The array is the computer.
Compute sits on the solar array itself, spread across the area that collects the sunlight.
One-pager / Updated Nov 2026
Project Vela: capability one-pager
Solar-powered AI inference in orbit. A public, capability-level summary. Print this page for a one-page brief.
The problem
AI is short on power, not ideas. Every new mega-data center competes for grid connections, cooling water and land. Above the atmosphere, sunlight is abundant and nearly continuous.
The pitch
V3 is designed to provide the capability to move a majority of future inference off Earth, so serving it doesn't require new mega-data centers. We aim for an order of magnitude beyond other published orbital compute proposals on coverage, not horsepower. We won't match terrestrial training clusters on raw compute; we target inference, the fastest-growing segment.
The approach
Passive cooling.
The panel radiates heat to space from both faces. No pumps, no fluid loops, no separate radiators.
Mature manufacturing.
Built on established, high-volume chip processes and solar cells.
Inference-first.
Many concurrent small-to-mid-size models: the agent firehose.
How we measure
Tokens per solar watt: useful AI output per watt of sunlight, across concurrent models, counting only what we measure. See the benchmark.
Where things stand
- V1: off-the-shelf parts on a solar panel, measured on the bench.
- V2: custom chip. Design target.
- V3: the panel itself becomes the computer. Design target.
- Vela-Scout 83: proposed rideshare demonstrator, not committed.
- Project Vela is not an operating satellite service.
Want the engineering? We share architecture, test data and the full case with researchers, partners and investors under NDA. info@project-vela.com →
Project Vela. Infinite Sunshine. Effortless Intelligence.