SpaceX Wants to Turn Orbit Into a Power Plant for AI

SpaceX is not merely trying to launch more satellites. It is trying to change the price of intelligence. With a public offering that valued the company at more than two trillion dollars and handed Elon Musk the war chest to push deeper into artificial intelligence infrastructure, the rocket maker is now betting that the next great data centre will not sit in Virginia, Texas or Ireland, but above the atmosphere, where the sun never goes out and electricity is no longer the first constraint.

That is the wager behind its orbital data centre plan: compute powered by sunlight, cooled by space, and scaled through a launch system SpaceX itself controls. If the idea works, it could redraw the economics of AI, weaken the grip of terrestrial power grids, and push the value of cheap electricity to the centre of corporate life. If it fails, it will join a long list of grand technology schemes that ran into the hard limits of physics, heat and cost.

A capital raise built for a different kind of empire

The scale of SpaceX’s public debut tells its own story. The company raised about $75 billion by selling roughly 555 million new shares at $135 each. On paper, that implied a valuation of around $1.8 trillion on a fully diluted basis. In the market’s shorthand, the figure was larger still: more than two trillion dollars, enough to make SpaceX one of the biggest companies on Earth and turn Musk into the first person many observers would describe as a trillionaire.

But the point of the offering was not simply to mint a new billionaire class or create another line item in the index funds. The sale involved only a small sliver of the company, roughly four to five per cent of the total shares outstanding. Musk kept about 82 per cent of the voting power, which means the public market bought into the story without taking control of the plot. That matters. This was not a surrender to Wall Street. It was a financing event for an industrial strategy.

The capital raise also revealed how carefully SpaceX has managed supply. The float was restrained, retail investors got a substantial allocation, and demand was fierce from the start. The result was a market debut that looked less like a conventional listing and more like a rationed entry into a very expensive future. The message was plain: the company does not need public investors to run the business day to day, but it does need their money to fund the next phase of infrastructure.

That phase is not only rockets and satellites. It is compute. More precisely, it is the industrialisation of compute at a scale that treats intelligence as a utility, not a luxury.

The real bottleneck in artificial intelligence is power

The current AI debate is full of noise, much of it self-serving. People argue about model quality, training data, regulation, labour disruption and the ethics of synthetic media. Those are all real questions. But underneath them sits a simpler one: how much power can the system reliably turn into useful cognition?

That is why the comparison with the human brain matters. A brain runs on about 20 watts — roughly the draw of a refrigerator bulb — and over a year uses about 175 kilowatt-hours, or around $20 of electricity for the average American household. From that small energy budget, a knowledge worker can generate roughly $65,000 in annual economic value. The ratio between the cost of powering intelligence and the value it produces is huge, and for most of history it could not be engineered at scale.

Biology produced intelligence through reproduction, education and time. There was no alternate factory for minds. AI changed that. It allowed firms to manufacture some forms of cognition from electricity rather than flesh. That is not a sentimental point. It is the foundation of the current technology boom, and it is why the most important firms in the sector are no longer just software companies. They are energy conversion businesses.

Once intelligence becomes an output that can be priced against electricity, the old corporate map looks dated. A call centre, a coding shop, a legal back office, a research team and a design studio all begin to compete on the same basis: who can turn the cheapest energy into the most useful work. That is the competition SpaceX is moving towards. Orbital compute is not a science-fiction flourish; it is a bid to win the energy race before rivals do.

Why orbit changes the equation

Terrestrial data centres are already straining against the grid. The problem is not lack of ambition. It is that the Earth is a bad place to collect uninterrupted solar power at massive scale. The sun sets. Clouds form. Weather shifts. Seasons change. Half the planet sits in darkness at any given moment. Even the best ground-based solar farms and battery systems remain tethered to weather, land use, transmission lines and local politics.

Above the atmosphere, those excuses vanish. Solar panels in orbit can receive constant radiation without clouds or night. Waste heat, the main engineering problem in dense computing, can be shed into the vacuum through radiators. A properly designed orbital system can in principle sit in continuous sunlight and convert that flow into sustained compute.

The source material points to the scale of the opportunity. The sun delivers about 170,000 trillion watts to Earth. Humanity uses roughly 20 trillion watts across the whole of civilisation — transport, manufacturing, homes and digital infrastructure included. In other words, we are already operating on a tiny fraction of the energy arriving at the planet. Orbit offers a way to capture a larger share of that stream without fighting the Earth’s daily cycle of dark and light.

