The Data-Centre Supercycle: From Earth to Space

 And then there is space

Human ambition is not limited to Earth. Space exploration expands our field of vision—and our economic possibilities—to a scale as vast as the universe itself.

Space offers new frontiers for scientific research, energy generation, communications, manufacturing, resource exploration, tourism and, eventually, digital infrastructure. Data centres may one day be constructed in orbit, where they could benefit from extensive solar exposure and an unobstructed view of the coldness of deep space.

However, space is not a zero-temperature refrigerator. In a vacuum, there is no air to carry heat away through convection. Orbital data centres would therefore require large and carefully designed radiators to release their waste heat. Cooling may be achievable, but it remains one of the greatest engineering challenges facing orbital computing.

The opportunity is nevertheless real. Certain orbits could offer near-continuous solar power without clouds, changing weather or competition for terrestrial land. Orbital facilities could also process the enormous quantities of data generated by satellites before transmitting the most valuable information back to Earth.

The European Commission-backed ASCEND study has already examined the feasibility and environmental impact of large-scale data centres in space. Its findings suggest that orbital computing could eventually contribute to European digital sovereignty, provided that launch, maintenance, radiation protection and thermal-management challenges can be overcome. Space-based data centres are therefore moving from science fiction towards serious technological investigation.

But the more important relationship between AI and space works in both directions. Space may eventually provide a location for AI infrastructure, while AI will be indispensable to humanity’s exploration of space.

The farther humans travel from Earth, the less practical direct human control becomes. Communication delays, interruptions and the impossibility of immediate physical assistance mean that spacecraft, exploration vehicles, habitats and life-support systems must be capable of making critical decisions autonomously.

AI will navigate unfamiliar terrain, detect equipment failures, manage energy and oxygen, monitor astronauts’ health, maintain habitats, coordinate robots and respond to dangers before instructions can arrive from Earth. NASA already uses autonomous systems to allow spacecraft and planetary rovers to continue operating when they are out of contact with mission control. The Perseverance rover for example performsmost of its driving autonomously.

Human survival beyond Earth will therefore depend on intelligent machines capable of perceiving, reasoning and acting without constant supervision.

This will create a layered computing architecture. Onboard and orbital systems will make urgent local decisions. Earth-based data centres will train the models, simulate missions, analyse scientific discoveries, validate software, coordinate fleets and provide strategic intelligence.

Earth will teach; space will execute.

For many years, terrestrial data centres will remain essential to supporting orbital infrastructure. Over time, computing systems in space may acquire enough operational autonomy to become less dependent on Earth—and perhaps less constrained by the slow, territorially fragmented regulatory machinery created for terrestrial technology.

That possibility will force a profound legal and political debate. Space should not become a lawless refuge for uncontrolled AI. But neither should legislation written for yesterday’s technologies prevent humanity from developing the infrastructure needed for tomorrow’s exploration.

Orbital data centres will not eliminate the need for terrestrial facilities. They will add another layer to the global—and eventually interplanetary—computing system. Space will not reduce the demand for data centres. It will multiply the number of places in which computing power is required.

The next users of AI will be machines

Even before humanity establishes a substantial presence beyond Earth, the world may already be underestimating data-centre demand because most forecasts remain too focused on human behaviour.

Today, artificial intelligence is largely used by people. We ask a chatbot a question, generate an image, analyse a document or instruct a digital agent to complete a task. These interactions may be computationally demanding, but they are intermittent. Humans sleep, pause and disconnect.

Intelligent machines will not.

The next great wave of AI will come from autonomous vehicles, humanoid robots, industrial machines, delivery systems, drones and other forms of physical AI. These systems will operate continuously, processing their surroundings through cameras, microphones, radar, lidar and other sensors.

In effect, every autonomous machine will produce and interpret a constantly changing film of its environment. It must repeatedly answer questions that humans process almost instinctively: What am I seeing? What is moving? What will happen next? Is that person about to cross the road? Can I safely pick up this object? How should I react if the situation changes?

That is an entirely different category of computational demand.

From human prompts to continuous machine perception

An autonomous car does not simply calculate a route. It must identify vehicles, pedestrians, road markings and unexpected hazards; predict their probable movements; interpret the intentions of other road users; and select a safe course of action within milliseconds.

A humanoid robot faces an even more complex problem. It must understand language, recognise objects, maintain balance, coordinate its limbs, calculate force and distance, and adapt its behaviour to environments that were never perfectly represented in its training data.

Safety-critical decisions will generally be processed locally because a car or robot cannot wait for a distant cloud server before avoiding a collision. But this does not reduce the importance of data centres. It changes their role.

The machine acts locally, but it learns globally.

