The Biggest Construction Project Humanity Has Never Attempted

Part 1 of 30 · Series date: September 8, 2026

An industry above our heads could one day turn sunlight into electricity, useful computation, and services for Earth. Getting there will require far more than a bigger rocket.

Imagine an ordinary morning

Picture a possible morning a few decades from now.

A rancher in New Mexico checks a map on her phone. A warning shows an unusual patch of heat in the mountains beyond her grazing land. A satellite has spotted it, compared it with earlier images, and sent an alert. Most of the raw pictures never traveled to Earth. A computer above the atmosphere examined them first.

Across town, a clinic receives the results of an overnight research task. Some calculations ran in a conventional data center. Others ran on machines in orbit, where the job could wait for an available communications link. The clinic does not need to know which machine did which part. It needs a useful, verified result.

Far away, a robotic vehicle near the Moon's south pole pauses beside a dark crater. It checks its remaining energy and the next available communications window before moving on. Its work depends on an infrastructure network: power, navigation, computing, and a way to call for help.

This is an imagined future, not a description of a system operating today. Each scene depends on technologies with different levels of maturity. Some pieces have already been demonstrated. Others remain engineering proposals. A few may never make economic sense.

But the scenes share a powerful idea. Space could become more than a place where we put instruments and send explorers. It could become a place where we build useful industrial capacity.

The first question is not whether that future looks exciting in an illustration. It does.

The question is whether we can build something above Earth that does a worthwhile job better, or at a competitive cost, than the alternatives below.

The cloud has a power cord

We have given the internet a wonderfully misleading name: the cloud.

It sounds light and almost weightless. Yet the services behind it depend on buildings filled with equipment. Computers need electricity. Networks need cables and switches. Equipment needs cooling, maintenance, security, and replacement parts.

Artificial intelligence adds another demanding customer to that system. Training an AI model means running a large amount of computation to adjust the model's internal settings. Using the trained model also consumes energy. Every generated answer, image, or video requires physical machines to perform work.

The International Energy Agency's April 2026 report projects worldwide data-center electricity consumption of about 950 terawatt-hours in 2030 in its central case, compared with an estimated 485 terawatt-hours in 2025. The projection covers all data centers, including AI workloads. It is a scenario-dependent forecast, not a measured future total. IEA: Key Questions on Energy and AI, executive summary, page 10.

The precise future total matters. But the local question can matter even more: can a particular place deliver a large amount of reliable power to a particular site?

A company cannot plug a giant computing campus into an ordinary wall socket. It needs a suitable power supply and the infrastructure that connects it. Even when electricity is available across a region, delivering more of it at one location can require substantial construction.

That helps explain the appeal of looking upward. If demand for useful computation keeps growing, could some machines operate closer to a vast energy source that does not require a fuel train or gas pipeline?

The Sun is already there. The difficult part is everything we must build around it.

Two ideas, not one

Before going farther, we need to separate two concepts that are often mixed together.

The first is a solar-powered space data center. Solar panels generate electricity. Computers aboard the spacecraft use it. Communications equipment sends information to customers or other satellites.

The second is space-based solar power delivered elsewhere. A spacecraft collects sunlight, converts it into electricity, and uses a transmitter to send energy to a receiver. That receiver might be on Earth, on the Moon, or aboard another spacecraft.

One system exports information. The other exports energy.

That distinction changes the design. A computing satellite can use electricity beside its own solar panels. A power-exporting satellite must also justify the size, efficiency, and cost of its transmitter and receiving system. Delivering useful power to Earth involves more than keeping an onboard computer running.

Think of two farms. One uses its solar panels to run a greenhouse on the property. The other builds a business around selling electricity to distant customers. They share some equipment, but they do not share every cost or requirement.

The same applies in orbit. Success in one field could help the other through better panels, lightweight structures, or cheaper launches. It would not prove that both businesses work.

This series follows both paths—and asks where they might eventually meet.

What we have already learned

Solar-powered spacecraft are not a new invention. Solar arrays and batteries are standard elements of many satellite power systems. NASA's spacecraft power reference describes the technologies engineers already use to generate, store, and manage electricity. Scaling them up remains a separate challenge. Source: NASA, Small Spacecraft Power.