That is why the orbital data centre thesis has attracted so much attention. It promises not just cheaper power but a different relationship between power and time. Ground-based compute is interrupted by the clock. Orbital compute is not. That has consequences for training large models, for running inference, and for the economics of services that need high uptime. It also has consequences for the balance sheet, because electricity is not merely an input. It is the input that decides whether the business is viable at all.

The argument is not that every data centre will move into space. It is that the frontier of scaling AI may shift there first, where the physics is friendlier and the energy supply looks less like a constraint and more like a gift.

SpaceX’s advantage is not the idea — it is the stack

Many companies can dream about space-based computing. Very few can build the stack needed to attempt it. That is where SpaceX has a real edge, and not a small one.

The company already makes the rockets, through Starship, that would be needed to launch bulky hardware at a lower cost than most competitors could manage. It has spent years building and flying thousands of Starlink satellites, which gives it experience in mass-manufacturing spacecraft rather than treating each launch as a one-off event. It has industrial relationships around solar cells, and through Tesla it benefits from an ecosystem that understands batteries, power systems and high-volume manufacturing. On the demand side, xAI supplies a hungry internal customer for computing power, while outside players reportedly rent capacity from that broader network at eye-watering rates.

This is the sort of vertical integration that gives companies strategic options other firms simply do not have. A startup might have a concept and some venture capital. A hyperscaler might have money and engineers but no launch business. A launch company might have access to orbit but no reason to build a compute platform. SpaceX can move through all of those stages at once.

That does not make success likely. It makes failure more expensive if it comes. But it also means the company is not trying to build a toy demonstration. It is trying to own the supply chain from Earth to orbit to compute.

The strategic logic is easy to see. If the cost of launch keeps falling, and if satellite manufacturing becomes more standardised, then the first company that can combine cheap lift, cheap solar collection and useful compute capacity in orbit may enjoy a long run of advantage. The key word is may. The market loves certainty when it is selling shares, but industrial reality does not obey marketing.

Still, the stack matters. In technology, as in war, systems beat slogans. SpaceX has a system.

Terrestrial rivals are chasing the same energy problem from below

It would be comforting to pretend that orbital data centres are a standalone moonshot, but they are better understood as one answer to a question every serious tech company is now asking: where does the power come from?

Microsoft has already taken the unusual step of backing a restart at Three Mile Island, the nuclear plant in Pennsylvania that became a symbol of atomic fear after its partial meltdown in 1979. That move tells you how serious the power crunch has become. A company that once built cloud services in data centres now needs to think like a utility operator, a reactor investor and a grid planner.

The same pressure is visible elsewhere. Data centre developers face years-long waits just to secure a grid connection. Solar, natural gas, battery storage and nuclear restarts are all being pulled into the same race. Investment decisions that once revolved around software adoption now hinge on interconnection queues, transformer availability and the cost of megawatts. The old age of software-only abundance is gone.

This is why the AI build-out has become so strange. Firms that preached asset-light growth are suddenly desperate for steel, copper, silicon and concrete. The cloud is becoming a power business by another name. It is less a software revolution than a return to industrial reality.

SpaceX’s orbital idea should be judged in that context. It is not a distraction from the terrestrial power race; it is a bid to step outside it. If ground-based computing must compete for land, permits, cooling water and grid access, then orbit offers a clean break from those constraints. The catch is that new constraints appear in their place: launch mass, maintenance complexity, latency and repairability.

Terrestrial rivals are not sitting still. They are buying turbines, hedging with nuclear, and locking up power contracts. SpaceX is making the opposite bet: that the cleanest source of scalable energy is already above us, and that the hard part is not finding electricity but hauling the hardware there cheaply enough to matter.

The economics of abundant cognition

The most important thing about AI is not that it can imitate a human. It is that it can produce work at a cost structure humans cannot match. Once a machine can draft documents, write code, analyse data, summarise meetings and triage support tickets for the price of electricity, the market for many forms of white-collar labour begins to change.

That shift is often described in moral language, but the economics are more brutal than that. Wages are the price of scarce cognition. For most of human history, thought was expensive because it depended on rare human minds. AI turns thought into an engineering output. If competition drives the price of that output down towards the cost of power, then the entire labour market for knowledge work comes under pressure.

