Google and SpaceX want AI data centers in orbit

AI computing is already seeking its next frontiers

According to TechCrunch, Google and SpaceX are reportedly discussing a project that is as ambitious as it is speculative: installing computing and data storage in orbit. The information, reported based on a report referring to discussions between the two groups, has not reached the stage of an official announcement or an industrial timetable. But it says a great deal about the state of the market: the battle for artificial intelligence is no longer being fought solely over models, chips or interfaces; it is now shifting toward access to a resource that has become critical: compute.

Since the explosion of generative AI at the end of 2022, major technology players have been engaged in an accelerated race for infrastructure. Microsoft is investing heavily in its Azure data centers to power OpenAI. Amazon is pushing AWS and its in-house chips. Meta is multiplying its purchases of Nvidia GPUs. Google, for its part, is betting both on its TPUs and on expanding its cloud capabilities. The bottleneck is well known: training and inference for modern models require colossal computing power, ever-larger storage volumes and, above all, a stable power supply.

In this context, the idea of moving part of this infrastructure beyond the planet may seem like science fiction. Yet the economic logic underlying it is very real. Suitable land for new data centers is becoming scarce, access to electricity is becoming a limiting factor, and grid connection times are growing longer in several regions of the world. In Europe and France alike, these pressures are already visible: delayed projects, debates over water use, energy consumption and the local acceptability of hyperscale sites.

What the report says about Google and SpaceX

According to TechCrunch, Google and SpaceX are reportedly discussing the possibility of deploying infrastructure in orbit capable of providing computing and storage. The scenario being considered remains unclear: at this stage, it is neither a commercial product nor a detailed public investment plan. Neither group has formalized a specific program comparable to a conventional industrial rollout. The significance of the information therefore lies less in the existence of an approved project than in the fact that companies of this size are seriously exploring such an option.

The pairing is not insignificant. Google has leading expertise in cloud infrastructure, large-scale data centers and AI accelerators. SpaceX, for its part, has profoundly changed the economics of space through the reusability of Falcon 9 launch vehicles and the deployment of the Starlink constellation. If any company is now capable of making repeated launches of heavy equipment into orbit credible, at least on paper, it is SpaceX.

The hypothesis rests on several theoretical promises. In orbit, solar energy is available in abundance, without weather-related alternation comparable to that on the ground. Land, meanwhile, ceases to be a constraint in the traditional sense. And certain physical limitations of terrestrial sites, such as proximity to overloaded power grids or conflicts over water use, take on a different nature. For companies facing explosive demand for computing capacity, these potential advantages are enough to justify exploratory studies.

But it must be repeated: we are still in the realm of discussion. The report cited by TechCrunch describes neither a validated architecture, finalized costs nor a deployment timeline. The journalistic interest of this information lies precisely in what it reveals about the mindset of hyperscalers: faced with AI pressure, even the most extreme options are now entering the realm of possibility.

Why compute has become the key battleground

The AI sector long emphasized model performance. Yet for the past two years, the market's real hierarchy has been taking shape elsewhere: in the ability to secure GPU supply chains, build data centers faster than competitors, and find the electricity needed to run them. Nvidia symbolized this shift with record revenues from its AI-dedicated chips, while cloud giants announced capital expenditure amounting to tens of billions of dollars.

The problem is not just the cost of components. A modern AI data center combines high-density servers, complex cooling systems, very fast network interconnections and electricity consumption that can reach levels comparable to those of heavy industrial infrastructure. Training large models, as well as large-scale inference for millions of users, turns AI into a matter of energy planning.

In France, this debate is gaining momentum. Data center projects are running into trade-offs over land availability, grid connection, climate targets and water management. In Ireland, the Netherlands and certain areas of Germany, authorities have already had to more strictly regulate the establishment of new sites. Europe wants to attract AI investment, but it is simultaneously discovering the material cost of that ambition.

