LEO attracts billions: Satellites to space data centers race heats up

Low Earth Orbit (LEO) has evolved from a connectivity frontier into a contested layer of critical compute infrastructure, with major technology firms and startups committing billions to satellite-based data centers and in-orbit AI compute nodes. Early demonstrations in late 2025 and early 2026 have crystallized investor interest and regulatory debate.

The technical front saw a milestone when Starcloud—backed by NVIDIA partnerships—launched Starcloud-1 in November 2025 carrying an NVIDIA H100 GPU; subsequent reports indicate the platform ran and queried large language models on orbit, marking the first high-powered LLM operations in space. Industry coverage and company statements framed the event as a proof of concept for in-orbit preprocessing and model inference that could reduce downlink needs.

Proponents point to persistent solar exposure, lower atmospheric losses for power generation, and passive radiative cooling in vacuum as potential cost and sustainability advantages versus terrestrial data centers. Critics and cautious analysts counter that launch costs per kilogram, radiation hardening, maintenance, in-orbit servicing and space traffic management impose significant technical and economic hurdles; the likely near-term outcome is hybrid architectures combining ground and orbital compute.

The race has regulatory and geopolitical dimensions: startups have filed ambitious constellation plans while incumbents such as SpaceX have outlined large-scale orbital ambitions, prompting FCC filings and international discussion about spectrum, debris mitigation and data sovereignty. National and strategic reviews of LEO capacity underscore the growing economic scale of the space sector and the attendant policy trade-offs.

For investors and market participants, the short-to-medium term will be shaped by four signals: regulatory approvals for orbital compute constellations, demonstrable cost curves for launches and operations, partnerships between hyperscalers and space system integrators, and progress on orbital servicing/traffic management. Should these align, the sector could unlock substantial new markets for cloud providers and chipmakers; if not, many projects may remain experimental for several years.