Behind every deployed robot or autonomous vehicle will be an enormous central computing system used to train models, create synthetic environments, simulate rare events, evaluate failures, validate software updates and convert the experience of one machine into improved intelligence for an entire fleet.

Waymo reports nearly 200 million fully autonomous road miles, but its vehicles have travelled billions more inside virtual environments. Its world-model technology generates camera and lidar simulations in real time, including unusual and dangerous situations that would be difficult to reproduce safely on public roads. Waymo’s experience demonstrates how physical deployment creates an even larger demand for centralised simulation. 

Every physical mile can produce many more virtual miles. Every mistake can be replayed thousands of times. Every new environment can create millions of simulated variations.

This creates a powerful machine-data flywheel: more machines generate more experience; more experience improves the models; better models make more machines commercially viable; and those additional machines generate still more data and demand for computing power.

If AI systems eventually approach or exceed human capabilities across a wide range of defined tasks—the possibility often associated with technological singularity—the demand could become partly self-generating. AI systems would use computing power not only to serve humans, but to design products, run scientific experiments, create simulations, coordinate other machines and improve future generations of AI.

Even without assuming a full singularity, persistent autonomous systems are enough to produce a second compute curve that is no longer limited by the number of humans, the length of the working day or our capacity to type questions.

Efficiency will not reduce demand

It is tempting to believe that better chips and more efficient models will moderate this expansion. History suggests the opposite.

When computing becomes cheaper, businesses discover more uses for it. When AI inference becomes more efficient, developers do not necessarily use less computing power; they deploy larger models, process richer information and run them more frequently. Video generation, advanced reasoning and autonomous agents can already require hundreds or thousands of times more energy per task than a basic text response.

This is the rebound effect of the AI economy: efficiency reduces the cost of intelligence, and cheaper intelligence creates more demand for intelligence.

The International Energy Agency estimates that global data-centre electricity consumption will rise from approximately 485 terawatt-hours in 2025 to around 950 terawatt-hours in 2030. Consumption by AI-focused facilities is expected to triple over the same period. Data-centre demand is therefore already growing far faster than overall electricity consumption. 

Those forecasts are extraordinary, yet they may still only partially capture the effect of millions of autonomous systems entering everyday economic life—or the longer-term emergence of orbital computing.

The limiting factor will not be algorithms alone. It will increasingly be access to electricity, grid connections, advanced chips, transformers, cooling systems, water, fibre networks and suitable land. The countries that secure these assets will possess the industrial foundations of the machine-intelligence economy.

Europe is falling behind the capital race

Europe remains rich in scientific talent, industrial expertise and private savings. It has globally important companies in semiconductor equipment, advanced manufacturing, automotive engineering, pharmaceuticals, aerospace, robotics and energy technology.

But it is losing ground in the infrastructure and capital mobilisation that turn research into global technological power.

American companies operate many of the leading closed-weight frontier systems: OpenAI’s GPT models, delivered through ChatGPT; Anthropic’s Claude; and Google’s Gemini. Chinese laboratories have built formidable open-weight model families, including DeepSeek, Alibaba’s Qwen and Moonshot AI’s Kimi, which developers can download, adapt and deploy.

The distinction is not absolute—American companies also publish open-weight models, while some Chinese systems remain proprietary. Nevertheless, the strategic picture is clear. The United States and China are setting much of the pace in frontier models, computing infrastructure and AI investment.

The 2026 Stanford AI index concluded that the performance gap between leading American and Chinese models had effectively closed, although the United States still produced more top-tier systems. Tracked private AI investment in the United States reached approximately $285.9 billion in 2025, compared with about $20.9 billion across Europe.

Europe’s problem is therefore not simply a lack of innovative people. It is a failure to finance scale.

Mario Draghi’s competitiveness report estimated that the European Union needs an additional €750–800 billion of productive investment every year—equivalent to 4.4–4.7 per cent of EU GDP. That figure covers the entire competitiveness agenda, including energy, digitalisation, defence, transport and decarbonisation; it is not an AI budget. But it reveals the scale of Europe’s capital problem. Draghi also observed that Europe has ample savings but fails to channel them efficiently into productive, high-risk investment. 

In 2022, EU households saved approximately €1.39 trillion, compared with €840 billion in the United States. Europe does not principally lack money. It lacks the financial machinery that can transform savings into moonshot capital at American speed and scale.

American markets can finance vast technological bets through equity, bonds, venture capital, private credit and hyperscaler cash flow. China can mobilise capital through a powerful combination of private enterprise and state-directed industrial policy. Both systems can produce waste, duplication and speculative bubbles—but they can also build infrastructure before demand has been proven beyond doubt.

Europe’s fragmented capital markets, bank-centred financing system and cautious institutional investors are far less suited to ventures whose collateral consists of intellectual property, talent and an uncertain technological future.

