Why Meta is betting on solar power from space for its AI data centers

Meta Platforms signed an agreement with space energy startup Overview Energy on April 27, 2026, to secure up to 1 gigawatt of power from space-based solar energy for its data centers. The deal marks the first time a major tech company has committed to buying power generated in orbit, underscoring a growing crisis: AI is demanding so much electricity that conventional power sources and grid infrastructure cannot keep up.

Overview Energy, which emerged publicly this year with backing from EQT Foundation, plans to launch satellites into geosynchronous orbit—a fixed position about 22,000 miles above Earth—to collect solar power. The satellites will then beam that power down to Earth using near-infrared lasers, where it will be converted into usable electricity. The company plans to test this end-to-end system in 2028 with a mission closer to Earth.

How the system works

Overview Energy's approach sounds like science fiction, but the core idea is practical. Satellites in space are always in sunlight, unlike solar panels on Earth, which only work during the day and can be blocked by clouds. The satellites collect solar energy and convert it into near-infrared light—wavelengths invisible to the human eye—then transmit it as a laser beam to receiving stations on the ground.

These receiving stations are built onto or near existing conventional solar farms. When the satellite's beam arrives, ground equipment converts the light back into electricity. This approach lets Meta use power infrastructure that already exists, rather than requiring entirely new power plants and grid connections.

The near-infrared wavelengths are considered safe for transmission through the atmosphere and sidestep some regulatory hurdles that earlier space solar concepts ran into. According to the Department of Energy, satellites can capture far more energy than ground-based panels because there's no atmosphere to block sunlight and no day-night cycle to interrupt collection.

The immediate pressure driving the deal

Meta is not waiting passively for space solar to mature. The company has begun deploying its AI computing equipment in temporary tent structures at locations where electrical power already exists, even though standard data center buildings would take longer to construct. This unusual approach reflects a stark fact: AI workloads are growing so quickly that traditional construction and power infrastructure cannot keep pace.

Earlier this year, Meta also signed a 20-year contract with energy company Vistra to purchase nuclear power—about 2,600 megawatts worth, enough to power a major city. In other words, Meta is pursuing multiple strategies at once: tent deployments for speed, nuclear power for stability, and now space solar for long-term capacity.

The broader context here is that AI is pushing infrastructure beyond its normal limits in ways we have actually seen before. During the early days of the commercial internet in the 1990s, telecom companies suddenly needed to lay fiber-optic cables across the country—work that would normally take decades was compressed into a few years because demand was so urgent. Something similar is happening now with electricity: AI demand is forcing utilities, power companies, and tech firms to pursue solutions they would never have considered on a normal timeline.

What this means for power and computing

A 1 gigawatt capacity is substantial. To put it in perspective, that equals the output of a large nuclear reactor or a sizable natural gas plant. However, the actual power delivered will depend on how efficient the satellites are and how often the system operates reliably, facts that remain to be tested at commercial scale.

Meta is not getting exclusive access to Overview Energy's satellites. Instead, the agreement gives Meta early access to some of the capacity, with the understanding that Overview Energy will sell power to multiple customers. This arrangement spreads the enormous cost of building and launching satellites across several buyers, making the project more financially feasible.

Worth noting: the 2028 demonstration date is still years away, so Meta's immediate need for electricity cannot be solved by space solar. This deal functions as a medium-term bet—a way to secure future power capacity—rather than a solution to today's bottleneck. For the next several years, Meta will rely on nuclear deals, terrestrial renewable energy, and that tent-based computing strategy.

The larger shift in infrastructure

What Meta is doing—combining makeshift tent deployments with exotic power sources like space solar—shows how thoroughly AI is disrupting normal infrastructure planning. Data centers traditionally take years to design, permit, and build. But hyperscale companies now operate on quarterly timelines for deploying new AI computing capacity. Those timelines clash directly with the pace at which utilities can expand the power grid.

Meta's two-track approach—quick tactical solutions now, longer-term novel power sources later—acknowledges both the urgent present and the likelihood that the electricity grid will continue to lag behind demand. This pattern suggests that as AI systems grow larger and more demanding, having access to enough reliable power will become as important to competition as the chips themselves.

If space-based solar power actually works at gigawatt scale, it would change where data centers can be located. Today, companies must build near existing power grids. With satellite power beamed directly to the ground, that constraint loosens considerably, potentially opening new geography for data center placement and driving down costs tied to power scarcity.