By 2030, orbital AI computing may undercut Earth-based data centers
A new forecast from the research group 33FG suggests that by the end of this decade, the economics of placing AI computing hardware in space may surpass the cost-efficiency of Earth-bound data centers. The conclusion comes from a detailed comparison of orbital solar power generation and current launch prices, revealing how declining space logistics costs could radically shift the global compute market.Why space may soon beat Earth on energy price per watt
According to 33FG’s modelling, delivering computing hardware and solar power modules to high orbit currently costs around $2000 per kilogram. At today’s price levels, such orbital installations can generate electricity at an estimated $18–26 per watt. Although still more expensive than the $12/W typical for terrestrial data centers, the margin is narrowing faster than previously expected.The key variable is launch cost. If delivery expenses fall by 50%, the price of orbital energy drops to parity with Earth-based electricity. At $500/kg — a threshold analysts believe may soon become realistic — “space power” could become roughly 30% cheaper than traditional energy sources used in AI data centers. At $100/kg, the advantage grows further, reaching an estimated 50% cost reduction per watt.
The role of Starship and orbital refueling
The analysis identifies SpaceX’s Starship as the primary catalyst capable of pushing launch economics into this new territory. Designed for full reusability and powered by in-orbit refueling, the vehicle could reduce delivery costs to the point where orbital computing platforms become financially attractive. Instead of being a distant sci-fi concept, large-scale off-Earth compute infrastructure suddenly looks feasible within a single decade.Multiple missions delivering modular AI hardware, radiators, and large solar arrays could form a network of orbital compute nodes operating above weather patterns and atmospheric losses. With continuous sunlight across most high orbits, satellites avoid the intermittency issues faced by ground solar installations — a major factor in power stability for AI workloads requiring 24/7 operation.
Why AI workloads are a perfect match for orbital environments
AI computation is increasingly dominated by energy consumption rather than chip scarcity. As the cost of training and inference grows, companies are pressured to find cheaper electricity sources and more thermally stable environments. In orbit, heat can be radiated efficiently into space, and solar energy is abundant. This combination creates a natural incentive for energy-intensive AI clusters to migrate off-planet once launch costs fall low enough.Unlike general-purpose cloud data centers, AI compute clusters operate predictably and can be scheduled to align with power availability. This makes them well-suited for orbital solar infrastructure, which offers stable output on high orbits with minimal shadowing. The absence of weather, dust, and atmospheric interference further enhances long-term reliability.
What must happen before orbital compute becomes mainstream
Despite its promise, several challenges remain. Hardware upgrades are more difficult in orbit, debris management requires strict protocols, and network latency may limit which workloads can be executed off-planet. However, analysts argue that AI training — especially batch workloads — is far less latency-sensitive than user-facing cloud services, making it an ideal candidate for orbital deployment.The cost curve is the decisive factor. If Starship or other next-generation reusable launch systems achieve the projected price milestones, the economics flip rapidly. Terrestrial data centers would struggle to compete with orbital installations that have near-zero ongoing power costs and unmatched solar exposure. This could reshape global compute geography and push high-value workloads beyond Earth’s surface.
Conclusion: the decade when space becomes the cheapest place to compute
33FG’s projection paints a future in which the frontier of AI computation is not just cloud platforms or domestic supercomputers — but orbital arrays operating above the planet. With launch costs falling and solar infrastructure maturing, space may soon offer the lowest-cost environment for large-scale AI processing. If current trends hold, the first commercially viable orbital AI clusters could appear as early as 2030.Editorial Team — CoinBotLab