Orbital Data Center: When AI Compute Moves Off-Planet

The thesis: why put a data center in orbit

The newest frontier of the space economy is moving compute itself off-planet. The argument is physical, not romantic. Training and serving large AI models is bounded on Earth by two scarce resources — electrical power and cooling water — and demand is outrunning the grid. Orbit sidesteps both.

• Near-continuous solar power. A dawn-dusk sun-synchronous orbit sees near-continuous sunlight — no night, no clouds, no grid hookup. Google's Project Suncatcher research notes that in such an orbit a solar panel can be up to 8x more productive than on Earth and produce power nearly continuously, reducing the need for batteries (Google Research).
• Free radiative cooling. Waste heat radiates straight to the roughly 3-kelvin cold of deep space — no water, no chillers, no town next door competing for the supply.
• The wall, and the catch. Orbit removes the power and water constraints at the cost of launch, radiation and on-orbit servicing. Those are the real engineering problems, and every serious program is essentially a bet that launch keeps getting cheaper and chips keep getting tougher.

You can see the orbital-compute thesis laid out interactively in the live 3D map under the Data Center pillar.

Who is actually building it

The honest picture is early-stage: a handful of real flight programs plus one large feasibility study. Here is who is doing the work, with primary sources.

Note: this site's live panel labels ASCEND "ESA — ASCEND." The verified funder is the European Commission via Horizon Europe, with the study led by Thales Alenia Space; ESA's role was indirect.

The connective tissue: lasers, not ground stations

An orbital data center is useless if its only path home is a bandwidth-constrained dish on the ground. The emerging answer is optical inter-satellite links that mesh compute nodes into existing constellations.

• Starcloud + Starlink. Starcloud has contracted with SpaceX for 50+ Starlink "Mini Laser" terminals across 25+ satellites. Each terminal supports up to 25 Gbps at distances up to 4,000 km, linking Starcloud's compute into the Starlink LEO mesh and bypassing ground stations. Starcloud-2, slated to fly on a SpaceX Falcon 9 in January 2027, will be the first to carry the laser hardware (BusinessWire; SpaceNews).
• Suncatcher's optics. Google's design uses free-space optical links; a bench-scale demonstrator achieved 800 Gbps each way (1.6 Tbps total). The research models an 81-satellite constellation at 650 km altitude within a roughly 1-km cluster radius (Google Research).

This laser backhaul is also why the broader orbital infrastructure layer — launch cadence and the Starlink mesh — matters so much to whether orbital compute ever scales.

Does the hardware survive? The radiation question

The skeptic's first objection is radiation: cosmic rays and trapped protons flip bits and degrade silicon. The most concrete public data point so far is encouraging. In Suncatcher testing, Google reported that its Trillium TPU's High Bandwidth Memory tolerated a cumulative dose of 2 krad(Si) before irregularities — nearly three times the expected shielded five-year mission dose of 750 rad(Si) — with no permanent failures at doses up to 15 krad(Si) (Google Research).

That is one chip family in a lab, not a proven multi-year on-orbit record — but it is the kind of result that turns "impossible" into "hard engineering problem." Serviceability and thermal management remain open: companies like Redwire frame orbital data centers as integrated power-and-thermal systems, drawing on heritage hardware such as IROSA solar wings delivering 20+ kW each on the ISS (Redwire whitepaper).

Public-market exposure: adjacency, not pure-play

Adjacency, not endorsement — not financial advice. Projects and companies named are independent and not affiliated with this site. Prices and figures are for information only.

There is no clean way to "buy orbital compute" on a public exchange. The pure-play builders are private (Starcloud) or units of giants (Google). The closest listed adjacencies are infrastructure and launch suppliers, not orbital-compute pure-plays:

• PL — Planet Labs (NYSE): Google's build-and-operate partner for the Suncatcher TPU prototypes.
• RKLB — Rocket Lab (Nasdaq): launch plus the satellite buses these compute platforms ride on.
• RDW — Redwire (NYSE): the in-space power, structures and thermal systems they depend on.
• SPCX — SpaceX: launch and the Starlink laser backhaul behind the Starcloud integration. This site's UNOFFICIAL premise treats SpaceX as listed under ticker SPCX (treated as listed 2026-06-12). SpaceX is privately held; there is no confirmed public listing under SPCX. This is fan framing, not fact.

As of mid-2026, these adjacency names had reportedly rallied roughly 97%–190% year-to-date amid SpaceX-IPO anticipation — which underscores that they are sentiment-driven adjacencies, not orbital-compute pure-plays (Stocktwits/TradingView). Information only, not investment advice.

Live quotes for these names sit in The Space Economy pillar and in the app's Economy panel.

What you can see in the live map, and where to go next

The numbers above come alive in the live 3D orbital map. Open the Data Center pillar to read the why-orbit thesis, the builders table, and the exposure cards; the cards deep-link into the Economy panel for live adjacency quotes. The map propagates thousands of real orbits in your browser, so you can also watch the Starlink mesh that orbital compute would lean on for backhaul.

Continue across the four-pillar tour of the space economy:

• The Space Economy — listings, adjacencies and market context, with the SPCX unofficial premise.
• Orbital Infrastructure — the launch cadence, constellations and mesh that make orbital compute possible.
• Space Colonization — the longer arc of living and working beyond Earth.

Published by Asakasa Technologies SRL — Applied Intelligence Lab, Bucharest, Romania. CUI 54809478 · Reg. Com. J2026036098000 · EUID ROONRC.J202603609800. This is an INDEPENDENT FAN / reference project, not affiliated with, endorsed by, or sponsored by SpaceX, Nvidia, Google, Planet, Rocket Lab, Redwire, ESA, Thales Alenia Space, or the European Commission. UNOFFICIAL · NOT FINANCIAL ADVICE.