The Tech Industry Is Debating a Radical Solution to AI's Power Problem: Data Centers in Space

The artificial intelligence boom is colliding with the physical limits of planet Earth. As technology giants race to train ever-larger models, they are running headfirst into strained power grids, massive water consumption requirements for cooling, and years-long delays for land permits.[3]

The cloud, it turns out, requires a staggering amount of ground. But a growing faction of aerospace engineers and tech billionaires are proposing a radical solution to the terrestrial bottleneck: taking the cloud off Earth entirely and building data centers in space.[3]

The concept of "orbital data centers" has rapidly moved from science fiction to a multi-billion-dollar corporate battleground. SpaceX CEO Elon Musk has made space-based AI compute a core pillar of his company's future, recently unveiling plans for a massive Texas manufacturing facility dedicated to churning out server-packed satellites.

But the vision is deeply polarizing. On Tuesday, SoftBank Group founder Masayoshi Son publicly dismissed the concept, bluntly telling shareholders that the AI arms race will be won on solid ground, not in orbit.[1][2]

The debate underscores a critical inflection point in the tech industry. With the energy demands of artificial intelligence projected to double by the end of the decade, the infrastructure required to support it is forcing companies to look for power in increasingly extreme environments.

The engineering case for orbital data centers rests on two distinct advantages: infinite energy and infinite cooling. In Low Earth Orbit, roughly 400 to 1,400 kilometers above the surface, satellites can be positioned in sun-synchronous orbits that experience near-continuous daylight.

Without atmospheric interference, weather, or the cycle of night and day, solar panels in space can capture significantly more energy than those on the ground. This allows operators to harvest "stranded" solar energy that never reaches the Earth's surface.

Furthermore, the vacuum of deep space acts as a perfect heat sink. Terrestrial data centers consume millions of gallons of fresh water to cool their densely packed server racks. In orbit, servers can be cooled using passive radiative thermal management—venting waste heat directly into the near-absolute-zero environment of space at zero operating cost.

SpaceX is aggressively pursuing this architecture. Earlier this month, Musk revealed a draft design for the company's first-generation AI satellite. The proposed spacecraft is a behemoth, featuring a wingspan of 70 meters—wider than a Boeing 747.

According to the company's specifications, each satellite is designed to handle 150 kilowatts of AI computing at peak performance. That capacity is roughly equivalent to a modern Nvidia GB300 server rack used in cutting-edge terrestrial facilities.

To achieve scale, SpaceX is constructing a massive 11-million-square-foot "Gigasat" factory in Bastrop, Texas. The facility is intended to mass-produce these orbital compute nodes, with Musk forecasting a goal of deploying one gigawatt of space-based AI computing power per year by 2027.

SpaceX is not alone in the vacuum. A wave of well-funded startups is already proving the hardware can survive the journey. In late 2025, the Y Combinator-backed startup Starcloud made aerospace history by launching a satellite equipped with an Nvidia H100 GPU—the first time a data-center-class chip was operated in orbit.[5]

Other entrants are designing purpose-built satellite payloads with multiple GPU nodes and high-bandwidth laser communication links, aiming to offer commercial AI inference directly from space within the next few years.

Despite the momentum, prominent tech leaders are pouring cold water on the orbital dream. SoftBank's Masayoshi Son argued this week that the fundamental economics of space compute are flawed, noting that electricity is only a fraction of the total cost of running a data center.[1][2]

The true expense lies in the silicon itself—the GPUs and networking hardware. Son pointed out that any savings generated by free solar power would be quickly erased by the exorbitant costs of launching heavy equipment into orbit, the communication latency between space and Earth, and the logistical nightmare of maintenance.[1][2]

In a terrestrial data center, a failed chip can be swapped out by a technician in minutes. In orbit, a dead GPU is permanent dead weight. Analysts at Deutsche Bank estimate that due to these hurdles, orbital data centers will not reach cost parity with ground-based facilities until well into the 2030s.[6]

"He who strikes first wins," Son told his shareholders, emphasizing that the critical window for AI dominance is the next few years. SoftBank is backing up that philosophy with capital, committing tens of billions of dollars to build massive, conventional data centers in places like France, where nuclear power provides a stable, terrestrial energy source.[1][2][6]

For those who agree that land is a bottleneck but view space as too expensive, alternative frontiers are emerging. Startups like the Peter Thiel-backed Panthalassa are developing floating data centers designed to operate in the open ocean.[4]

These maritime facilities aim to generate their own electricity from wave power and cool their servers using cold seawater. Proponents argue that the ocean offers the same isolation from municipal power grids and zoning disputes as space, but at a fraction of the deployment cost and complexity.[4]

Ultimately, the push for orbital data centers reveals the sheer magnitude of the AI industry's physical footprint. Whether the servers of tomorrow are anchored in the ocean, built next to nuclear reactors, or floating in the silent vacuum of Low Earth Orbit, the cloud has outgrown the traditional warehouse.[3]