Super Intelligence master guide › Energy, work, learning and safety series › Article 0010
Computing in orbit would mean placing processing equipment in space and operating it as part of a spacecraft or orbital infrastructure. The appeal is to combine access to solar energy with useful computation, or to process data close to the space instruments that produce it. The case depends on the complete system: power, cooling, communications, radiation tolerance, launch, maintenance and the value of the work performed.
Moving a computer into space does not remove its need for supporting infrastructure. It changes that infrastructure. A promising solar arrangement is only one part of the proposal, and a successful small demonstration would not by itself establish that a large orbital computing service is practical or economical.
This article explains the mechanisms and the questions a reader should ask about orbital computing. It uses hypothetical cases to compare workload choices, power budgets and transport costs. No case describes a completed commercial deployment, and no date is assigned to future large-scale infrastructure.
The “Super Intelligence” series considers both current AI and future possibilities. Artificial superintelligence remains hypothetical. Orbital computing is a physical infrastructure question that can be evaluated without assuming that current models have reached that threshold or that space is the necessary destination for all future computing.
What counts as orbital computing?
The term can cover very different activities. A spacecraft may process measurements for its own mission. A group of satellites might share some processing. A proposed orbital facility could provide computing services to customers on Earth. These activities have different reasons for being in space and should not be assessed as one category.
A mission computer supports the spacecraft’s own work. A broader computing service needs customers, input data and a way to return useful outputs. The first might be justified because the task already exists in orbit. The second must explain why moving a service away from Earth creates more benefit than cost.
This distinction prevents an important reasoning error. Evidence that electronics can perform work on a spacecraft does not establish the economics of a large data centre in orbit. The relevant scale, workload and support requirements may be different. A useful proposal identifies precisely which kind of computing it intends to provide.
Follow the task from input to output
For any proposed workload, ask where the input originates, where processing occurs and where the result is needed. A camera in space and a customer on Earth create one path. A customer uploading a large dataset from Earth to an orbital computer creates another. The same processing hardware can have a different value depending on that path.
The workload also has a timing requirement. Some results can wait; others must arrive promptly. This affects communications and operations. An orbital system should be judged against the actual task, rather than the vague idea that more available solar energy automatically makes every form of computing attractive.
Solar energy is a design advantage to examine
Solar generation is an established part of spacecraft design, but performance depends on the system and its environment. NASA’s spacecraft power guidance identifies eclipse periods, changing sunlight angles and degradation over a mission’s life as relevant considerations. Space should therefore not be described as uniformly and permanently sunlit. NASA: Spacecraft power subsystems.
An orbital computing proposal needs to state its orbit and power assumptions. How much electricity is available when the workload needs it? What happens during periods with less generation? How much power is required by components other than processors? These questions turn a general energy promise into an operating budget.
Capacity is different from delivered energy
A solar array’s rated power is a capacity figure. Energy depends on the power supplied over time. A target described as gigawatts of capacity added each year is a construction or deployment rate. It is not itself the quantity of energy generated during that year.
Consider an idealised array supplying one megawatt for five hours. That produces five megawatt-hours before any losses included in the simplified model. If the array supplies less during another interval, the energy total changes. The example shows why a large capacity target cannot substitute for a time-based power assessment.
Count the support systems
A spacecraft may need electricity for communications, controls and other equipment as well as computation. In a hypothetical orbital service, a one-megawatt power supply does not imply one megawatt available to processors. The useful processing allocation is what remains after the other demands and the planned margin are considered.
This is a general budgeting principle. A terrestrial facility also has supporting demands, but their form and constraints differ. The comparison should include them on both sides. Leaving support systems out of the orbital case while including them in the ground case would make the comparison unfair.
Why cooling remains a major question
Computing equipment generates heat, and it must operate within suitable temperature limits. Heat can move through the spacecraft’s solid structures by conduction, while heat rejection into space depends on radiation. The surrounding vacuum provides no convective cooling. Heat moving within a spacecraft and heat leaving it are therefore distinct parts of the design. NASA: Thermal control.
The phrase “cold space” can be misleading when used as a complete cooling explanation. A computer still needs a path by which generated heat leaves the equipment. An orbital design must describe that path, the conditions under which it works and the mass and geometry of the supporting hardware.
