SoftBank wants to invest €75 billion in data centers in France

SoftBank is waving up to €75 billion for data centers in France, a massive signal in the new AI battle

The announcement is spectacular because of its size, but even more so because of what it reveals. According to TechCrunch, which reports statements from the Japanese group, SoftBank says it can invest up to €75 billion to build data centers in France. The project mentioned would target up to 5 gigawatts of additional capacity, a scale rarely mentioned in Europe for infrastructure tied to intensive computing and artificial intelligence. At this level, this is no longer a simple real estate plan or a conventional cloud capacity expansion: it is an industrial projection that simultaneously touches on energy, land, networks, semiconductors, digital sovereignty, and geopolitical competition.

The €75 billion figure naturally calls for caution. The conditional is appropriate, and SoftBank has not, at this stage, presented a detailed site-by-site deployment schedule comparable to a fully locked-in investment program. But even taken as an ambition, the announcement carries considerable weight. It places France at the center of a reconfiguration of the European AI infrastructure market, at a time when American players are accelerating with tens of billions of dollars, hyperscalers are seeking energy footholds in Europe, and governments are trying to reconcile industrial attractiveness, technological sovereignty, and grid constraints.

The most important point is probably elsewhere than in the sum itself. The real strategic substance of this announcement is electrical capacity. In generative AI, in training large models as in large-scale inference, the question is no longer only who has the best algorithms or the latest GPUs. It is also who can power, cool, connect, and operate infrastructure of several hundred megawatts, or even several gigawatts. In this game, France has real strengths: a low-carbon electricity mix dominated by nuclear power, a heavy industrial tradition, a structured electricity transmission network, well-established telecom and cloud operators, as well as a growing political will to make the country a European hub for AI.

The fact that this announcement comes from SoftBank is not insignificant. Masayoshi Son’s group built its reputation on gigantic bets, sometimes visionary, sometimes excessive, often structuring for the global technology ecosystem. Its recent history, from the Vision Fund to Arm, including the setbacks of WeWork and other emblematic holdings, requires any investment promise to be read with a double lens: that of strategic ambition and that of concrete execution. But in AI, SoftBank has regained a central role, notably since Arm’s return to the stock market and the renewed acceleration of global investment around chips, servers, and computing capacity.

In this context, France no longer appears merely as one market among others. It is becoming a possible concentration ground for the physical layers of European AI. That is what makes the announcement important beyond its communication effect. It confirms that the next battle is no longer being fought only over models, but over infrastructure. And in Europe, that battle now involves very concrete questions: where to find 5 connectable gigawatts, at what cost, on what timeline, with what permits, with which cloud partners, and with what political acceptability.

Why France is attracting intensive computing projects: electricity, land, networks, and a political window

To understand the significance of the announcement, we need to go back to the transformation of France’s role in the European digital economy. For a long time, France was seen as a major digital consumer market, with telecom champions and a few cloud players, but less central than Ireland, the Netherlands, Germany, or the Nordic countries in the geography of very large data centers. That hierarchy is changing, under the combined effect of AI and energy.

The first French advantage is electrical. Data centers intended for AI have consumption profiles very different from those of traditional data centers. Where conventional infrastructure hosts more varied and sometimes smoother workloads, AI clusters concentrate thousands, or even tens of thousands, of GPUs and accelerators, with very high power densities. The need is not only for abundant electricity, but for electricity that is relatively stable, decarbonized, and available over the long term. On this front, the French nuclear fleet remains a major argument. It allows France to show a lower electricity carbon intensity than many European neighbors, which has become a leading criterion for major cloud operators, for companies concerned about their environmental footprint, and for investors who want to finance infrastructure compatible with credible ESG trajectories.

The second advantage is land and logistics. Data center megaprojects are not limited to a few buildings. They require large plots of land, manageable permitting timelines, the possibility of connection to the high-voltage power grid, very high-capacity fiber access, water resources or cooling alternatives, as well as a relatively predictable regulatory environment. France has industrial or peri-industrial zones capable of hosting this type of infrastructure, notably in regions where there is a push to reindustrialize or make use of existing brownfield sites.