The source material argues that the eventual benefit may flow to consumers. That is likely true, but only after a fight. First the surplus tends to go to firms that own the infrastructure and the models. Then, if competition does its job, the savings spread outward. Medical advice becomes cheaper. Legal assistance becomes less exclusive. Tutoring, design and engineering support become available to people who could never have paid human experts at prevailing rates.

The deeper point is that GDP may become a poor guide to what is happening. GDP counts output by price times volume. If AI pushes the price of cognitive tasks down while multiplying the amount done, measured economic activity could look flat even as actual abundance rises. The economy can get better while the statistic gets worse. That is a common pattern when a new technology destroys old rents.

It is also why many of the loudest arguments about AI miss the central issue. They treat it as a labour story. It is really an energy story. The companies that can convert sunlight, uranium, gas, copper and silicon into cognition at the lowest cost will shape the next phase of growth.

Why the old energy story still controls the new economy

Civilisation has always expanded through energy mastery. Human and animal muscle came first. Biomass followed. Water wheels, windmills and early mechanical systems improved throughput. Coal and steam then multiplied output in a way that transformed cities, transport and trade. Oil and gas carried the industrial model into the twentieth century, while electricity made power more flexible and more widely usable.

The post-war period intensified the pattern. From roughly 1950 onwards, the world entered a phase of extraordinary growth in energy consumption, industrial production and material output. That same period also brought pollution, warming oceans, biodiversity loss and environmental strain. Energy abundance has never come free. It has always carried a bill, and nature eventually sends it.

Renewables have improved the picture. Solar and wind are cheaper than they were, more scalable, and better understood. But they still depend on land, weather and grids. They help a great deal, yet they do not remove the basic arithmetic of intermittency. Nuclear offers more consistent generation, but it remains politically and commercially difficult. Fusion remains a promise waiting for a date.

SpaceX’s orbital concept sits at the edge of this larger energy story. It is an attempt to leap over some of the old bottlenecks rather than squeeze through them. That is why it should be taken seriously even by people who doubt the timeline. Big industrial shifts often begin as bad jokes. Railways were once dismissed as overbuilt toys. Commercial aviation was once a sideshow. The point is not that every audacious idea works. It is that the next industrial regime usually looks absurd before it looks obvious.

The present regime is already straining. If AI keeps growing, it will need more power. If power stays scarce, AI stays expensive. If AI stays expensive, its broader social effects remain limited. The entire chain is constrained by electricity.

What orbital data centres would have to get right

For all the romance around orbit, the engineering list is unforgiving. A useful space-based data centre must launch cheaply, survive radiation, manage heat, handle repairs, avoid catastrophic failure and deliver enough compute to justify the deployment. That is before anyone asks how the system connects to Earth-bound users in real time.

Latency is one of the first obvious problems. Some tasks, like batch training or offline inference, can tolerate delay. Others cannot. Real-time applications — trading systems, interactive services, many consumer tools — care about speed and reliability. An orbital system may be best for workloads that are compute-heavy but not latency-sensitive. That limits the initial market.

Heat dissipation is another issue. Space is cold, but cold space is not the same thing as easy cooling. High-density compute still generates waste heat that must be pushed away through radiators. That hardware adds mass. Mass adds launch cost. Launch cost remains one of the biggest barriers to the entire concept. SpaceX can reduce it, but not abolish it.

Then there is maintenance. Terrestrial data centres can be repaired by teams of technicians. Hardware in orbit cannot. Redundancy becomes essential. Fault tolerance becomes a design principle, not a luxury. Every component has to be treated as if a truck roll will never happen, because it won’t.

This is why scepticism is healthy. The orbital data centre pitch can sound like a glossy answer to a difficult problem, but engineering is not impressed by adjectives. It cares about mass, thermal load, failure rate and cost per usable unit of compute. If SpaceX cannot beat the terrestrial alternatives on those measures over time, the plan will remain a prestigious experiment.

Still, the company has one advantage that should not be underestimated: the willingness to use a rocket factory as an industrial platform rather than a transport service. That is a serious idea. It may be wrong. It is not silly.

