Meta Is Beaming Solar Power From Space Now. And That's Not Even the Crazy Part.
So Meta announced two energy deals recently on April 27, 2026 and honestly I had to read the headline twice because I thought it was from a science fiction newsletter. Satellites beaming power from space to solar farms on Earth. Batteries that can store electricity for over 100 hours.
Let me explain what actually happened and why it matters, because the coverage I've seen is either too technical or just hype. I'll try to make it simple.
The basic problem is this: Meta's AI is absolutely eating electricity. In 2024 alone, their data centers used more than 18,000 gigawatt-hours of power. To give you some idea of what that number means, that's roughly enough electricity to power about 1.7 million American homes for a full year. And that was 2024. The AI workloads have only gone up since then. So Meta has been scrambling to find clean, reliable power that doesn't disappear when the sun goes down or the wind stops blowing. These two new deals are their answer to that or at least the beginning of one.
The Space Solar Deal (Yes, Really)
The first partnership is with a startup called Overview Energy, based in Ashburn, Virginia. They only came out of stealth mode in December 2025, so they're a very new company in terms of public profile.
Here's what they're building. Overview wants to put satellites in orbit that collect solar energy in space, where there are no clouds, no night, no atmosphere getting in the way. Then they convert that energy into near-infrared light and beam it down to solar farms on Earth. The solar farms pick up that light and turn it into electricity. So you get solar power at 2am. Or on a cloudy Tuesday. Or in the middle of winter. The satellite doesn't care, it's above all of that.
I know what you're thinking. Lasers from space? Isn't that dangerous? Actually, Overview's CEO Marc Berte says the beam is wide and low-intensity infrared, not a focused laser. He told TechCrunch you can literally stare into it with no ill effects.
The clever part of the whole design is that Overview isn't asking Meta to build new infrastructure. They're using solar farms that already exist. The satellite supplements them. Keeps them producing at night. Meta's existing solar investment doesn't become useless after sunset, it works around the clock. That's a much easier sell than "tear everything down and build something new."
Meta has agreed to take up to 1 gigawatt of power from this system. The orbital demonstration is planned for 2028, and if it works, commercial delivery to the US grid could start as early as 2030. So this isn't something you'll see powering data centers next year. But the fact that Meta is willing to put money behind it now, before it's proven at scale, says something about how urgent the energy situation is for AI companies.
The Battery Deal That's Actually More Immediately Interesting
The second deal, signed April 22, is with Noon Energy, a California-based startup founded in 2018.
Here's the problem with normal batteries. Your typical grid-scale lithium-ion storage, the stuff deployed everywhere over the past few years, maxes out at around four hours. Maybe eight or twelve if you have a newer system. That's fine for smoothing out peak demand in the evening after a sunny day. It does nothing for a stretch of cloudy weather lasting three days, or a week where wind just doesn't blow.
Noon Energy's system is designed for 100+ hours of storage. Some of their pilot systems have demonstrated 200 hours. Let that sit for a second. That's over eight days of stored power.
How do they do it? Their technology is a reversible solid oxide fuel cell. When there's extra electricity from solar or wind, the system converts it into a carbon-based chemical fuel and stores it. When you need power back, you run the process in reverse, using oxygen from air to convert the stored fuel back into electricity. It's modular, you can keep adding storage capacity as the data center grows. And it uses carbon and oxygen, not lithium, not cobalt, not the rare metals that make regular battery supply chains complicated.
Meta has reserved up to 1 gigawatt of this, 100 gigawatt-hours of storage capacity. That's one of the largest commitments to ultra-long-duration storage in history. The first project is a 25 megawatt pilot, expected to be finished by 2028. Noon Energy says this is probably going to be their first commercial deployment. So the deal is basically making them a real company in that sense.
One thing I want to flag though: Noon is still in pilot stage. The 200-hour storage claims come from their demo projects, not commercial deployments at scale. Meta is betting on this working. That's not the same thing as it having already worked. Nobody knows if it scales the way they're hoping.
Why This Is Actually a Big Deal for Meta's AI
People sometimes think "Meta" and think Facebook or Instagram, but the energy story is really about Llama, their AI models, and the infrastructure required to train and run them. Meta is competing with OpenAI, Google, and others for AI dominance, and the compute arms race has a direct power bill attached to it.
Meta has committed to building 30 gigawatts of renewable power capacity. They were also ranked as the largest corporate clean energy buyer globally in 2025, contracting 10.24 gigawatts of clean energy in that single year alone. They also have agreements for 7.7 gigawatts of nuclear energy across deals with Vistra, TerraPower, Oklo, and Constellation Energy. So these two new deals aren't random experiments, they're part of a massive, coordinated effort to make sure their AI operations don't end up dependent on fossil fuels or the unpredictability of the regular grid.
The reason that matters so much is that AI workloads can't really pause. If you're training a model that takes weeks of compute, you can't stop for three days because there's no wind and the batteries ran out. You need always-on power. Short-duration batteries can bridge a few hours. Noon's technology can potentially bridge multiple days. That's a fundamentally different kind of reliability.
And the Overview partnership solves a different piece of the same puzzle. Meta has been building a lot of solar capacity, but solar only works when the sun shines. If you can supplement those farms with orbital solar that beams down light at night, your existing infrastructure becomes more useful overnight, literally.
Okay But Is Any of This Real
Yes and no. The deals are real. The technology is real in the sense that it exists and has been demonstrated at small scale. But both partnerships are targeting 2028 for their initial demonstrations. That's two years away. A lot can go wrong. Overview Energy is a four-year-old startup that was in stealth until six months ago. Noon Energy raised only about $45 million in total funding before this deal, that's not nothing but it's not big for a company that needs to eventually build gigawatt-scale manufacturing.
Meta's VP of energy and sustainability Nat Sahlstrom said something I thought was pretty honest for a corporate press release: "Both technologies are early, and that's exactly why they're worth supporting now." That's not the usual "this will change everything" language. It reads more like a calculated bet, if even one of these works at scale by 2030, Meta gets a real energy advantage over competitors who waited.
So yes it's real. And yes it's genuinely interesting technology. But I wouldn't say the problem is solved. Meta is buying options on the future, not flipping switches.
What This Means for Everyone Else
The part that doesn't get enough attention is what happens when a company with Meta's resources bets on an unproven startup. Noon Energy, after this deal, suddenly has reason to build out production capacity for a gigawatt-scale supply chain. Overview Energy now has a named customer with a target of 1 gigawatt when they prove out their orbital system. Both of these things make the underlying technology more likely to actually get built, regardless of whether Meta ends up as the main customer.
For now, if you're curious whether Meta's AI is going to run on space solar power next year, NO. That's a 2028 at earliest, 2030 more realistically kind of story. But for a company burning through 18,000+ gigawatt-hours per year and growing fast, putting money into this now is probably the only sensible thing to do.
The power problem for AI is real. These deals don't solve it yet. They're just the most interesting attempt I've seen so far.