Electricity and heat budgets belong together
A power budget showing substantial electricity for processors should be connected to a thermal budget. The system must be able to handle the heat associated with its planned workload. If the thermal arrangement limits operation, available solar electricity alone does not establish the amount of computation that can be sustained.
Imagine a hypothetical module whose power arrangement supports a high processing rate, but whose thermal design supports only a lower sustained rate. The usable service is governed by the combined system. The excess power capability does not automatically become more output because another requirement constrains the workload.
The electricity argument should retain capacity, duration and reliable delivered energy; a generation rating does not settle the workload’s complete energy and service requirements.
The orientation problem illustrates trade-offs
A proposal may need to expose one surface to sunlight while keeping another arranged for heat rejection. Antennas may require a useful orientation for communication. These requirements should be examined together rather than treated as independent ideal conditions.
This is a conditional design example, not a claim that every spacecraft has the same layout problem. Its educational value is the trade-off: changing one arrangement can affect another. A convincing system description shows how several requirements are met at once, instead of describing each component under its most favourable separate assumption.
Radiation tolerance is a system requirement
Electronics in space operate in a radiation environment that must be considered in their design. NASA’s radiation-hardness assurance material describes work including environment definition, component selection and testing, layout and radiation-tolerant design. Suitability depends on the mission and lifetime rather than a simple universal label. NASA: Radiation-hardness assurance.
For orbital computing, the relevant question is what performance and reliability the complete system can sustain in the selected environment. A powerful processor on Earth is evidence about its terrestrial capability. It does not by itself establish the conditions under which that processor should be operated in space.
Include failure response in the service
A computing customer wants a usable result, not merely equipment that has been delivered. A proposal should explain how errors are detected, what work can be restarted and how service continues when a component fails. These are operational requirements that connect hardware reliability to the customer’s experience.
Suppose a hypothetical processing job can be restarted without losing much time. It may tolerate an interruption differently from a task that needs a continuous result. The workload’s recoverability changes how failures affect its value. That is why reliability should be evaluated alongside the intended use rather than only as a component statistic.
If remote operation makes intervention difficult, limited authority when recovery is difficult becomes especially relevant; monitoring cannot substitute for preventing an action that would be costly to reverse.
Communications can change which workloads make sense
An orbital service needs to move data and instructions through a communication system. NASA’s communications guidance discusses the relationship between data rate, bandwidth, signal conditions and practical spacecraft constraints. Communications are therefore part of the computing service, not a free connection assumed after the processors are chosen. NASA: Spacecraft communications.
A proposal should account for inputs, outputs, operational messages and any communication between processing units. It should also explain what happens when a planned link is unavailable. A job completed in orbit has little customer value if the result cannot reach the place that needs it within the required period.
Data location can favour onboard processing
Consider a hypothetical observation instrument producing a large stream of measurements. Some of those measurements might be repetitive or irrelevant to the question being studied. If a validated processing method can identify useful portions onboard, the system might send a smaller result to Earth.
That possibility creates a reason to compute near the source of data. It should still be tested carefully, because discarding information can remove material that later proves valuable. The proposal needs an account of what is kept, what is reduced and what evidence supports that choice. Processing location is a decision about information as well as energy.
Earth-origin workloads require another case
Now consider an organisation whose input dataset is already on Earth. Sending it into orbit creates an additional movement before processing begins. If the result is then needed on Earth, another communication step follows. The orbital service must explain why its advantages outweigh those steps.
This does not prove such a workload is impossible. It identifies the comparison required. A smaller input, a longer acceptable turnaround or a specialised task might create a different case from a large, continuously changing dataset. “Computing” is too broad a category to settle the question without the workload details.
Worked example: processing data close to an instrument
Imagine a hypothetical satellite that gathers 100 units of raw data during an observation period. Its connection can return 20 units within the useful reporting interval. The mission team is considering whether onboard processing could reduce the amount that needs to be sent.
One option sends selected raw data based on a predefined rule. Another produces a compact summary and preserves the raw record for later retrieval. A third attempts a more complex analysis onboard. Each option changes the information available to the team on Earth. The assessment should therefore begin with the mission’s question, not simply a desire to reduce traffic.