Computing near the source of space data has also been tested. During the private Axiom Mission 1 in 2022, an AWS Snowcone device aboard the International Space Station demonstrated edge-computing work, including screening images for sensitive content. That was a specific demonstration—not an independent orbital hyperscale cloud. Source: Axiom Space, AWS Snowcone research.

Power transmission has its own early milestones. Caltech's Space Solar Power Demonstrator tested lightweight structures, solar cells, and wireless power-transfer technology. Its MAPLE experiment demonstrated wireless power transfer in space and a transmission detectable on Earth. These were experimental results, not delivery of utility-scale electricity. Source: Caltech, first mission results.

These examples establish something important: the field contains real hardware and real tests.

They also show how carefully we need to read progress reports. A computer operating on a crewed station is not the same as a free-flying data center. A detectable energy signal is not the same as a power plant serving a town.

Early achievements deserve attention for what they accomplish. Inflating them makes the next engineering step harder to understand.

Infrastructure is the real invention

A railroad is more than a locomotive. It needs track, stations, repair facilities, signaling, workers, and customers. A power grid is more than a generator. It needs transmission, controls, protection, and maintenance.

An orbital industry would face the same basic truth.

It would need regular transportation, dependable communications, reliable power, repair or replacement plans, and customers willing to pay for a service. It would also need rules that allow different operators to share a limited and increasingly valuable environment.

That is why the biggest breakthrough might not look dramatic. A standard connector that lets robots replace equipment could matter enormously. So could a reliable way to test a repaired module, move a workload between spacecraft, or retire a failed satellite safely.

These are the quiet details that turn a collection of impressive machines into infrastructure.

One spacecraft can be a demonstration. A thousand spacecraft that require constant special treatment can become a very expensive demonstration.

A functioning industry needs repeatability.

The first customer may already be in space

The most persuasive early use for orbital computing may not be answering questions from a laptop on Earth. It may be helping a satellite handle its own information.

Consider an Earth-observation instrument. It collects images, but not every picture is equally useful. Clouds may hide the ground. A camera may capture a scene that has barely changed. Some images may contain an urgent event while others can wait.

In a possible onboard workflow, a computer checks the images, marks useful regions, and prioritizes what to send. The satellite still needs clear rules about accuracy, uncertainty, and which original data to preserve. But it no longer treats every byte as equally urgent.

The principle is familiar. A person watching a security camera does not normally call the police to describe every empty minute. The useful service is noticing a meaningful event and communicating it promptly.

This is a proposed application pathway, not a claim that every satellite already works this way. Its economic appeal is straightforward: some information is more valuable when processed before it travels.

That could make space-native customers a more natural starting point than ordinary cloud customers. A spacecraft may benefit from nearby processing even when an Earth-based user would prefer a conventional server.

Why the Sun does not settle the argument

Sunlight has no fuel bill. Collecting and using it still has a bill.

Solar arrays must be manufactured, transported, deployed, controlled, and eventually replaced. Their output depends on their design, orientation, temperature, and operating conditions. Depending on the orbit, a spacecraft may pass into Earth's shadow and need stored energy or a different work schedule. Source: NASA, Small Spacecraft Power.

For a business, the relevant question is not simply the price of sunlight. It is the cost of useful work delivered over the system's life.

Suppose two computers can complete the same job. One has cheaper energy but costs far more to install and replace. The other has a regular electricity bill but sits in a building where a technician can change a failed part. Either might win, depending on the numbers.

An orbital computer must earn back more than its launch cost. Its revenue must support the complete system that keeps it productive.

This is why a promise of dramatically cheaper electricity should lead to another question: cheaper per unit of energy collected, or cheaper per useful result delivered to the customer?

Those are not always the same thing.

The coldest place can still overheat your computer

Space sounds like an ideal place to cool equipment. The reality is less convenient.

A fan works by moving air. Outside a spacecraft, there is essentially no surrounding air to carry heat away. Heat can move through the spacecraft's materials and internal cooling system, but it must ultimately leave through radiation.