The third factor is political. For several years, the French executive has been trying to establish the country as a European AI hub. That involves public research, start-ups, talent, but also increasingly the heavy layers of infrastructure. The discourse has changed: it is no longer only about having good labs or a few unicorns, it is about securing access to compute. In a market where training and operating advanced models depend on resources that are increasingly scarce and costly, sovereignty can no longer be thought of only at the software level.

This reorientation is not unique to France. Germany is also multiplying announcements around chip factories, sovereign clouds, and computing capacity. The Nordic countries are highlighting their decarbonized electricity, their cooling-friendly climate, and their historical experience in hosting large data centers. Ireland, despite growing grid tensions, remains a major foothold for American hyperscalers. The Netherlands has long played a central role in European digital hubs. But France can put forward a rare combination: market size, power grid, industrial base, public support, and geographic position.

It should also be recalled that the question of AI infrastructure in Europe has become more acute since 2023. The explosion in demand for Nvidia GPUs, the rise of generative models, production inference needs, and growing pressure from companies to have capacity located on the continent have highlighted a structural lag. Europe does not lack talent in mathematics, research, or engineering. It more often lacks massive, fast, and competitive access to compute. That is the deficit that announcements like SoftBank’s claim to address.

Finally, France benefits from a diplomatic and economic context favorable to this kind of narrative. In the global competition between the United States and China, Europe is seeking its place. It wants to attract investment without reducing itself to a mere host ground for foreign players. It wants to secure strategic capacity without cutting itself off from international capital. This tension runs through all current European industrial policy. SoftBank’s announcement fits exactly into that space: it promises colossal means, but immediately raises the question of the real governance of the infrastructure that could be built.

What the announcement actually says: €75 billion, 5 gigawatts, and many execution conditions

Reported by TechCrunch under the headline “SoftBank says it will invest up to €75 billion to build French data centers”, the announcement rests on two orders of magnitude that are enough to measure its scale: up to €75 billion in investment and up to 5 gigawatts of capacity. Taken literally, these figures would place the project in an exceptional category for the European market.

As a point of reference, 5 gigawatts represent power equivalent to that of several nuclear reactors or the consumption of a large metropolitan area, depending on uses and periods. In the data center world, people generally speak in megawatts per site or per campus. A large hyperscale campus may target a few dozen to a few hundred megawatts. The most ambitious next-generation projects, driven by AI demand, go higher, but mentioning 5 gigawatts at the scale of a national program remains colossal. In practice, that means not a single site, but potentially a portfolio of distributed campuses, with successive development phases, multi-year grid connections, energy partnerships, and complex territorial trade-offs.

The €75 billion amount must also be interpreted methodically. In digital infrastructure, costs vary greatly depending on the nature of the project: land acquisition, site preparation, electrical connection, substations, buildings, cooling systems, redundancy, network equipment, security, and above all, depending on the case, whether computing hardware is included or not. Yet in AI, the cost of servers and accelerators can quickly exceed that of the buildings themselves. If we are talking about infrastructure intended to host massive GPU clusters, the overall bill can become considerable. The figure put forward by SoftBank is therefore not absurd in absolute terms, but it assumes an ambition that goes far beyond the simple real estate construction of neutral data centers.

Another central question concerns the exact nature of the asset. SoftBank is not a data center operator in the traditional sense of the term, as Equinix, Digital Realty, Data4, CyrusOne, or NTT Global Data Centers may be. Historically, the group acts more as an investor, technology holding company, and architect of strategic bets. The announcement must therefore be read through several hypotheses: direct investment in data center assets, creation of joint ventures, financing of campuses operated by partners, or a broader structure linking infrastructure, cloud, chips, and AI. In the SoftBank universe, ecosystem logic often matters as much as the logic of the isolated asset.

