forecasts the spread between day-ahead and real-time prices, submits bids and automates participation. Data Factory, an early customer introduced through Vega Energy Advisors, said the system found value with almost no lift from its staff. Gridmatic says suitable users can cut energy costs by more than 5 percent.
Credit is part of the product, too. Traditional suppliers may demand 45 to 60 days of collateral from volatile, high-load businesses. A 2026 collaboration with OBM combines daily settlement and automated load control with Gridmatic's forecasts. Paying daily reduces the supplier's exposure and the customer's working-capital burden. It is a small operational change with a larger implication: energy optimization is as much about contracts and cash timing as algorithms.
The moat is accountability
Gridmatic sits between several categories that usually remain separate. It resembles a quantitative trading firm when it forecasts nodal prices, an energy-services company when it dispatches batteries, a utility when it supplies a factory, and an asset manager when it deploys fund capital. That mixture complicates the business. Regulatory licenses matter. Market errors are expensive. Growth consumes working capital. A software rival can ship a feature; a power marketer must also settle the invoice.
But those inconveniences may be the point. Gridmatic learns from positions it actually trades and assets it actually operates. The feedback loop connects model quality to measured financial outcomes. Its teams combine alumni of Google, Meta and Apple with veterans of Shell, NextEra, NRG and other energy firms. The culture described on its careers page is unusually engineer-led: no dedicated product managers or designers, end-to-end ownership, reproducible backtests and a preference for automating the industry's stubborn manual work.
A power forecast becomes valuable only when someone can act on it, finance it and live with the result.
The market around Gridmatic is widening. Solar and battery costs have fallen. Electricity demand is rising again under the weight of data centers, manufacturing and electrification. Meanwhile, weather-dependent generation makes the timing of supply more consequential. Traditional utilities, commodity traders, battery manufacturers and software specialists are all pursuing pieces of the same opportunity.
Gridmatic's wager is that the winning power company will orchestrate a fleet it does not necessarily own: wind and solar under contract, batteries operated for developers, and customers willing to flex demand. The company is not removing volatility from electricity. It is trying to forecast that volatility, package it into contracts and make it useful. On a grid that changes every five minutes, better timing may be the closest thing to a power plant.