Inference within mission limits.
Software intended to run approved AI workloads on compatible spacecraft accelerators. The host operator retains command authority. Reliability and hardware portability are targets for validation.
Satellite InferenceAI Inference in OrbitGet in touch PRODUCTS / DEVELOPMENT ROADMAP
Build the pathfinder.
Learn from the mission.
Our program connects spacecraft inference software with a compact hosted payload, then uses flight evidence to guide dedicated solar compute nodes.
01 / THE PRODUCT FAMILY
Software and hardware.
One staged program.
A hosted compute payload.
Jetson T5000, from the NVIDIA Thor family, is the reference platform. Jetson AGX Orin Industrial is the fallback; a Lite configuration remains an option if it better fits the host and workload.
Dedicated solar compute.
Node-1, a planned 10–20 kW compute spacecraft on a purchased high-power bus, comes first; the nodes of later clusters are designed at 100 kW. Both use a dawn–dusk orbit and require their own development and financing.
Capacity built from nodes.
A longer-term program to aggregate orbital compute. Architecture and deployment depend on demonstrated economics, available launches and contracted demand.
These are development plans. Hardware selection and qualification follow the workload, host interface and measured power, thermal and radiation performance. No manufacturer or host is represented as a partner.
02 / TWO EARLY ORBITAL STEPS
Orbit-0.
Software on an existing host.
The first objective is an approved software experiment using a GPU already in orbit. It targets earlier orbital learning while the dedicated pathfinder is developed.
Access, workload compatibility and operator approval must be agreed with a provider. Orbit-0 is a program target, not a reserved service.
SI-P1.
Our hosted pathfinder.
The hardware program targets a flight-ready unit with environmental test evidence and a defined host interface. The host provides the spacecraft services required to operate it.
The SI-P1 launch target is H1 2028, dependent on the host manifest, qualification and flight financing.
03 / THE FIRST PROGRAM
A 13-month plan.
Defined gates.
The plan covers software development, Orbit-0, SI-P1 build and tests, and a host reservation. The SI-P1 launch also needs a host slot and separate flight funding.
| Target window | Work | Required gate |
|---|---|---|
| Before the start | Satellite Inference Inc. formed; IP assignments and planned patent filings. | Formation and ownership complete. |
| Months 1–4 | Requirements, host interface, team onboarding and supplier quotations. | Agreed workload and a feasible host configuration. |
| Months 3–8 | Design-partner discussions and a first paid ground evaluation. | Customer-defined acceptance criteria and commercial scope. |
| Months 4–6 | Orbit-0 software experiment on an existing orbital GPU. | Provider access, compatibility and approval. |
| Months 6–8 | Host contract and deposit; radiation and thermal-vacuum tests of the engineering model. | Host agreement and satisfactory test results. |
| Month 8 | Confirm the flight: host slot and flight funding. | Host agreement and test results in hand. |
| Months 9–13 | Flight-unit qualification and delivery to the host. | Qualification accepted. |
Windows count from the program start. Flight funding is not yet committed. Host, lab and launch pricing remain subject to quotations.
04 / LONGER-TERM CAPACITY TARGETS
Follow the evidence.
Expand the useful capacity.
- 2029–30
Node-1
A dedicated solar compute demonstrator, following pathfinder evidence and a separately financed development program.
- 2031–32
First MW-class cluster
Aggregate useful capacity only when launch access, infrastructure costs and customer needs support it.
- BY 2035
100 MW-class ambition
A conditional scale objective requiring substantial new capital, fleet deployment and customer commitments.
- ECONOMIC GATE
Three costs must improve together
Launch cost to the operating orbit, spacecraft cost per watt and mass per kilowatt define the scale economics.
Future nodes are modeled in a dawn–dusk sun-synchronous orbit with approximately 96% annual sunlight. Seasonal eclipses remain, and the chosen orbit requires mission-specific validation. The first pathfinder follows its selected host.
The program is launch-agnostic. A 20 kW node is planned at under a tonne, within rideshare and medium-lift capacity. Cluster economics improve sharply with heavy-lift prices, and only the largest clusters need heavy-lift vehicles. No launch provider is represented as a partner.