Meta Reserves 1 GW of Space Solar to Power Its AI Data Centers

Meta announced two energy partnerships this week aimed at powering its expanding fleet of AI data centers: a deal with Overview Energy for up to 1 gigawatt of space solar power, and a separate agreement with Noon Energy for up to 1 GW / 100 GWh of ultra-long-duration energy storage.

The space solar piece is the headline. Overview Energy plans to put satellites in geosynchronous orbit roughly 22,000 miles above Earth, where the sun never sets, and beam the collected power down to existing terrestrial solar farms.

The combined commitment makes Meta one of the most aggressive corporate buyers of next-generation clean energy infrastructure — and a real-world test case for whether exotic technologies can scale fast enough to keep pace with AI's compute demands.

Context: Why AI Just Broke the Power Grid

Training and serving frontier AI models is enormously power-hungry. A single hyperscale AI data center can pull more than a gigawatt — comparable to a mid-sized city. Meta, OpenAI, Microsoft, Google, and Amazon have all reported that their data center buildouts are now constrained primarily by power availability, not chips or land.

The traditional fixes — natural gas turbines, more grid interconnections, nuclear power purchase agreements — are getting harder to come by. Interconnection queues in the U.S. now stretch five to seven years in many regions. Meta has already responded by contracting more than 30 GW of clean energy across solar, wind, and nuclear, and is one of the largest corporate nuclear buyers in U.S. history with 7.7 GW under contract through Vistra, TerraPower, Oklo, and Constellation.

The space solar and long-duration storage deals announced this week target a different problem: not how much power, but when. Solar farms produce nothing at night. Wind is intermittent. AI workloads run 24/7 and don't pause for clouds.

What's Actually in the Deals

The two partnerships address different technical challenges:

Overview Energy (space solar). Overview's satellites will collect solar energy in geosynchronous orbit — where sunlight is constant and unfiltered by atmosphere — and transmit it as microwaves to receiver stations co-located with existing solar farms on Earth. Meta has reserved capacity for up to 1 GW of delivered power. The first orbital demonstration is planned for 2028, with commercial delivery potentially starting around 2030.

Noon Energy (long-duration storage). Noon's reversible solid oxide fuel cell architecture uses carbon-based storage to deliver more than 100 hours of discharge — far longer than lithium-ion batteries, which typically max out around 4-8 hours. Meta has reserved up to 1 GW / 100 GWh of capacity. An initial 25 MW / 2.5 GWh pilot demonstration is targeted for 2028.

Together, the two technologies are complementary. Long-duration storage handles the daily and weekly cycles where existing renewables fall short. Space solar — if it works at scale — addresses the structural problem that ground-based solar is unproductive for half of every day.

Industry Impact

For Meta, the deals are a signal that the company is willing to spend on speculative infrastructure to lock in long-term power optionality. None of this technology will be delivering electrons in 2026 or 2027. The bet is that 2028-2030 will be when the first commercial volumes arrive — exactly when Meta's planned AI capacity hits another step-change.

For Overview Energy and Noon Energy, the agreements are validation. Space solar in particular has been a niche idea for decades, with skeptics arguing that launch costs and beaming efficiency would never make it competitive with ground-based renewables. Falling launch costs (largely thanks to SpaceX and Rocket Lab) and improved photovoltaic efficiency have changed the math, but the economics remain unproven at scale. Meta's commitment, even on a reservation basis, is the largest corporate vote of confidence the technology has received.

For the broader AI sector, the news reinforces an emerging pattern: the largest AI labs are no longer waiting for utilities to solve their power problem. Microsoft restarted Three Mile Island. Amazon bought a nuclear-adjacent data center campus. Google has signed multiple small modular reactor deals. Meta is now reaching past nuclear into orbital infrastructure.

What's Next

Both projects have aggressive but realistic timelines. Overview Energy's first orbital demonstration in 2028 will be the first real test of whether space solar can deliver meaningful power at commercial efficiency. If the demo succeeds, Meta's 1 GW reservation gives Overview a credible anchor customer to scale into.

Noon Energy's 25 MW pilot in 2028 is technically less novel — long-duration storage technologies already exist — but the carbon-based architecture and 100-hour duration would set a new commercial benchmark if it performs. Expect early data from the pilot to flow back into the broader storage market quickly.

The deals are also non-exclusive. Meta will keep buying nuclear, wind, solar, and battery capacity in parallel. The space solar and long-duration storage commitments are additional, not substitutes.

Bottom Line

AI's growth is now constrained by physics — specifically, by how much electricity utilities can deliver to data centers and when. Meta's response is to write checks for technologies that don't exist yet at scale, betting that orbital solar and 100-hour storage will be commercial within five years. If you're tracking AI infrastructure, watch the 2028 demonstrations closely. They'll be the first real evidence that the "exotic energy" thesis isn't just a press release.