The indirect link with Arm also deserves attention. Even if current AI data centers are largely dominated by Nvidia GPUs, CPU architecture, energy efficiency, interconnects, and workload specialization are becoming increasingly important topics. SoftBank, which controls Arm, has every interest in positioning itself where the next generation of computing infrastructure is taking shape. Supporting, financing, or catalyzing massive capacity in Europe can strengthen its influence across the entire chain, from chips to platforms.

Masayoshi Son’s style must also be taken into account. SoftBank’s founder has often stood out for very long-term, very large-scale statements, sometimes judged excessive at the time they are made. His early bet on Alibaba became legendary. His Vision Fund then embodied a form of hyper-finance in technology, capable of deploying unprecedented sums, but also of fueling fragile valuations. Since the tech market reversal in 2022, SoftBank has sought to regain a more disciplined trajectory, while reaffirming its exposure to AI. In this framework, a €75 billion announcement in France can be read as a message to several audiences: European governments, markets, industrial partners, and American competitors.

The challenge now is materialization. For a project of this size to move beyond the intention stage, precise answers will be needed on at least six fronts: the source of funding, operating partners, the sites under consideration, connection timelines, energy supply contracts, and the exact destination of the capacity. Will it be public cloud, colocation, dedicated AI supercomputers, capacity reserved for certain strategic clients, or a mix of these models? The answer will determine not only the economic viability of the program, but also its real impact on the European ecosystem.

In other words, the announcement is huge, but it is not self-sufficient. It opens up a field of possibilities. And that is precisely why it matters: because it now forces French and European players to think at the gigawatt scale, where public debate was still often framed in terms of support for start-ups, open-source models, or regulation of uses.

A European infrastructure battle against the United States: comparison with hyperscalers, Nvidia, and major AI plans

To measure the strategic significance of the initiative announced by SoftBank, it must be placed back in the global race for AI infrastructure. Since the arrival of ChatGPT at the end of 2022, competition has shifted toward the physical layers of digital technology at remarkable speed. The United States has taken a clear lead, not only thanks to its labs and cloud giants, but above all thanks to its ability to mobilize capital, secure energy, and build quickly.

Microsoft, Google, Amazon, and Meta have all revised their capital spending upward. In their latest fiscal years, these groups have committed or announced tens of billions of dollars in annual capex, a growing share of which is directed toward data centers, networks, and AI accelerators. Meta has repeatedly explained that its infrastructure investments were driven by generative AI. Microsoft has followed up with announcements of new campuses and energy partnerships. Amazon Web Services continues to expand its global footprint. Google is strengthening its cloud and TPU capacity. At this scale, the boundary between cloud infrastructure, AI infrastructure, and energy infrastructure is becoming increasingly blurred.

Nvidia, for its part, is not a data center operator, but the group has become the de facto chief orchestrator of this rush toward compute. Its H100 and then Blackwell GPUs, its DGX systems, its InfiniBand and Ethernet networks, as well as its CUDA software, structure the technical economy of AI. The relative scarcity of advanced chips has pushed operators to secure capacity much further upstream. In Europe, this has reinforced the idea that it is no longer enough to buy cloud on demand: entire infrastructures must be reserved, built, or financed to guarantee access to compute.

Faced with this, Europe is moving forward in a more fragmented way. Announcements exist, sometimes impressive ones, but they are often part of dispersed national or sectoral frameworks. In France, discussions around AI have intensified with the emergence of Mistral AI, with Microsoft’s investments in the country, with initiatives from Scaleway, OVHcloud, and other cloud players, as well as with the highlighting of sovereign computing capacity. In Germany, projects around semiconductors, industrial cloud, and high-performance computing are multiplying. The Nordic countries continue to attract energy-intensive infrastructure thanks to their natural conditions and electricity.