Suppose the task is identifying periods that deserve closer inspection. An onboard method that marks those periods may help the team prioritise downloads. The result needs to be tested against cases with known characteristics. If it regularly misses unusual periods, a small communication saving could come at a substantial cost to the mission’s purpose.
The team also considers uncertainty. Rather than force every observation into a confident category, the method might preserve ambiguous cases for later review. This could improve the trade-off between reduction and information loss, although the design would need its own evaluation. The example shows how computing policy can affect the usefulness of communications.
Next the team includes power and thermal limits. A more demanding analysis may reduce traffic but consume resources needed elsewhere. A simpler selection rule might offer a better overall service even if it is less sophisticated. The strongest model is not automatically the strongest choice for the whole mission.
Finally, the team evaluates the result delivered to Earth. Can the scientist or analyst answer the intended question more effectively? Were important records retained? Did the processing reduce the total time required to obtain the useful result? These are better measures than the amount of code or the raw processing rate.
Worked example: an orbital service compared with a ground service
Consider a hypothetical organisation that wants to run a large but flexible analysis once each week. It compares a ground computing service with a proposed orbital service. The task is the same for both: use the same input, apply the same method and deliver a result that passes the same checks.
The ground service quote includes computing, storage and transfer. The orbital quote includes processing but requires separate payment for input transfer and depends on a stated delivery window. Comparing only the two processing prices would be misleading. The customer needs a complete cost and timing comparison.
Suppose the orbital option has a lower processing charge in the scenario. The customer then adds the time to prepare and transmit the input, the cost of the communication arrangement and the expected effort for handling interruptions. The result may still be attractive, or the additional requirements may remove the apparent saving. The example does not assume either outcome.
Now vary the input. A compact set of parameters may be easier to move than a continuously changing large dataset. The customer can test both cases and ask whether the proposal only works under one. This reveals the workload boundary instead of allowing a favourable example to become a claim about all computing.
The comparison should also consider the service period. A system that cannot be easily upgraded may have a different useful life from an accessible ground installation. A customer should know what capability is being purchased and for how long. Lower cost in the first week does not settle the economics of the full arrangement.
A clear evaluation would state the conditions under which the orbital option wins and the conditions under which it does not. That conditional answer is more useful than a universal claim. It allows the customer to judge whether its actual workload falls within the favourable case.
Launch costs are only part of the lifecycle
An orbital system must be manufactured, prepared, transported, deployed and operated. Some components may need additional missions, replacement or end-of-life handling. A proposal should explain which of these costs are included rather than use the launch price as a substitute for the entire lifecycle.
A hypothetical low launch price improves one part of the economics, but cannot establish the complete case by itself. If preparation, supporting structures or communication dominates the cost, reducing transport may have a smaller effect than expected. Conversely, if transport dominates, a reduction could matter greatly. The cost breakdown reveals the difference.
Mass should include the complete system
A computing module includes more than processors. The accounting should include whatever power, thermal, communication, structure and other components are needed to make it operate. A cost stated per kilogram of computing hardware can mislead if it omits the supporting mass that must also be delivered.
For a worked arithmetic exercise, suppose a complete hypothetical package has five tonnes of processing-related equipment and fifteen tonnes of supporting equipment. Launching only the five-tonne figure would not deliver the described system. The relevant mass is twenty tonnes, subject to the design’s actual packaging and delivery constraints.
The numbers are illustrative rather than an estimate for orbital AI. Their purpose is to show why component-only accounting can understate a system requirement. The same habit applies to ground facilities: the whole service depends on supporting infrastructure that should be visible in the comparison.
Maintenance and replacement change the service period
If a component fails, an orbital proposal needs a route to recovery. Possible approaches could include spare capacity, replacement missions, a serviceable design or an operating arrangement that accepts some reduction. Each should be considered as part of the proposal rather than assumed available without cost.
A system may also become less attractive as computing requirements change. The evaluation should therefore distinguish physical survival from economic usefulness. Equipment that still functions may no longer offer the service customers want at a competitive cost. A lifecycle comparison includes that possibility without claiming to know the pace of future improvement.
An orbital service depends on delivery, assembly and replacement logistics; delivered hardware must be integrated into an operating service and remain supportable after the initial launch.