The long shadow of the public market

The public offering itself is part of the story because it changes the stakes. A private company can indulge in long gestation periods and speculative projects without facing daily scrutiny from the market. A public company can still do that, but every decision is now filtered through shareholders, analysts and the broader public appetite for risk.

SpaceX, however, entered the public markets with unusual control intact. Musk’s voting power remains dominant, so the company can pursue a plan that may not satisfy conservative investors in the short term. That structure is useful for moonshots, but it also reduces the normal checks that public markets impose. If the orbital compute thesis drags on, the company can still push ahead because the capital and the control are both concentrated.

That concentration is one reason the offering mattered so much. It gave SpaceX access to an enormous amount of fresh capital while preserving the founder’s authority over the strategic direction. Many firms use public money to lower the cost of capital. This one appears to be using it to build an industrial empire across launch, satellites, energy and intelligence.

That is a larger ambition than launching internet satellites or making better chips. It is an attempt to control the scarce inputs that will matter most if AI becomes a central economic utility. Those inputs are power, manufacturing capacity, launch access and demand. SpaceX has a claim on all four.

Whether the market has really understood this is another matter. Investors tend to price visible revenue, not latent industrial option value. They can see rockets. They can see satellites. They can see AI spending. They are less good at seeing what happens when those pieces are fused into a single system. That blindness creates opportunity for the company and risk for the crowd.

The political economy of cheap intelligence

If SpaceX and its rivals succeed, the consequences will not stop at shareholder returns. Cheap intelligence changes the politics of labour, education and industrial policy.

Start with work. If a growing share of routine knowledge tasks can be done by machines, the bargaining power of many white-collar professions weakens. Not all at once, and not evenly, but enough to unsettle the old hierarchy. Roles that once looked secure may become easier to outsource to software. The salaries that once signalled scarcity may begin to look more like temporary rents.

Then consider education. If every student can have an always-available tutor that is cheap, patient and personalised, the old logic of access begins to break down. That sounds positive, and it is, but it also puts pressure on institutions that charge heavily for credentials while offering patchy value. Universities and training providers will have to justify themselves more aggressively.

Law and medicine will face similar shifts. Not because AI replaces judgment, but because it cheapens the preparation and routine analysis that surround judgment. A law firm that once relied on armies of junior staff may need fewer hands. A medical practice may use AI to triage and advise more cheaply. Even if the final decision stays human, the cost of getting to that decision falls.

For governments, the question is what to tax, what to regulate and what to subsidise. If the main bottleneck becomes energy, not human labour, then industrial policy shifts toward power generation, transmission, storage and hardware supply chains. That is a familiar sort of politics in a new costume. It is also why the energy debate will become more central, not less, as AI systems improve.

Space-based compute pushes this logic further. It says the world’s cheapest scalable power may not be on the ground at all. If that proves true, then the fight over AI supremacy becomes a fight over access to orbit.

The real gamble is not on satellites but on civilisation’s next input

The cleanest way to understand SpaceX’s orbital data centre plan is not as a gimmick, and not even as a moonshot. It is as a bet that the next stage of economic growth will be determined by who can most efficiently turn energy into cognition.

That is why the sun matters so much in this story. The sun is not just a backdrop. It is the ultimate upstream input. Earth receives far more energy than civilisation currently uses, and the limiting factor is not supply in the cosmic sense. It is extraction, conversion and storage. Orbit changes extraction. It may also change conversion. If SpaceX can keep pushing launch costs down while building reliable satellite compute hardware, it could make orbital energy feel less like a fantasy and more like an industrial location.

That possibility is exactly what makes the plan so consequential. It forces everyone else to think differently about the resource base of the digital economy. Cloud computing was once treated as a software problem. AI has turned it into a power problem. SpaceX is now suggesting that the answer may lie in moving the power source closer to the sun and the hardware farther from the weather.

The risks remain large. The timeline may be long. The cost of orbital capacity may stay higher than terrestrial alternatives for years, perhaps longer. But the direction of travel is hard to ignore. If intelligence is becoming a commodity, then the winner will be the firm that can make that commodity cheapest at the largest scale. If the cheapest energy is in orbit, then orbit stops being a curiosity and becomes a battleground.

That is the real wager behind SpaceX’s trillion-dollar gambit: not that rockets are glamorous, but that the future of compute will belong to whoever can industrialise sunlight before anyone else figures out how to compete.