But the scale gap remains striking. That is where SoftBank’s announcement changes the conversation. By mentioning 5 gigawatts, it sets the bar at a level that forces Europe to compare itself no longer with its own standards, but with those of major American complexes or the most ambitious plans in the Gulf. It is no coincidence that the AI battle is now also attracting infrastructure funds, energy companies, specialized developers, and states. The data center has become a hybrid strategic asset: at once real estate, industrial, energy, and geopolitical.

This dynamic recalls other moments of technological shift. At the beginning of the cloud era, the advantage came from software pooling and the ability to industrialize servers. In the current phase, the advantage goes to those who can industrialize entire chains of power: chips, servers, liquid cooling, substations, fiber, orchestration software, electricity contracts, and sometimes dedicated energy production. AI is bringing digital infrastructure into a logic close to that of the great industrial complexes of the 20th century.

France is precisely seeking to position itself at the intersection of these trends. The country can point to its nuclear fleet, its place in European networks, its geographic centrality between the north and south of the continent, and the presence of major public and private clients likely to consume local compute. But competitors are not standing still. Spain is gaining momentum in renewables and interconnections. Italy is seeking to attract more infrastructure. Germany retains industrial depth unique in Europe. Ireland and the Netherlands, despite grid constraints, maintain expertise and a mature ecosystem.

The real lesson from the international comparison is therefore this: if Europe wants to exist in AI beyond regulation and research, it must learn to think in gigawatts, not only in models. SoftBank’s announcement, whether it fully or partially materializes, acts as a revealer of this new scale of competition.

What this implies for France and the French-speaking market: industrial opportunity, energy tensions, and the question of sovereignty

For France, the potential interest of such a program is obvious. An investment of several tens of billions of euros in data centers could generate spillover effects in construction, electrical engineering, networks, cybersecurity, maintenance, cooling, operating software, as well as across the entire subcontracting chain. Direct permanent employment effects are often more limited than political figures suggest, because data centers are capital-intensive infrastructure more than labor-intensive ones. On the other hand, indirect effects can be significant if this capacity then attracts AI clients, labs, software vendors, cloud providers, and computing industry players.

For the French-speaking market, the issue is particularly important. French, Belgian, Luxembourgish, French-speaking Swiss, French-speaking Canadian, and French-speaking African companies are watching the question of compute localization closely. Many want to use generative AI, but still hesitate over data hosting, regulatory compliance, sovereignty guarantees, or network performance. If France were to become a major node of AI capacity in Europe, it could strengthen its role as a regional platform for cloud services, specialized French-language models, sector-specific tools, and infrastructure compliant with European requirements.

This prospect is all the more sensitive because the French language remains underrepresented in part of the major dominant systems, despite recent progress. The availability of local compute does not by itself guarantee the emergence of French-speaking champions, but it can reduce a structural handicap: access to GPUs and inference at competitive cost. For start-ups, labs, or B2B software vendors developing specialized models, proximity to very large-scale infrastructure can make a decisive difference.

There are, however, three major areas of tension.

• Energy tension: 5 additional gigawatts represent an enormous challenge for the grid. Even in a country with a solid nuclear base, connecting such capacity implies massive investment in electricity transmission and distribution, territorial trade-offs, and potentially competition with other industrial uses.
• Land and environmental tension: data centers consume space, mobilize materials, require cooling solutions, and can trigger local opposition. The acceptability of a campus of several hundred megawatts is not automatic, especially if the value created locally appears abstract to nearby residents.
• Sovereignty tension: welcoming foreign capital to build strategic infrastructure is useful, but raises the question of control. Who will operate the sites? Who will have priority over capacity? What guarantees for European players? How will this fit with trusted cloud or digital sovereignty ambitions?

These tensions should not be read as prohibitive obstacles, but as the real parameters of the new AI economy. In France, public debate has often pitted innovation against regulation, or start-ups against large groups. Infrastructure forces another reading, a more material one. A country may have excellent researchers, promising start-ups, and an AI-friendly policy; if it cannot provide electricity, land, and connection timelines, it risks remaining dependent on capacity built elsewhere.

