SpaceX eyes orbital data centres for AI surge

Plans taking shape inside SpaceX point to an unusual next step in the global race for computing power: data centres operating in orbit by 2026. The company has begun recruiting specialists in power systems, thermal engineering, networking and on-orbit operations, signalling an effort to pair its satellite infrastructure with space-based computing at a time when demand for artificial intelligence processing is colliding with limits on Earth.

People familiar with the hiring say the work focuses on designing modular compute platforms that could be launched, assembled and upgraded in space. The concept leans on two existing pillars of the company’s business: the broadband satellite constellation Starlink, which already provides global connectivity, and the heavy-lift launch system Starship, built to place unusually large payloads into orbit at lower marginal cost.

The move comes as cloud providers and AI developers grapple with electricity shortages, land constraints and regulatory scrutiny tied to terrestrial data centres. Training and running large language models requires dense clusters of processors that draw vast amounts of power and produce intense heat. In many regions, grid upgrades are lagging demand, while community resistance and water usage concerns are complicating approvals for new facilities. Space offers theoretical advantages: near-constant solar energy, natural radiative cooling, and freedom from zoning and land acquisition hurdles.

Engineers working on the project are understood to be exploring designs that rely primarily on solar arrays for power generation, with energy storage to handle orbital night cycles. Thermal management is central, as heat must be dissipated through radiation rather than air or liquid cooling. The absence of gravity changes how fluids behave, forcing a rethink of cooling loops that are routine on Earth. Reliability is another challenge, with systems expected to operate autonomously for long periods and withstand radiation exposure.

While the idea of orbital data centres has circulated in academic and industry discussions for years, SpaceX’s involvement gives the concept new credibility because of its integrated capabilities. Few organisations can combine frequent, lower-cost launches with a global satellite network and in-house spacecraft manufacturing. Analysts note that the company’s internal demand could also play a role, as satellite operations, imaging and network management increasingly rely on advanced analytics and machine learning.

The commercial rationale extends beyond internal use. If viable, space-based computing could offer low-latency processing for satellite data, secure environments for sensitive workloads, and services for regions where ground infrastructure is weak or politically constrained. For AI developers, it could represent an alternative supply of compute capacity at a time when competition for advanced chips and power contracts is intense.

There are, however, significant uncertainties. Launch costs, though falling, remain far higher than building on land, and any hardware failure in orbit is expensive to fix. The economics depend on Starship achieving regular, reliable flights with payloads large enough to amortise costs. Regulatory frameworks are also evolving, covering spectrum use, space traffic management and the environmental impact of increased launches and orbital assets.

Environmental considerations cut both ways. Supporters argue that moving some compute workloads off-planet could reduce pressure on water resources and grids. Critics counter that higher launch rates increase emissions and debris risks, and that space-based systems may eventually need disposal plans to avoid worsening congestion in low-Earth orbit.