Radiators provide surfaces designed to release that energy. Their performance depends on temperature, materials, orientation, and what they face. A spacecraft also receives heat from sunlight and its surroundings. Engineers must manage the whole balance. Source: NASA, Small Spacecraft Thermal Control.

For a data center, this matters because computation produces waste heat. Adding more processors means finding a way to reject more heat, not merely generating more electricity.

Large radiator surfaces can add mass and deployment complexity. Running equipment hotter may help heat rejection but create other design constraints. Scheduling less work during difficult thermal conditions may protect hardware while reducing revenue.

There is no need to invoke mysterious physics. Ordinary engineering is demanding enough.

An orbital computing platform is therefore a power plant, communications node, and thermal system wrapped around computers. The chips may receive the attention, but they do not operate alone.

A useful future does not require replacing Earth

Many discussions present a choice that is too simple: keep data centers on Earth or move them to space.

A more useful question is where each job belongs.

Some computing tasks involve repeated, rapid exchanges among many processors. Others can be divided into largely independent pieces. Some customers need immediate responses. Others can wait hours for a batch of results. Some datasets begin on Earth. Others begin aboard instruments in orbit.

These differences suggest a possible mixed system. Ground facilities could handle work that benefits from dense networks and easy maintenance. Orbital machines could process space-generated data or other jobs that earn enough value from their location.

That is a design argument, not a forecast of inevitable adoption.

The same restraint applies to space solar. Earth-based energy technologies will continue improving while orbital proposals develop. A space power station must compete with the alternatives available when it enters service, not just with the equipment available when someone first sketches it.

In 2024, NASA assessed two conceptual space-solar systems aimed at operation in 2050. Under its assumptions, they were more expensive than terrestrial sustainable alternatives. It identified important gaps in assembly, autonomy, power transmission, manufacturing, and launch economics. Source: NASA, Space-Based Solar Power assessment summary.

That finding is neither a permanent ban nor a promise. It is a reason to demand better evidence as designs evolve.

What people on Earth might gain

If orbital computing finds the right applications, its benefits could reach people who never visit a launch site.

Possible services include faster interpretation of environmental observations, better support for remote scientific instruments, and more capable spacecraft. Some may improve disaster awareness or help operators prioritize inspections. None of these outcomes follows automatically from putting a processor in orbit. The complete service must work, and someone must act on its results.

If space-based power becomes practical, it could create options for users who are difficult to serve through conventional infrastructure. A remote installation or lunar operation might value electricity differently from a city beside an established grid.

The largest ambitions go further: additional energy supply, orbital factories, and a supporting network for sustained activity beyond Earth. Those possibilities belong in the conversation, but as scenarios to investigate rather than benefits already secured.

The public should also ask who receives the gains. A system can be technically impressive while providing little local benefit. Better services, affordable prices, shared scientific knowledge, and responsible operations require decisions about ownership and access.

Innovation creates possibilities. It does not decide fairness by itself.

Moving industry upward does not erase responsibility

There is a temptation to imagine orbit as empty space where Earthly concerns no longer apply.

That is a poor foundation for an industry.

Any serious proposal needs to explain how it will avoid collisions, handle failures, retire hardware, and protect other users. It should account for manufacturing and transportation impacts, not just the absence of a smokestack beside the computer.

Power transmission raises additional questions about receivers, beam controls, interference, and public trust. Computing raises questions about sensitive data, security, and service continuity.

These concerns do not prove that the technologies should be rejected. They define work that must be included in the design and budget.

A cheap system that leaves its cleanup bill to everyone else is not truly cheap.

The useful goal is not to place industry beyond responsibility. It is to extend responsible infrastructure into a new environment.

The missing middle between invention and everyday life

Return to the imagined warning on the rancher's phone. Between a hot patch in a satellite image and a useful alert lies an entire chain of decisions.

Someone must establish that the instrument is working. Software must distinguish an unusual signal from familiar sources of heat. A network must carry the result. A service must translate coordinates into a location the rancher recognizes. The warning must reach someone who can check it and, if necessary, contact the appropriate responders.

