The Startup Selling Data Centers Their Own Private Power Grid

There is a race happening in the power business, and the prize is the AI boom. Data centers are devouring electricity at a pace the American grid was never built to handle, and the companies racing to build them face an uncomfortable math problem: new power plants take years, and the reliable ones are getting harder to buy. This week, a startup founded by two SpaceX alumni made its boldest claim yet about how to solve it: skip the grid entirely.

Ambrosia Energy announced on Tuesday that it has begun work on a 10 gigawatt-hour battery factory near Austin, Texas. The 150,000-square-foot facility is expected to reach full production within two to three years, and co-founder and CEO Ben Longmier said it will let the company build about 300 megawatts of off-grid power plants per year. The customer is a familiar one: data centers. The fuel is solar panels paired with in-house battery packs.

The announcement, made onstage at TechCrunch Disrupt in San Francisco, marks the startup's transition from ambitious pitch to industrial-scale bet. Ambrosia emerged from stealth in June with backing from DFJ Growth and a thesis that sounded almost too simple: pair solar panels with lithium-ion batteries, and sell the combination to hyperscalers at $100 per megawatt-hour, cheaper and faster than a natural gas plant.

The pitch, in one number

That number is $100 per megawatt-hour. It is the price Ambrosia says it can charge for round-the-clock electricity from solar plus storage, and it matters because of what it sits next to. Financial advisory firm Lazard estimates that a new combined-cycle gas turbine, the most efficient kind, costs about $107 per megawatt-hour to build and operate. In other words, Ambrosia claims it can undercut gas on price while delivering cleaner power.

The speed comparison is even sharper. Gas turbines currently carry a five- to seven-year order backlog. Ambrosia's ambition, in the words of co-founder and president Sara Spangelo, is a plant that goes from contract signing to power on in 12 months, at any scale. In a market where hyperscalers are signing decades-long deals for power that may never arrive on time, a 12-month path is not a marginal advantage. It is a different business.

How you make solar run all night

The engineering argument rests on two choices. The first is to stop cycling batteries the way the rest of the industry does. Most grid-scale batteries charge and discharge over two- or four-hour periods, a fast cycle that stresses the cells. Ambrosia instead trickle-charges its batteries through the day and discharges them slowly overnight, reducing wear on the system.

The second is vertical integration. Ambrosia buys battery cells from outside suppliers but builds its own packs in-house, and Longmier says the packs contain roughly one-hundredth the parts of competing systems. That simplification has pushed the cost of the full storage system down to about 1.3 times the price of the bare cells, compared with five to seven times for typical stationary storage. The startup has taken the same simplification approach to solar mounts and other components.

Together, these choices are what make the $100 figure plausible even with massive overbuilding. Ambrosia's first commercial plant, under construction in West Texas, pairs 300 megawatts of solar panels with 660 megawatt-hours of batteries to deliver just 30 megawatts of continuous power. That is a ten-to-one ratio of solar nameplate capacity to delivered output: most of the panels exist to guarantee the last few cloudy days, not to serve the load on an average day. The economics only work because solar panels and batteries have become so cheap that the extra capacity costs less than waiting seven years for a gas turbine.

Ambrosia's bet is that the cheapest way to keep a data center running all night is to buy ten times more solar panels than you need.

What is already running

The company is not working entirely on paper. Ambrosia began building its pilot-scale plant in West Texas in January, and Longmier said sections of it have now operated continuously for more than 100 days, generating 200 kilowatts of electricity around the clock. The first full commercial plant is beginning work later this year, targeting 99.3 percent uptime.

Two people are driving it. Longmier and Spangelo previously founded Swarm, a satellite startup they sold to SpaceX, and both later worked on Starlink. Spangelo says the playbook is familiar: building power plant modules is, in her words, kind of like deploying a satellite constellation, you launch four, you learn, you iterate. The company has said it wants to reach gigawatt scale by the end of the decade, and Longmier has floated partner land access that could, in theory, support a 30-gigawatt plant.

Why it matters beyond one startup

The deeper story is structural. The United States built its power grid around predictable, slow-growing demand, and the AI buildout has blown past those assumptions in a matter of years. Grid interconnection queues stretch for years in many regions, and PJM, the grid operator for much of the eastern U.S., has taken emergency measures tied to data center demand. In that environment, a company that can plant a power station next to a data center and turn it on in a year is not just selling electricity. It is selling time.

The skeptics have real ammunition. A 99.3 percent uptime target sits well below the availability levels major cloud providers promise their own customers, and Ambrosia has not yet proven it can hold that rate through a bad winter at gigawatt scale. The economics depend on Texas sun and cheap land; replicating them in cloudy, expensive regions is an open question. And $100 per megawatt-hour is still a company target, not a delivered invoice.

But the directional signal is hard to dismiss. A decade ago, the idea that solar plus batteries could compete with gas on firm, 24-hour power would have been laughed out of a utility boardroom. Now a startup is building a battery factory on that exact premise, with venture backing and a pilot that has run for a hundred days. Whether Ambrosia itself succeeds, the price of waiting for the grid is falling fast.