Europe is pouring billions into AI development and infrastructure. GPU (graphics processing units) access is expanding rapidly through cloud platforms and GPU-as-a-service (GPUaaS) providers, becoming a key enabler of AI development and deployment.
The underlying assumption is straightforward: scale compute, and you scale capability.
Yet, despite the efforts made by EU Member States, from sovereign cloud initiatives to federated data infrastructure, the European AI landscape remains constrained by a critical bottleneck: dependence on GPUs largely designed by non-European players such as NVIDIA and manufactured by Asian foundries, primarily Taiwan’s TSMC, over which Europe has no chance of chip independence at either layer in the short term.
This has direct implications for European technological sovereignty.
The computer boom
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
The semiconductor industry is undergoing a structural boom, driven by accelerating demand for AI workloads, including agentic systems, robotics, and automated operations.
According to Deloitte, the global semiconductor market is projected to reach approximately $975 billion in annual sales in 2026, with generative AI chips alone expected to contribute around $500 billion in revenue.
GPUs, originally designed for rendering graphics, have become the backbone of modern AI systems due to their ability to process large-scale computations in parallel.
This makes them essential for training and deploying LLM and Agentic AI systems. GPUaaS allows organisations to rent access to this compute by the minute or hour, significantly lowering the cost and complexity of ownership.
However, the GPUaaS ecosystem is dominated by US-based hyperscalers and semiconductor providers. Companies such as Amazon, Google, and Microsoft control a significant share of global cloud infrastructure.
This concentration has elevated the importance of semiconductors as a strategic asset. Governments are increasingly balancing export controls with domestic capacity-building efforts, as AI technology ownership and chip access have become critical to national security, supply chain resilience, and technological sovereignty.
Europe is scaling without ownership
In response, the European Commission has introduced initiatives such as the AI Continent Action Plan, aimed at strengthening Europe’s AI capabilities across large-scale AI data and computing infrastructure, access to large and high-quality data, developing and deploying AI in strategic sectors, strengthening AI skills and talent, and regulations.
This includes €20 billion in funding for up to five AI gigafactories, as part of a broader €200 billion investment ambition under InvestAI by public and private actors. Earlier sovereign-infrastructure efforts include the EU’s €180 million sovereign cloud contracts, awarded to providers such as Scaleway, StackIT, and Post Telecom.
The push is already visible in infrastructure. As of 2026, Europe operates 14 supercomputers and 19 AI Factories under the EuroHPC JU (Joint Undertaking), backed by roughly €10 billion of combined Commission and Member State funding over 2021–2027.
European providers, such as French cloud leader OVHcloud, and initiatives like Deutsche Telekom and T-Systems, which in early 2026 launched T Cloud Public alongside their Munich-based Industrial AI Cloud running 10,000 NVIDIA Blackwell GPUs, are moving toward sovereign cloud and AI infrastructure.
Yet at the base of these efforts lies a persistent constraint: dependence on external chip suppliers such as Nvidia and AMD.
AI compute remains highly concentrated in a small number of dominant players. NVIDIA is currently the worldwide chip provider leader for AI GPUs, forming the backbone of most large-scale AI systems.
The company holds roughly 85% of the AI GPU segment, a figure analysts expect to drift toward 75% by 2026 as AMD and custom silicon scale, while together with firms such as Advanced Micro Devices (AMD),
Broadcom and Qualcomm are the industry at the performance-critical layer of AI infrastructure.
At the infrastructure level, US cloud providers dominate GPUaaS in Europe: AWS, Microsoft Azure, and Google Cloud collectively account for the bulk of European cloud capacity, with hyperscalers controlling roughly 70% of European cloud infrastructure revenue.
Efforts such as the Chips JU, the implementation body of the EU Chips Act, aim to strengthen Europe’s semiconductor ecosystem with coordinated efforts across member states and private actors. However, semiconductor manufacturing remains structurally difficult to localise due to extreme capital intensity and supply chain concentration.