Implementation is also moving too slowly. According to the independent European Policy Innovation Council’s July 2026 tracker, only 60 of Draghi’s 383 recommendations had been fully implemented. A further 98 were partially implemented, while 225 remained unfinished.

Europe is not without options

The European Union has launched InvestAI, intended to mobilise €200 billion, and is establishing 19 AI factories. It is also pursuing large-scale AI gigafactories and aims to triple European data-centre capacity within five to seven years. These initiatives re important foundations, but announcements must now become powered, connected and operational infrastructure.

A recent scenario essay, Europe 2031: What Getting AI Wrong Means for Us, warns that if the continent remains on its present trajectory, it could “lose the ability to shape its own future”. The document is a scenario rather than a prediction, but its strategic message is difficult to dismiss: technological dependence will eventually become economic and political dependence. Europe would struggle to protect its values, finance its welfare systems or exercise genuine geopolitical influence. 

Europe’s best opportunity may not be to reproduce Silicon Valley company by company. It is to apply AI throughout the industries in which Europe already possesses customers, specialised data, engineering knowledge and regulatory legitimacy.

Finnish entrepreneur Peter Sarlin represents that possibility. He co-founded and led Silo AI, which AMD acquired for approximately $665 million in cash. Sarlin has argued that Europe can build lasting advantages by embedding AI across its industrial economy, where European companies already control customer relationships and valuable data.

His latest venture, Qutwo, is developing software intended to orchestrate workloads across classical, quantum-inspired and, eventually, quantum systems. Quantum processing units may one day accelerate particular optimisation, sampling and simulation tasks, just as GPUs powered the generative-AI revolution. But quantum remains a strategic possibility, not yet a replacement for conventional computing.

Europe should invest in that future without using it as an excuse to neglect the data centres, grids and accelerators it needs today.

Why Europe needs America

Europe cannot rebuild the entire AI stack quickly enough through isolation. Nor should it confuse sovereignty with technological autarky.

It needs American capital, accelerators, cloud expertise, frontier models and experience in constructing hyperscale infrastructure. The Draghi report itself recognised that continued access to the latest American AI models and processors is essential to Europe’s competitiveness.

But partnership must not mean permanent dependency.

Europe should propose a transatlantic AI, space-computing and data-centre compact under which American companies invest in large-scale European infrastructure located on European soil, operating under European law and supported by long-term agreements on energy, cybersecurity, data protection and access to advanced technology.

The partnership should ultimately extend beyond Earth. Europe and the United States possess complementary strengths in space launch, satellite manufacturing, communications, AI, scientific research and advanced computing. Together, they could establish the technical and legal foundations of orbital data infrastructure before China or another power determines those standards for them.

Europe can offer industrial customers, skilled engineers, research institutions, strategic locations, semiconductor expertise, political stability and access to one of the world’s largest markets. It also retains indispensable technological leverage through companies such as ASML and through its strength in advanced manufacturing, robotics, aerospace and scientific research.

European governments should accelerate permitting for strategically important data centres, grids and low-carbon electricity generation. European pension funds and insurers should be given better vehicles for investing in AI infrastructure and scale-ups. The European Investment Bank should take more first-loss and project risk. Regulatory obligations should be harmonised across the Single Market so that a company approved in one member state does not have to navigate 26 additional interpretations.

Europe can also turn trust into a competitive advantage. Public opinion is not simply anti-AI: 62 per cent of Europeans view robots and AI positively in the workplace, while 84 per cent believe the technology requires careful management. Europeans are asking for safeguards, not technological paralysis.

The correct response is therefore not deregulation for its own sake. It is predictable, harmonised regulation that protects citizens while allowing companies to build. Trust can accelerate adoption when rules are clear; fragmented and constantly changing requirements destroy it.

The choice is partnership or irrelevance

AI sovereignty should mean that Europe can choose its partners, maintain critical infrastructure, protect sensitive data and continue operating during a crisis. It should not mean attempting to manufacture every chip, train every model and build every platform without allies.

The United States needs a technologically capable Europe as a strategic partner, industrial market and democratic counterweight. Europe needs America’s capital and AI ecosystem if it is to participate meaningfully in the next era of computing.

The coming data-centre supercycle will not merely support more internet searches or office applications. It will provide the memory, simulation and collective learning capacity of an expanding population of intelligent machines—first on Earth and eventually beyond it.

Today, humans are the principal customers of artificial intelligence. Tomorrow, machines will increasingly become customers of intelligence themselves.

That is why the real demand for data centres may be far greater than today’s forecasts suggest—and why Europe must act before the infrastructure, capital, energy and power of the machine age become permanently concentrated elsewhere.