An orbital computer improves that chain only if it improves one or more of those steps without weakening the others. Sending an uncertain result sooner can help. Sending it with unjustified certainty can create a new problem.

This is the missing middle in many technology stories. We hear about an invention and then leap to its promised benefit. Infrastructure occupies the space between them. It includes unglamorous work such as maintaining contact lists, testing software updates, and deciding what happens when a customer cannot connect.

The most important invention might therefore be a dependable service assembled from several technologies, rather than any one spectacular machine. A good alert system does not ask its customer to become an orbital engineer before it can be useful.

A future large enough for ordinary people

Much of the public conversation focuses on billionaires, rockets, and giant investments. Those subjects deserve scrutiny, but they can make the possible economy seem remote from ordinary work.

Consider an illustrative supplier that develops a connector a robot can attach reliably. Its market could include research spacecraft, communications platforms, and a future computing station. Another business might test whether satellite software can recover from corrupted memory. A third might turn complex observations into maps that land managers can actually use.

None of those businesses needs to own a launch vehicle. Each needs to solve a problem someone is willing to pay to have solved. Their opportunities would depend on actual procurement and demand; this is a map of possible roles, not a claim that these jobs have already appeared at scale.

The distinction matters for education as well. Building space infrastructure calls for software and physics, but also electrical work, manufacturing quality, logistics, accounting, technical writing, and public administration. A system can fail because its purchasing records are wrong just as surely as because its equations are wrong.

We should measure opportunity by the useful work created and services improved, not merely by the height at which the equipment operates.

Learn in steps that leave something valuable behind

An ambitious program is stronger when each stage has a purpose of its own. A small image-processing experiment should teach something about image processing, even if a giant orbital cloud never follows. A lightweight power module should have potential customers beyond one speculative megaproject.

This creates an important test for a development roadmap: if the next grand stage is canceled, what useful capability remains?

The answer might be a tested component, a better scientific dataset, or a smaller service with repeat customers. If the answer is only a pile of equipment useful after five more unproven breakthroughs, the plan carries more interlocking risk.

This does not mean every research experiment must earn money. Public science often creates knowledge that no single customer would finance. It means the purpose should be clear. Buying knowledge, demonstrating a capability, and subsidizing a commercial service are different decisions.

The path upward becomes more credible when people can explain what each step accomplished, what it cost, and why the next step deserves to happen.

What would prove this?

The next convincing milestones should be understandable without a promotional video.

For orbital computing, an operator should demonstrate useful work over a meaningful period, report power consumption and failures, and explain how the result reaches a customer. Claims of lower cost should include launch, equipment, communications, lost capacity, replacement, and disposal.

For power transmission, researchers should measure energy at the receiver, explain the full conversion chain, and show safe, controlled operation. A commercial claim needs more than a successful signal: it needs dependable delivery at a defensible cost.

For both, repeat performance matters. One successful run is encouraging. Reliable service under changing conditions is much stronger evidence.

Independent customers and verifiable operating records would help separate an industry from a sequence of experiments. So would clear reporting when a design fails. Failures can improve engineering—but only if the lessons are understood.

The best question remains simple: what useful service has this system delivered, and what did delivering it really require?

The construction project begins with a choice

Calling this the biggest construction project humanity has never attempted is an invitation to imagine its potential scale, not a measured ranking. No finished orbital energy-and-computing economy exists for us to compare with the world's great infrastructure networks.

If it develops, it will probably emerge in pieces. A better onboard processor. A more reliable communications link. A lighter solar array. A robot that replaces a component without a person standing nearby. A customer who buys the service again because it worked.

Some pieces will succeed without leading to the grandest vision. Others will reveal limits that make a different path more sensible. That is not a disappointing version of progress. It is how useful systems take shape.

Look again at the imagined morning that opened this article. The rancher does not need to admire an orbital computer. She needs a trustworthy warning. The clinic does not need to celebrate a new cloud region. It needs sound results. The lunar robot does not need a slogan about abundance. It needs enough power to finish its task and get home.

Those practical needs are where the story becomes real.

The Sun provides the opportunity. Engineering, economics, and human judgment will decide how much of it we can use.

Next article