Meta is gambling on space-based power generation! Locking in Overview Energy's future production capacity to "feed" AI data centers.

Meta Platforms (META.US) is seeking to power its artificial intelligence (AI) data centers with CECEP Solar Energy collected in space in a novel way to meet its almost endless power demands. Meta announced on Monday that it has reached an agreement with the startup Overview Energy to obtain up to 1 gigawatt of solar power from space. 1 gigawatt of power is roughly equivalent to the generating capacity of a nuclear reactor. Meta declined to disclose the financial terms of the agreement. According to the agreement, Meta will have priority access to Overview Energy's future generation capacity. Both parties expect commercial power supply to be achieved by 2030. Nat Sahlstrom, Vice President of Energy and Sustainability at Meta, stated that the company hopes Overview Energy can provide them with clean and uninterrupted energy. Overview Energy plans to collect sunlight using satellites orbiting the Earth and convert it into power to support grid operations. The company's vision of a space-based power station that can transmit energy back to Earth is still in the theoretical stage. The startup is continuing to develop and test its underlying technology and plans to conduct its first orbital demonstration in 2028. Overview Energy is betting on the idea of "never losing the sun in space." If their efforts to continuously collect and transmit energy back to Earth are successful, they will offer a solution for Clean Energy Fuels Corp to overcome some limitations of ground-based solar panels, as ground-based solar energy generation efficiency is usually affected by weather, day and night cycles, and seasonal changes. For Meta, this move is part of its large-scale AI investment spree. The company is investing billions of dollars to ensure the energy, infrastructure, and computing power capabilities needed to support its AI plans. Meta's AI ambitions have largely relied on natural gas so far, as the company believes natural gas is more stable and reliable compared to some cleaner forms of energy. Meta is currently building its largest AI data center campus in rural Louisiana, USA, and supporting the development of 10 new gas-fired power plants for the campus. With the exponential growth in AI's demand for computing power in recent years, the "three highs" pain points of ground-based computing centers have long been an industry consensus: large data centers consume as much electricity annually as a medium-sized city, cooling costs account for 30% of total operating expenses, and coverage blind spots in remote areas, oceans, and airspace are challenging to overcome. The space environment provides a natural solution - the efficiency of CECEP Solar Energy in space without atmospheric obstruction is over 30% higher than on the ground, and zero-cost cooling can be achieved in a vacuum environment, and low-earth orbit satellite networking can achieve millisecond-level global coverage. However, the costs of satellite launch and construction are currently high, and there are no practical projects yet. It is worth mentioning that Meta is not the only company trying to meet the energy needs of data centers in space. Musk, Bezos, and others have previously discussed deploying computing facilities in orbit. Musk has announced the TeaFab chip project, which will achieve an annual capacity of 1 terawatt (TW) of AI computing power, with plans to start production in 2027, with 80% of the chips being used for space computing centers. Earlier this year, Musk's commercial space giant SpaceX submitted an application to the Federal Communications Commission (FCC) to deploy up to 1 million satellites, aiming to build the world's first "orbital data center network" to provide space computing power for AI models. In addition, reports indicate that Bezos' space exploration company Blue Origin announced last month its entry into the field of space AI infrastructure. The reports state that on March 19, Blue Origin submitted official application documents to the FCC seeking approval for its large satellite network deployment plan, planning to deploy a total of nearly 52,000 highly specialized satellites in Earth orbit. Unlike traditional communication or navigation satellites, this large fleet of satellites is specifically designed to handle complex artificial intelligence computing tasks and will carry advanced processing chips to directly perform high-intensity AI data computations in space.