Meta working on tech that can solar power AI even when there's no Sun

As artificial intelligence continues to scale at an unprecedented pace, the energy demands behind it are becoming impossible to ignore. Consequently, Meta is now turning to experimental but potentially innovative solutions, announcing new partnerships aimed at rethinking how clean energy is generated and stored.

In a new initiative, the company has teamed up with Overview Energy and Noon Energy to tackle two of renewable energy’s biggest constraints: intermittency and storage. The goal is to build an energy backbone capable of sustaining AI infrastructure at scale while reinforcing grid reliability. This comes as data centre energy demands surge, with projects like the proposed 9GW "Stratos" campus in Utah highlighting the scale of power infrastructure being built to support AI and cloud computing.

Traditional solar energy comes with an obvious limitation; it only works when the sun is shining. That leaves vast infrastructure underutilised for large portions of the day. Overview Energy’s approach aims to change that by moving solar collection into space. The company is developing satellites positioned in geosynchronous orbit. From there, they should be able to capture uninterrupted sunlight and beam energy back to Earth. Ground-based solar farms would then convert that signal into electricity, effectively allowing them to operate even at night.

Reportedly, Meta has already reserved up to 1 gigawatt (GW) of capacity from this system. An orbital demonstration is planned for 2028, with early commercial deployment potentially beginning around 2030.

On the storage side, Noon Energy is developing long-duration storage system designed to hold energy for over 100 hours, far longer than typical lithium-ion batteries. The technology uses solid oxide fuel cells combined with carbon-based storage to extend how long renewable energy can be used. Under the partnership, Meta has reserved up to 1GW/100 gigawatt-hours (GWh) of storage capacity, with a pilot project expected by 2028. If proven viable, this approach could enable continuous clean energy supply for data centres and AI operations.

These partnerships are part of a wider push by Meta to secure long-term, sustainable energy sources for its growing infrastructure footprint. Both space-based solar and ultra-long-duration storage remain early-stage technologies, with key demonstrations still years away. However, if these bets pay off, they could unlock a step-change in how energy is produced, distributed, and stored, helping solve one of the biggest bottlenecks facing the AI era.