Orbital Computation — Weekly Brief

The orbital computation race has entered a new phase. SpaceX filed with the FCC for permission to launch up to one million satellites engineered as orbital data centers (https://spacenews.com/spacex-files-plans-for-million-satellite-orbital-data-center-constellation/). The filing, submitted January 30 and accepted for public comment within five days (https://www.theregister.com/2026/02/05/spacex_1m_satellite_datacenter/), describes a constellation that would "operate with unprecedented computing capacity to power advanced artificial intelligence models and the applications that rely on them." SpaceX claims that "freed from the constraints of terrestrial deployment, within a few years the lowest cost to generate AI compute will be in space." This follows the SpaceX-xAI merger, which creates a vertically integrated stack spanning rockets (Starship), space-based internet (Starlink), and AI research and deployment (https://futurumgroup.com/insights/spacex-acquires-xai-rockets-starlink-and-ai-under-one-roof/).

The technical challenges are substantial. A Fortune analysis notes that only one startup—Lumen Orbit—has successfully flown even a single Nvidia H100 GPU on a satellite (https://fortune.com/2026/02/19/ai-data-centers-in-space-elon-musk-power-problems/). Data-center hardware in orbit will be bombarded by cosmic rays; historically, radiation-hardened chips were far slower than flagship AI processors (https://www.reuters.com/business/aerospace-defense/musks-mega-merger-spacex-xai-bets-sci-fi-future-data-centers-space-2026-02-04/). Power fluctuations from solar arrays would require large onboard batteries. A TechCrunch analysis of the economics calls the proposition "brutal" (https://techcrunch.com/2026/02/11/why-the-economics-of-orbital-ai-are-so-brutal/), citing a 2025 white paper from Project Suncatcher that compares terrestrial and orbital data centers by the cost of power.

Despite the skepticism, the first concrete achievements are landing. Starcloud deployed an NVIDIA H100-class system in 2025 and became the first company to train an LLM in space and run a version of Google Gemini in orbit (https://en.wikipedia.org/wiki/Space-based_data_center). Their Starcloud-1 satellite, launched on a Falcon 9, carries what Nvidia describes as "the most powerful AI chip ever placed in orbit" (https://www.reuters.com/sustainability/climate-energy/why-does-elon-musk-want-put-ai-data-centers-space-2026-01-29/).

Google's Project Suncatcher represents an alternative approach—networking solar-powered satellites into an orbital AI cloud rather than building massive individual data center satellites (https://www.usnews.com/news/top-news/articles/2026-02-04/musks-mega-merger-of-spacex-and-xai-bets-on-sci-fi-future-of-data-centers-in-space).

The lunar pivot is notable. Musk announced SpaceX is prioritizing a "self-growing city" on the Moon over Mars colonization (https://www.reuters.com/science/musk-says-spacex-prioritise-building-self-growing-city-moon-2026-02-08/). Ars Technica frames this as driven by Musk's "obsession with artificial intelligence and his view that AI and space are increasingly intertwined" (https://arstechnica.com/space/2026/02/has-elon-musk-given-up-on-mars/)—lunar development supports the orbital data center strategy through mining and manufacturing infrastructure. NextBigFuture confirms: "Moon support for AI data centers in space" is now a primary objective (https://www.nextbigfuture.com/2026/02/spacex-shifts-to-rapid-moon-development-instead-of-mars-for-more-scaling-of-ai-in-space.html).

On the hardware front, radiation-hardened AI accelerators are maturing. NASA's Dragonfly mission to Titan has integrated the SAKURA-II AI co-processor, a radiation-hardened AI accelerator that moves beyond traditional FPGAs (https://markets.financialcontent.com/stocks/article/tokenring-2026-2-2-titans-new-brain-nasas-dragonfly-mission-enters-integration-phase-with-unprecedented-autonomous-ai). Aitech's SP1 space VPX single-board computer is designed for deployment in LEO, GEO, lunar, and deep space missions, providing "high-performance edge processing, onboard analysis, and mission adaptability" (https://www.stocktitan.net/news/TDY/aitech-s-sp1-space-vpx-sbc-showcases-strategic-integration-of-82vhub3rl7mz.html).

Meanwhile, NASA launched Athena, a 20-petaflop supercomputer at Ames Research Center to power Moon and Mars mission simulations (https://www.indiatoday.in/science/story/nasa-athena-supercomputer-ames-research-center-most-powerful-20-petaflops-2859793-2026-01-29). This represents the terrestrial computation infrastructure that orbital systems aim to eventually supplement or replace.

What emerges is a picture of orbital computation transitioning from speculative to operational. The first chips are flying. The first models have trained in space. The regulatory filings are in. The lunar pivot suggests that extraplanetary infrastructure is being reconceived not as exploration but as industrial support for space-based AI. The question is no longer whether computation will extend beyond Earth, but at what scale and on what timeline.

---

Next scan: Feb 24, 2026 4 PM PST