As AI growth strains terrestrial power and water resources, tech giants are looking toward two radical frontiers: orbit and nuclear fission . While companies like SpaceX explore satellite-based processing, others are doubling down on Earth-bound nuclear energy to sustain massive computing demands.
Starcloud-1 and the dawn of orbital AI
In 2025, the startup Starcloud demonstrated the viability of orbital processing by launching Starcloud-1, which handled queries from Google’s Gemma LLM. this proof of concept is part of a larger movement toward space-based infrastructure designed to alleviate the strain on Earth's digital ecosystems.
Google is also developing Project Suncatcher, which intends to use a satellite constellation in a dawn-dusk sun-synchronous orbit to harness solar energy. This project aims to deploy Tensor Processing Units in space to handle massive computational loads through wireless interconnection and microwave links to Earth.
The -250°F cooling crisis in vacuum
Operating AI hardware in the vacuum of space introduces extreme thermal management difficulties. While space temperatures can drop to minus 250 degrees Fahrenheit , the absence of an atmosphere means traditional cooling methods like conduction and convection are impossible. The lack of air means heat cannot be moved away from chips via air, forcing a total reliance on infrared radiation.
According to the report, companies must rely on large surface area radiators to dissipate heat through radiation alone. Additionally, cosmic rays pose a theat to hardware, as ionizing radiation can damage solar panels, heat radiators, and processing chips.
SpaceX's 2028 target for 100 kW of compute
SpaceX has signaled its intention to enter the orbital compute market through its May 2026 IPO filing. The company aims to deploy AI compute satellites by 2028, targeting a density of over 100 kW of compute per metric ton.
However, SpaceX has noted that these technologies, alongside lunar economy development and human augmentation systems, remain unproven and highly complex.. This ambitious roadmap places SpaceX at the center of a potential new economy involving both orbital computing and lunar development.
Microsoft's $16 billion bet on Three Mile Island
While space offers a long-term frontier, the immediate solution for AI's energy hunger appears to be nuclear power on Earth. Microsoft has already committed $16 billion to restart Unit 1 at Three Mile Island to fuel its data centers.
Other tech giants are following suit with massive energy deals. Google has secured an 1,800-megawatt nuclear deal, and Amazon has invested $700 million into X-energy to support its compact reactor needs. These moves suggest that nuclear energy is the preferred near-term solution for meeting AI computing power needs.
Can radiation-hardened chips meet LLM power demands?
Several critical technical barriers remain unverified. Can new radiation-hardening techniques be developed fast enough to support AI-grade chips in orbit? This remains a primary concern for the feasibility of space-based large language model processing.
Furthermore, as the report notes, the industry has yet to prove that satellite networking, formation control, and the high costs of orbital deployment can be effectively managed. The technical complexity of these systems remains a significant hurdle for companies like SpaceX and Google.
Comments 0