Separate orbital computing from space-based power for Earth
Space-based solar power for terrestrial use would collect energy in space and transmit it to Earth. Orbital computing would use power to perform work in space. These proposals overlap in some engineering questions but have different outputs, delivery systems and economic comparisons.
NASA’s 2024 study of space-based solar power identifies significant questions about launch, manufacturing, assembly, maintenance and transmission to Earth. It examines the energy-beaming proposal rather than an orbital computing business. Its findings should not be presented as a direct validation or rejection of the latter. NASA: Space-based solar power study.
The shared lesson is to examine the full architecture. A favourable solar resource does not establish how useful output reaches the customer. For one proposal that output is delivered energy; for another it is a computing result. Treating those outputs as interchangeable would obscure the case that each must make.
Worked example: what a capacity target leaves unanswered
Imagine a hypothetical plan to add two gigawatts of computing-related power capacity each year. This describes an annual addition to capacity, not an annual energy total. If every addition were completed on schedule, the installed capacity would grow, but the energy delivered would still depend on when each part became available and how it operated.
Suppose one part enters service at the start of the year and another at the end. Counting both as operating for the entire year would overstate that year’s delivery. The timeline matters even before questions about availability and workload are considered. A target should therefore separate manufacturing, launch, commissioning and sustained operation.
Now include retirement. If older equipment leaves service while new equipment arrives, gross additions and net installed capacity differ. A statement about how much is launched each year should not be read automatically as an equal increase in the service available to customers. The accounting needs to state what remains operational.
The plan must also distinguish solar capacity, total electrical capacity and power assigned to processors. They may be related, but they are not interchangeable. A solar manufacturing target does not by itself establish how much computing equipment is transported, commissioned and supplied under real conditions.
For learners, the exercise is to draw three timelines: equipment production, deployment and usable service. Add the periods during which each module can perform the chosen task. This makes an ambitious scale claim easier to understand without inventing a forecast. The question becomes which stage the target actually measures.
A network of modules is another system to evaluate
A large computing proposal might involve several orbital modules rather than one self-contained unit. If the workload is spread across modules, they may need to exchange information. The design should explain how much communication the task requires and whether the planned connections support that exchange.
Some tasks can be divided into largely independent pieces. A hypothetical set of separate calculations might allow each module to process its own portion and return a compact result. Other tasks may require frequent coordination. Their useful performance could depend heavily on the links between modules, rather than only on the processing capacity inside them.
The difference matters because a total processor count can hide coordination requirements. Adding more modules does not necessarily produce a proportional increase in useful work. The task may have a stage that cannot be divided, or the communication process may become more demanding as the system grows. These are possible limits to examine, not a claim about a particular architecture.
A worked comparison can make this concrete. Suppose ten modules each perform an independent calculation in one hour. If no coordination is needed during the calculations, the total output can be considered as ten results for that period. If a different task requires each module to wait repeatedly for information from the others, the same hardware count no longer establishes the same throughput.
A proposal should therefore describe the workload’s structure. What can be parallelised? What has to happen in sequence? What information must move between units? Answering these questions helps connect available hardware with useful computation and prevents a capacity figure from becoming an unsupported performance promise.
A lifecycle exercise for replacement and spare capacity
Consider a hypothetical service that needs four working modules to meet its promised workload. It launches a fifth module as a spare. The spare adds cost and mass, but may help maintain service when one module is unavailable. Evaluating only the four productive modules would omit part of the design that supports reliability.
The spare also needs a defined role. Does it remain idle, share ordinary work or replace a failed module after a change in operations? Each arrangement affects available capacity and the response to failure. A component called a spare is not automatically ready for every kind of interruption.
Now suppose the service lasts for several years. The evaluation should include what happens if the spare is used early and another failure occurs later. A replacement mission might be possible under the proposal, but its timing and cost would belong in the lifecycle assessment. The original launch is only the beginning of the service period.
For an adult evaluating the proposal, the question is whether the price covers the service actually promised. For a student, the exercise illustrates why redundancy and useful output are different measures. More hardware may improve resilience without being fully counted as ordinary throughput. Both contributions can have value, provided the accounting keeps them visible.
What evidence would make a proposal stronger?