In this framework, orbit appears as an extreme answer to a very terrestrial problem: how can computing power continue to grow when physical constraints on the ground are mounting? The mere existence of such an avenue shows how far the sector's center of gravity has shifted. After the GPU shortage, the next competition concerns access to extreme infrastructure.

A fascinating project, but one facing massive obstacles

From a technical and economic standpoint, the idea of an orbital data center nevertheless raises considerable difficulties. The first is obvious: launch costs. Even if SpaceX has lowered the price of space access, sending computing racks, power systems, communications equipment and maintenance devices into orbit remains infinitely more expensive than building a site on the ground. Added to this is the constraint of reliability: in space, a failure cannot be handled with a simple on-site intervention.

Cooling is another major challenge. From afar, space seems cold. In practice, dissipating heat in a vacuum is a complex engineering problem, primarily relying on thermal radiation. Yet AI accelerators generate enormous amounts of heat. Designing systems capable of removing this energy efficiently, reliably and economically would represent a significant technological leap.

There is also latency and connectivity to consider. Computing capacity in orbit only makes sense if it can exchange data quickly with terrestrial networks and end users. For certain storage, deferred-processing or highly specialized computing applications, this may be feasible. For large-scale real-time inference, the issue becomes more delicate. The business model would therefore depend heavily on the tasks assigned to this infrastructure.

Finally, regulatory and environmental issues do not disappear when leaving Earth. Space debris, communications security, data sovereignty, liability in the event of an incident: all are issues that would take on a new dimension. In Europe, where cloud and data governance is already sensitive, orbital infrastructure operated by American giants would inevitably raise additional political questions.

What this avenue changes for the industry, including in Europe

Even if the project never goes beyond the exploratory stage, it already serves as a strong signal for the entire sector. It indicates that digital leaders now view AI infrastructure as a field for radical innovation. In other words, the next competitive breakthrough could come less from a spectacular new model than from the ability to guarantee abundant, available and relatively predictable computing over the long term.

For European players, this development is strategic. The European Union is investing in supercomputers, semiconductor factories and so-called sovereign cloud capabilities. But compared with the financial resources of Google, Microsoft, Amazon or Meta, the gap remains immense. If competition shifts toward increasingly capital-intensive infrastructure, the risk is that European dependency will grow even stronger.

France is not absent from the issue. Between ambitions around cloud, announcements on AI computing capacity and growing interest in next-generation data centers, the country is seeking to position itself. But the very idea of an orbital data center is a stark reminder of a harsh reality: AI leadership increasingly depends on combining software, hardware, energy and industrial logistics. In this area, companies able to align cloud, chips, launch vehicles, networks and capital have an overwhelming advantage.

The lesson of this matter is not that data centers will leave Earth tomorrow, but that tech giants are already exploring every option to avoid a lasting compute shortage.

Toward AI shaped by access to energy and space

The discussion reported by TechCrunch may still lead nowhere, or to a mere research program. But it opens a broader perspective: AI is entering a phase in which its limits are no longer primarily algorithmic. Models are advancing, certainly, but their deployment depends on physical resources that are increasingly scarce and costly. Electricity, land, supply chains, networks and potentially orbit itself are becoming variables of power.

This development could reshape the sector's hierarchy. Companies that control the entire stack, from chips to energy infrastructure via networks and cloud platforms, will be best placed to absorb the next wave of demand. Conversely, players that depend on third-party suppliers for every link risk being subject to capacity trade-offs, price increases and delays.

In the medium term, the most realistic answer is likely to remain terrestrial: denser data centers, new cooling systems, locations near abundant energy sources, increased use of nuclear power, hydropower or solar power, and software optimization to make better use of existing hardware. But the mere fact that orbit is entering the conversation shows that the industry is preparing for a scenario in which demand for AI compute continues to grow at an extraordinary pace.

If this trajectory is confirmed, the next decade of artificial intelligence could be marked less by a war of models than by a geopolitics of infrastructure. In this landscape, players able to turn energy, capital and even access to space into a competitive advantage will hold a growing share of global technological power.