Vision
A space economy operated by intelligence
Within a decade, space is critical infrastructure: constellations delivering communications, navigation, defense, observation and compute, flown by intelligence under human command.
Why space, and why not yet
We have already started handing real work to intelligence, but only in software. Code, text, analysis, decisions that live in bits. There, a mistake is a rollback and the cost of being wrong is a retry.
A spacecraft is not software. It is hardware, in orbit, wearing out. You cannot revert it and you cannot reach it. Handing real work to intelligence there is not a productivity gain, it is a transfer of responsibility for a physical asset whose failures are permanent.
Nobody transfers that responsibility on trust alone. They transfer it when they can prove, at any moment, what condition that specific machine is actually in. That is the real blocker: not the models, not the physics, but the fact that an operator cannot produce a current, defensible statement of the condition of any one of their spacecraft.
So every judgment stays human. Whether this behavior matters, whether that satellite still has years in it, whether it can still be disposed of: each one is worked out by a person, from scratch, one satellite at a time.
Which is why a fleet's size is still a function of its rota. More spacecraft means more engineers, and headcount is the least scalable input in the business. Break that coupling and the unit economics of operating in space change, because intelligence compounds and headcount does not.
Machines running the routine. Humans supervising, and taking the decisions that matter.
Why now
That coupling used to be affordable. Three forces are making it expensive at the same time, and each one is asking an operator a question about the condition of an individual machine.
Scale
Fleets are outgrowing the teams that fly them.
There are around 14,000 operational satellites today and ESA expects far more by 2030. Every launch adds work per satellite: more data, more rules, more knowledge held in a few people's heads. Which satellite needs attention right now, and can you say so without an engineer working it out?
Accountability
Operators are increasingly held to account.
Disposal is now enforced at the licensing stage rather than encouraged. In Europe, whole-lifecycle risk management for space infrastructure is moving through law and would reach anyone serving the European market. What is the condition of this specific asset, and will the number survive being shown to someone else?
Exposure
And the risk sits with the operator.
Very few satellites in growing fleets carry in-orbit cover, because underwriters price on engineering judgment rather than loss data. The exposure did not disappear. It moved onto the operator's own balance sheet. How much life is actually left, and what is it worth?
Three questions, one requirement. And the input already exists: the evidence needed to answer all three comes down in telemetry every pass, and data volume and compute now make it tractable per component, per satellite, continuously.
What is missing
Nobody keeps it.
Assurance is proved once, before launch, against expectations. Then the fleet flies for years, and the most expensive decisions lean on a case that was closed the day the rocket left the pad. An operator can hold six years of telemetry on a satellite and still need a senior engineer to spend an afternoon reconstructing context before anyone can say whether an unusual reading matters, or how much life is left in the component that produced it.
Not for lack of tools. Mission control systems, simulation, RAMS models and homegrown analytics all work. Each was built for one phase of the machine's life: design, test, operations, reliability, disposal. Each holds a partial view of the same spacecraft, and none of it stays current.
So the condition of a machine is never held anywhere. It is reconstructed, by people, whenever someone needs it. That is what keeps fleet size tied to headcount, and it is why nothing has been delegated: an operator will not hand work to a system that cannot show its reasoning, and no regulator, insurer or board will accept a number that was assembled for the occasion.
The gap
What is missing is not another view of the spacecraft. It is something that holds its condition and keeps it current.
Our mission
Build the intelligence layer for the space economy.
The second brain of a satellite fleet and of the organization that flies it. It reads the data coming down, organizes the company's knowledge, and reasons over both with physics-based and analytical engines: causes, consequences, options. It hands people a qualified result with a recommended action.
Delegation is graduated, never binary.
| Level | The layer | The operator |
|---|---|---|
| Inform | Surfaces qualified events and context | Interprets |
| Recommend | Proposes actions with reasoning | Decides |
| Act with approval | Prepares execution | Approves |
| Act within bounds | Executes routine operations and reports | Supervises, sets the bounds |
AI proposes. Physics grounds. Your engineers command. And it runs inside your perimeter: intelligence this critical has to be sovereign.
Why this compounds
We started where trust is won: live mission assurance. Mercury keeps the operational case live. Hydra keeps the reliability case live.
Operations is our data engine, not our product line. Every qualified event and every reliability estimate becomes structured evidence, and that evidence travels in two directions: backward, into how the next spacecraft is tested and designed, and forward, into how much of the routine a fleet is willing to delegate. Each rung earned makes the next one arguable.
What compounds across the industry is the architecture and the learning methods, not anyone's telemetry. Customer data never leaves their control. That is what lets a sovereign product scale.
Where this goes
Every spacecraft worth flying carries a live case as a matter of course, the way it now carries a current orbit. Operators run larger fleets with the teams they already have, keep assets in service years longer on evidence rather than assumption, and answer for both without rebuilding the number for each audience. Above them sits the layer that holds the record and, one rung at a time, does the work.
Evidence only accrues in real time
Every other input to this can be bought. Compute, models, engineers, capital. Evidence cannot. It accumulates one pass at a time, and a fleet that starts in 2029 starts from zero while one that started in 2026 has three years of flight history behind every decision.
That is a slow asset, and we are building it deliberately: a small number of operators, deep deployments, evidence accruing from the first pass.