A convincing proposal begins with a defined workload and a complete system description. It states the orbit, power budget, thermal arrangement, communication requirements and operating period. These details make it possible to understand what must be demonstrated before the project can provide its intended service.
The next step is evidence at the relevant scale. A component test can support a component claim. A small integrated demonstration can support a claim about that demonstration. A long-term commercial service needs evidence about sustained operation and delivered results. Each stage is valuable, but should not be silently promoted into the next.
Test the ordinary and difficult conditions
A demonstration under favourable conditions can reveal whether a concept is worth studying. An operational claim needs attention to less favourable conditions as well. What happens when a link is unavailable, a component is degraded or the workload differs from the example used to present the design?
These questions are proposed evaluation criteria, not assertions that a particular mission has failed them. Their purpose is to match confidence to evidence. A system that explains its limitations clearly may offer a more credible case than one with a large target and little information about how the target will be reached.
Compare equivalent useful work
The fairest ground-versus-orbit comparison uses the same task and quality requirement. If one service produces a less accurate or less timely output, the difference belongs in the assessment. A cheap answer that does not satisfy the user is not equivalent to a usable result.
A student can apply this principle with an ordinary classroom analogy. Compare two methods of solving the same problem, including the time needed to check the answer. The quicker method matters only if it meets the same goal. At infrastructure scale, the principle remains the same even though the components are more complex.
A fair terrestrial comparison starts by comparing equivalent useful work; processing, communications, support systems and rejected work belong inside a consistent measurement boundary.
How to learn from orbital computing without accepting a forecast
The topic brings physics, computing and economics into one problem. Students can practise distinguishing power from energy, identifying a system boundary and tracing input to output. Adults can use the same skills to evaluate an infrastructure announcement. Parents can encourage curiosity while asking which parts of an exciting claim have been demonstrated.
A useful exercise begins with a simple drawing. Put the solar array, processing module, thermal system and communication link on the page. Draw the resource and information flows. Then ask what happens if each connection is limited. This helps learners see why a system cannot be understood by looking at its most impressive component.
The next exercise is comparison. Choose one hypothetical workload and explain why its data location, timing and review requirements matter. The answer should name a condition that favours orbital processing and a condition that creates additional difficulty. Conditional reasoning is the skill being developed; there is no need to declare a universal winner.
For the transport side, continue with space transport and the logistics of a larger civilization. For the terrestrial investment chain, read AI demand and the future of energy.
Frequently asked questions about computing in orbit
Would orbital computing have unlimited solar power?
No system has unlimited usable power simply because it is in space. Available power depends on the array, orbit, sunlight conditions, conversion and the demands of the rest of the system. A proposal needs an operating budget rather than a broad statement about abundant sunlight.
The useful question is how much power is available for the intended workload over the required period. That connects the energy resource to a service a customer can actually use.
Is cooling easier because space is cold?
The environment does not remove the need for a thermal design. Heat must move away from the equipment through a suitable path, and the system must remain within its operating temperature limits. A statement about cold space does not describe that path.
For readers, the practical check is whether the proposal connects its processing load to its heat-rejection arrangement. Power and thermal capability should be assessed together.
Does computing hardware working in space prove a large orbital data centre is viable?
A demonstration supports the capability shown under its conditions. A larger service introduces additional requirements, including sustained power, communication, support equipment and operations. Its viability should be evaluated at the relevant scale.
This does not diminish small demonstrations. They can answer important questions and guide later work. The distinction prevents evidence about one stage from being mistaken for proof of an entire commercial system.
Which tasks might benefit from processing in space?
A task whose input is already produced by a space instrument may have a different case from one that must upload large amounts of data from Earth. Timing, recoverability, output size and the value of the result also matter.
These factors identify candidates for study rather than guarantee suitability. A useful proposal explains why its chosen task benefits from its location and how that benefit survives the complete system costs.
How does this connect to superintelligence?
More capable systems could create new computing demands, but capability claims do not settle where infrastructure should be built. The location question still depends on physical and economic conditions. Current AI and hypothetical future systems should remain distinct.
Return to the Super Intelligence guide to connect these infrastructure questions with adoption, delegation, learning and safety.
Previous: 0009 — Winning Through Adoption Across Industries · Next: 0011 — Space Transport and the Logistics of a Larger Civilization
