Orbital AI: Follow the Payload Through Power and Cost
MODELS & COSTS
Orbital AI: Follow the Payload Through Power and Cost
Starmind's AI1 design and the assumptions behind orbital computing costs.
A planned orbital compute payload has to become more than a launch announcement. Power delivery, heat rejection, links and service life determine whether it can do economical work. The SpaceX–NVIDIA proposal is useful to examine through those requirements and the competing cost assumptions.
In this article
The filing, payload design and factory plans are different stages of that proposal. Their scale makes it important to distinguish requested capacity, production targets and demonstrated hardware.
The proposed AI1 payload
Back on January 30, SpaceX filed an FCC application for up to one million orbital data center satellites at 500–2,000 km altitude. The filing projects that launching a million tonnes of satellites a year would generate 100 gigawatts of AI compute in orbit. Musk has said the same thing in a few venues now: "Within 2 to 3 years, the lowest cost way to generate AI compute will be in space."
In June, he revealed the first satellite's detailed design in a 30-minute video on X. The specs are... let me just put them in a table.
| AI1 satellite | Spec |
|---|---|
| Solar array | 70m panels, ~250 W/m² power density |
| Compute payload | avg 120 kW / max 150 kW — about one NVIDIA GB300 rack |
| Cooling | 110 m² deployable liquid-cooling radiator panels |
| Orbit | ~600 km LEO |
| Connectivity | 1 Tbps laser links between satellites + Starlink mesh |
| Ground latency claim | ~0.003 s |
| Compute silicon | NVIDIA Rubin GPU + Vera CPU (per the Aug 4 announcement) |
Musk notes the AI1 is structurally "much simpler than a Starlink satellite" — mostly solar cells, radiators, and a few laser links.
▲ The AI1 render Musk showed in June (source: SpaceX via X)
The cited plans describe a Gigasat factory in Bastrop targeting more than 1,000 AI satellites annually by the end of 2027, alongside the much larger Terafab proposal. The floor areas, chip-production volumes and $55–119 billion figures are project targets and cost plans. They are not evidence that those capacities are already operating.
Demonstrations and announcements
The timeline below separates orbital-GPU demonstrations from platform and constellation announcements. Announced scale is not deployed scale.
▲ Announcement timeline from an orbital H100 demonstration to a proposed large Rubin constellation.
The cited timeline covers Starcloud’s H100 flight the previous November and subsequent LLM work, the SpaceX–xAI combination in February, NVIDIA’s Space Computing announcement in March, and the AI1 reveal and SpaceX listing in June. A single orbital accelerator demonstration and a million-satellite proposal are different stages.
Power and thermal trade-offs
The proposed advantage is access to solar power without a terrestrial grid connection. The engineering trade replaces some ground constraints with launch mass, radiation exposure, orbital operations and heat-rejection requirements.
Solar irradiance outside the atmosphere is about 1,361 W/m², but that is not electrical output from a panel. Panel efficiency, orientation, eclipse time and degradation affect useful energy. Likewise, a cold radiative environment does not make cooling free: heat must reach radiators, and their area, temperature, mass and view to space matter. Orbital systems also require authorization and spectrum coordination. Simple 'free power, free cooling, no permits' descriptions omit these costs. The NASA thermal-control overview explains the radiator energy balance; NASA's solar-irradiance summary identifies the irradiance value.
▲ Left: NVIDIA's Space-1 Vera Rubin module claims 25× the on-orbit inference performance of an H100. Right: one AI1 satellite draws about as much power as a GB300 rack.
NVIDIA announced its Space Computing platform at GTC in March 2026 and describes Space-1 Vera Rubin as a module for orbital datacenters. The AI1 announcement also specifies Rubin and Vera. Planned platform adoption should be distinguished from sustained operation at scale.
Competing economic assumptions
The published economic cases use different system boundaries and launch assumptions. Their totals should not be combined as though they were bids for the same service.
Starcloud projects energy costs near $0.005/kWh and cites a 10-year, 40MW comparison of roughly $8.2 million versus $167 million on the ground. Lonestar claims 97% lower operating costs. These are company projections. Their inclusion of launch, replacement, communications and radiators must be checked before using the totals as an all-in comparison.
The skeptic's calculator: Andrew McCalip (Varda Space Industries) built a public calculator and the base case says orbital compute costs about 3× more per watt than ground. His summary: "The physics doesn't immediately kill it, but the economics are savage." The European Space Policy Institute calls the $10M-per-flight Starship assumption "unrealistic in the near-term."
▲ Left: launch cost is the whole game — $2,700/kg on Falcon 9 today, versus the sub-$100/kg that Starship's economics would need. Right: the "free solar" claim that anchors the optimist case.
Launch cost per kilogram strongly affects the calculation, but it is not the only variable. Hardware lifetime, shielding, radiator mass, utilization and replacement launches can change the result. The cited filings discuss deployments as early as 2028; the material here does not establish a validated break-even date at either $100/kg or $500/kg.
Other orbital-computing projects
The original reports also describe an October Starcloud-2 plan with Blackwell and AWS Outposts, Kepler optical relay satellites with compute nodes, and Axiom ISS-derived nodes. Google’s Project Suncatcher concerns radiation work on Trillium TPUs and a laboratory 1.6Tbps optical link. Aetherflux’s power-beaming demonstration and Lonestar’s lunar concept are additional plans. Each project’s current status requires separate confirmation.
The proposal leaves two practical questions for later deployments.
First, can the announced NVIDIA payload sustain useful workloads through radiation exposure and thermal cycles? A shared architecture across projects may help software reuse, but it does not establish an orbital industry standard or exclude competing accelerators.
Second, what is the delivered cost over a measured service life? Launch demonstrations, regulatory progress and a complete operating record would help answer that. A successful small mission alone cannot validate a million-satellite or 100GW production target.
The current evidence supports a planned compute payload and competing cost models. It does not yet support moving a general cloud budget to orbit on the assumption that power and cooling are free.
Sources and projection limits
Sources: SpaceX earnings call + X post (2026-08-04) · SpaceX FCC filing (2026-01-30) · NVIDIA Space Computing press release (2026-03-16) · SpaceX AI1 reveal (2026-06-08) · SpaceX IPO prospectus / record $75B IPO (2026-06-12) · Starcloud / Kepler / Axiom / Google Suncatcher / Aetherflux coverage via CNBC, WSJ, SpaceNews, Aviation Week, ETNews, TradingKey. Cost projections are exactly that — projections. The orbital-DC economics debate (McCalip calculator, ESPI analysis) is live and unresolved; treat vendor numbers as directional.
Related on this blog: Nvidia's reported $12.9B move on open weights · the $7,000 humanoid and the open physical-AI push · Google’s AI Satellite Is in Orbit. Can Space Beat Earth?
Comments
Post a Comment