This dependence already has a cost. For many European companies, access to the most advanced AI resources now goes through American clouds. That offers advantages in speed and maturity, but raises questions of market concentration, technological dependence, and value capture. A rise in French capacity could partially rebalance the balance of power, provided it also benefits the local ecosystem and does not merely serve as a simple rear base for outside players.

The role of French operators will therefore be crucial. OVHcloud, Scaleway, Data4, energy companies, local authorities, grid operators, integrators, and major industrial clients will all have an interest in positioning themselves. If SoftBank is truly seeking to deploy at scale, it will need strong local alliances. And those alliances will determine whether the announcement translates into a rise in the French ecosystem or into further outsourcing of value to international decision chains.

Beyond the announcement effect, the next frontier will be execution: energy, timelines, chips, and political trade-offs

The coming months will show whether SoftBank’s announcement reflects a structuring program being assembled or a statement of intent meant to carry weight in industrial and political negotiations. But even in the second hypothesis, it has already produced a tangible effect: it has shifted the center of gravity of the debate. The conversation on AI in Europe can no longer be limited to open-source models, software licenses, or the AI Act. It must now integrate the heavy realities of infrastructure.

The first test will be timelines. Building data centers is not instantaneous; building very large AI campuses is even less so. Between site identification, environmental studies, permits, construction, grid connections, and equipment installation, several years can pass. Yet the AI market evolves on a quarterly rhythm. There is therefore a risk of mismatch between industrial time and competitive time. The winners will be those who know how to compress that timeline without degrading project robustness.

The second test will be the hardware supply chain. Even if electricity and land are available, servers, GPUs, liquid cooling systems, network equipment, and critical components still have to be obtained. Market concentration around a few suppliers, especially for accelerators, remains a vulnerability factor. Any European AI infrastructure strategy runs into this reality: energy autonomy does not mean complete technological autonomy.

The third test will be political. In France as elsewhere in Europe, AI is simultaneously seen as a growth opportunity, a sovereignty issue, and a source of environmental tensions. Governments will have to arbitrate between attractiveness for investors, protection of territories, energy planning, and the expectations of local industrial players. The larger the projects, the more visible these trade-offs become. A 20- or 30-megawatt campus can remain relatively discreet; a program targeting 5 gigawatts becomes a matter of national industrial policy.

The fourth test will be the economic model. Investments in AI data centers rest on the assumption that demand for compute will continue to rise sharply for years. Many indicators point in that direction: the spread of copilots, agents, AI embedded in enterprise software, multimodal inference, robotics, industrial vision, simulation, digital twins. But the market remains young. GPU prices may evolve, architectures may change, software efficiency may improve, and some workloads could shift toward more compact or more specialized models. Operators will therefore have to build infrastructure capable of absorbing technological uncertainty.

For France, the opportunity is historic if it manages to turn its relative energy advantage into a lasting industrial advantage. The country has cards many envy: decarbonized electricity, a state capable of planning, scientific talent, a large domestic market, and a stated ambition in AI. But those cards matter only if they are converted into real assets, connected, financed, and used.

From this perspective, SoftBank’s announcement goes far beyond the framework of a financial promise. It functions as a revealer of the phase European AI is entering. After the battle of models comes that of power infrastructure. After demos and valuations comes the economy of substations, dark fiber, electricity contracts, and very high-density machine rooms. If France wants to be more than a usage market or a talent pool, this is the ground on which it will now have to prove its execution capacity.

What comes next will therefore depend less on the rhetoric of amounts than on the ability to make three time horizons compatible: the competitive urgency imposed by the United States, the long time of energy infrastructure, and European political time, often slower and more fragmented. If SoftBank manages to trigger even a significant part of this program, France could become one of the main AI computing nodes in Europe. If the project runs into the usual bottlenecks, it will at least have had one virtue: showing that the continent’s digital sovereignty will be decided less and less in interfaces and more and more in megawatts.