Make smarter decisions on the data center on-site power market

Data center on-site power is very hot right now, because the grid can't connect AI campuses fast enough. The bottleneck has simply moved to the equipment makers.

We've found a lot of interesting signals. For example, Brookfield agreed in October 2025 to invest up to $5B to deploy Bloom Energy's fuel cells at AI data centers. And in December 2025, Boom Supersonic, the startup building a supersonic airliner, unveiled a 42 MW gas turbine for data centers, with a first order of 1.21 GW from an AI cloud company. Builders want power on site so badly that big money is pouring in and new suppliers are coming from outside the power industry.

In this section, we look at why so many AI campuses now plan their own power, which technologies they pick first, and what long waits for turbines mean for the rest of the market.

An AI campus can be ready years before the grid can feed it. Every month of waiting leaves very expensive chips doing nothing.

Hyperscalers, AI labs, GPU clouds, colocation developers and utilities all feel this differently, from turbine lead times and gas supply to air permits and noise. We compare six types of companies and rank the pains that matter most to each.

This shows which problems buyers will pay extra to solve now, and which ones they'd rather leave to the grid.

Selling engines and turbines, renting mobile power by the megawatt, signing long power contracts, building and owning the plant, or running it as a service: on-site power companies earn money in many ways.

We compare who pays and who carries the risk. Equipment makers get paid upfront but depend on factory capacity. Developers that own the plant sell power for years and carry fuel and financing risk. Rental fleets charge more per megawatt-hour but can be sent home once the grid arrives.

You'll see which models still make money when the grid finally connects the campus, and which ones lose their customer that day.

Growth here depends less on new technology than on how slowly grids expand. We go through each force behind it.

The obvious drivers are bigger AI campuses, long grid queues and chips that need steady power around the clock. Less obvious ones matter too: lenders who now accept on-site plants as part of a project, cheap gas in places like Texas, and fuel cells and engines that can be added in small steps. Several of these feed each other, and we show how.

Before building or investing, you should know what makes a data center builder generate its own power, and what makes it wait for the grid.

Engine makers like INNIO get much of the attention in on-site power, but the market goes far beyond them.

Some of the most interesting players stay behind the scenes. Some rent fleets of trailer-mounted generators that can arrive in months. Others write the software that balances engines, batteries and solar, or rebuild jet engines into power plants.

They're hard to spot because most deals are private. A campus rarely names the company running its power plant, and many projects mix several suppliers under one developer.

Below, you'll find the market grids and ecosystem maps we built to show who supplies the equipment, who owns the plants, and where new players are getting in.

VoltaGrid, Williams, Caterpillar… a few companies are already powering AI campuses on site, or are close to it.

VoltaGrid deploys modular gas power plants and agreed to supply Oracle with 2.3 GW in Texas. Williams builds dedicated gas plants for big tech campuses, and its first one, for Meta in Ohio, is nearly finished. Caterpillar sells the engines and generators found at many sites. Few of these projects have run for years yet, so their long-term returns are still unproven.

We'll look at how these companies won their first campuses, secured equipment and fuel ahead of rivals, and planned for the day the grid arrives.