In January, the European Council threw its weight behind the creation of AI “gigafactories,” widening Europe’s supercomputing programme to include large-scale AI facilities. The move signalled serious intent to invest in advanced AI infrastructure. Yet with most leading AI models still built in the United States, and Europe’s AI challenges extending far beyond funding, it remains unclear whether this will be enough to close the gap.
There are currently 11 supercomputers in the EU managed by the European High-Performance Computing Joint Undertaking (EuroHPC JU). In November 2023, the Commission opened access to these supercomputers to speed up artificial intelligence development. The plan is to allow the EU’s world-class supercomputing resources to be used by European AI start-ups, SMEs and the broader AI community.
The latest move will help build EU competitiveness, but perhaps more importantly, security. Nicodemos Damianou, Cyprus deputy minister of research, innovation and digital policy, explained: “We’ve taken a bold and swift step towards proceeding with establishing AI gigafactories in Europe. AI is one of the most critical technologies of our time, defining our digital future, and investing in the needed infrastructure capacity for AI is essential for boosting Europe’s resilience, competitiveness, and sovereignty.”
All ICT is underpinned by computing power and it’s also a fundamental factor of critical infrastructure. Europe needs autonomy on logistics and helping startups by giving them access to supercomputers funded by the EU is one key way to reduce AI dependencies.
You might be interested
Enough to reduce dependence, but not to replace ChatGPT
Supercomputers allow researchers to carry out experiments they could not do any other way. For example, to do tests such as quantum sensing that analyses data at the atomic level. Supercomputers imaging sub-molecular levels in nature can examine things like cancer and at such a micro level researchers can see whether treatment is having any effect, or see treatment efficacy very, very quickly.
AI is one of the most critical technologies of our time, defining our digital future… – Nicodemos Damianou, Cyprus deputy minister of research, innovation and digital policy
By supporting the further development and scalability of AI models and reducing training time from months or years to a matter of weeks, the EuroHPC JU could ultimately reduce Europe’s dependence on these new technologies from other parts of the world. Latvia’s Tilde has already launched a 30bn parameter open-source model using EuroHPC resources. But can that compete with what’s coming from OpenAI, Anthropic, or Google?
Lost in translation?
Nik Kale, Principal Engineer, Cisco Systems and member of the Coalition for Secure AI (CoSAI) thinks the move is a step in the right direction. “The question of whether infrastructure investments translate into durable capability is exactly the right one to ask,” he said.
“AI gigafactories address a real bottleneck for Europe: access to sovereign computing. That matters, especially given supply-chain constraints and growing concerns about dependency on non-EU technology ecosystems. But computing alone does not create competitive AI capability. Without tight integration with talent, data access, deployment pathways and clear operating rules, gigafactories risk becoming underutilised national assets rather than engines of innovation. The core risk isn’t that Europe builds too little infrastructure, but that it treats infrastructure as the strategy rather than as one layer in a broader system,” explained Kale.
“Europe will benefit from gigafactories if it treats them as a foundation rather than the finished product. They provide essential compute, but governments and enterprises create real value when they turn that capacity into deployable, governable AI systems they can use every day. That is where adoption and sovereignty actually happen,” said Milos Rusic, Co-Founder and CEO at deepset.
Business use of AI is taking off
On a business level, a Eurobarometer survey showed that 20% of EU enterprises with 10 or more employees used AI technologies to conduct their business in 2025. The highest shares of these enterprises in 2025 were in Denmark (42%), Finland (37.8%) and Sweden (35%). At the other end were Romania (5.2%), Poland (8.4%) and Bulgaria (8.5%).
But almost all EU countries have recorded increases in the share of enterprises using AI technologies year on year.
Not just chatbots
The most common use of AI technology by EU enterprises was to analyse written language. This was followed by AI used to generate pictures, videos, sound/audio. But this over-emphasis on generative AI could be a mistake, argues CEPS director of research, Andrea Renda, who thinks Europe needs to stop trying to out-America America or beat China at its own game.
“Unless specific safeguards are in place, betting on the existing model of gigafactories may mean, for the EU, giving up its ambition to adopt a vision of trustworthy AI, which is at once human-centric, sustainable and resilient. Placing faith in the Nvidialed architecture is likely to replicate the current energy- and water-hungry approach to generative AI, and betting on an approach to GenAI that is falling short of the extraordinary achievements it repeatedly promised,” he says in CEPs latest paper.
Unless specific safeguards are in place, betting on the existing model of gigafactories may mean, for the EU, giving up its ambition to adopt a vision of trustworthy AI. – Andrea Renda, CEPS director of research
In other words, the race to build the largest language models with the most parameters is one the EU cannot possibly win on current trajectories. Europe is also caught in the middle of the US–China semiconductor war. Tit for tat export controls on rare earths and chips, chip supply concentration and strategic competition have left the EU dependent on unreliable supply chains. Building gigafactories doesn’t change that on its own.
EU tries to imitate the US — but that doesn’t work
Scott Dylan, Founder, NexaTech Ventures: “My view is that gigafactories are necessary but insufficient. They’re part of the answer. But Europe really needs a more honest conversation about what technological independence actually means in practice when you’re a decade behind in computing, two decades behind in venture capital scale, and entirely dependent on foreign chip supply. That’s not to be defeatist, it’s the starting point for a realistic strategy.”
Like Andrea Renda, Mr Dylan believes the biggest problem is that the EU is attempting to replicate an American model – but without the American-scale capital markets, energy abundance and semiconductor independence. “The EIB partnership announced in December is a step, but €20bn spread across five gigafactories over several years doesn’t match what Microsoft alone is spending on AI infrastructure in 2025,” he said.
Cathedrals in the desert
Jason Hookey, Executive Counsellor at Info-Tech Research Group said that “Although AI gigafactories can help to reduce the capacity gap in Europe, without being part of a broader industrial strategy for making it easier to adopt AI, they will not address the innovation gap.
“The risk is that these gigafactories could turn into “cathedrals in the desert,” meaning that they do not lead to good economic outcomes because of their prestige value alone. In order for gigafactories to have an economic impact, they need to function like utilities that provide low-cost, usable services to small- and medium-sized enterprises (SMEs), rather than just high-end research and development (R&D) labs. “The hub-and-spoke model can improve reach and reduce risk, but organisations will need strong governance to make workloads truly portable across borders,” said Mr Hookey.
“Geopolitics is the current major limiting factor; while gigafactories give operational sovereignty, they do not give technological sovereignty. As long as Europe continues to “build the car but rent the engine” i.e., relies on imported silicon, export controls will continue to govern Europe’s ability to build an AI-based economy,” he concluded.
Lets learn from China
This concern is understood at European Commission level. Executive Vice-President for Tech Sovereignty, Henna Virkkunen, has said that these gigafactories will be purchasing chips “from outside the EU because we don’t have our own production yet.”
Mr Dylan thinks the EU could learn by looking East rather than West: “China’s response to American restrictions has been instructive. Rather than building equivalent infrastructure, Beijing has focused on efficiency gains – DeepSeek’s models trained at a fraction of typical costs, Huawei’s chip clustering approach that compensates for individual chip limitations through scale and cheap energy. The EU might learn something from this pragmatism.”
While Mr Renda gives a clear instruction: “The EU should launch a moonshot on AI to explore more trustworthy solutions, starting with the infrastructure layer and its adjacent software layers. Rather than adhering to the dominant, possibly over-hyped wave of generative AI, the EU should ensure that it continues and revives its commitment to more trustworthy approaches. These include neuromorphic AI and neuro-symbolic AI, which may prove to be winning alternatives in the age of physical AI.”