As a result, GPU and advanced chip dependency is unlikely to materially decrease in the short to medium term, with Europe still reliant on Nvidia for roughly 85% of its AI GPUs.
GPUaaS expands access to compute. It does not change who controls it.
Economic capture
Beyond technical dependency, economic power follows control of computing distribution. Hyperscalers capture disproportionate value by intermediating access to scarce GPU resources, leaving European users exposed to externally set pricing, capacity allocation, and margin structures.
The current CapEx (capital expenditure) for just Google, Amazon, Meta, and Microsoft is about $725 billion on AI infrastructure in 2026, up 77% from $410 billion in 2025; this number exceeds the GDP of many European countries, representing levels that Europe cannot realistically match in AI investment.
While this enables access for European firms, it reinforces structural dependency.
By contrast, European spending on sovereign cloud infrastructure is forecast to reach roughly $12.6 billion (around €12 billion) in 2026, according to Gartner, an 83% jump from 2025, but still an order of magnitude below US hyperscaler CapEx.
Among multiple criticisms of the European AI Continent Plan, calls have been made to reallocate those investments to already established companies that can compete against US investment capacity.
French champion Mistral, for instance, has confirmed a €1 billion CapEx plan for 2026, and has separately committed to a €1.2 billion Swedish data centre and a Paris-area facility powered by 13,800 NVIDIA chips.
A group of 18 European Parliament lawmakers raised questions about the development of the AI gigafactory, warning that it risks deepening Europe’s dependence on US GPU chips and arguing that the data centre space is “concentrated and dominated by a single supplier.”
This dependency extends to jurisdiction. Access to cloud-based GPU infrastructure is ultimately governed by legal and regulatory frameworks outside Europe’s control, particularly in the United States.
Even when data is hosted locally, control over infrastructure providers introduces geopolitical risk, where access to critical AI resources can be influenced by policy decisions beyond Europe’s control.
The illusion of scarcity
The constraint is not purely one of supply. Industry studies suggest GPU usage in Kubernetes clusters averages just 5%, and 71% of enterprises cite GPU utilisation inefficiency as a major barrier to scaling AI workloads.
Many companies rely on Kubernetes as the orchestration platform allocating compute across servers; the result is GPU infrastructure that remains structurally underutilised due to inefficient allocation, fragmented demand, and suboptimal workload distribution, eroding return on investment.
In practice, organisations often overprovision compute to secure access or manage several workflows, reinforcing the perception of scarcity while reducing overall system efficiency.
For Europe, this creates an additional risk: even with increased control over infrastructure, inefficient allocation and monetisation could turn large-scale investments into underperforming assets. GPUaaS improves access, but does not inherently optimise how compute is used.
Where Europe can still win
It would be reductive to dismiss Europe’s infrastructure investments as purely illusory.
Building sovereign compute capacity creates tangible optionality: it develops domestic engineering talent, establishes regulatory leverage over data flows, enables the development of European models, and creates the institutional capacity to absorb sovereign hardware once European chip alternatives mature.
The EuroHPC JU’s supercomputer network, for instance, already enables research communities and smaller member states that previously did not have meaningful access to frontier computing.
The debate around AI infrastructure in Europe remains focused on access and capacity. The European Commission is allocating effort to building its own AI stack, yet the more fundamental issue lies in control over allocation.
As long as compute resources are distributed through external platforms, Europe’s ability to shape how AI is deployed remains limited, regardless of how much infrastructure capacity is built.
In this context, GPUaaS should be understood as a short-term enabler rather than a structural solution. Expanding access to compute accelerates adoption, but does not shift control over the underlying system.
For Europe, the path toward a sovereign AI ecosystem lies in acknowledging its structural constraints and directing resources toward areas where control is realistically achievable, rather than attempting to compete head-on with an industry that holds significantly greater investment capacity and technological maturity.
Long-term competitiveness will depend less on replicating the full infrastructure stack and more on securing strategic positions of leverage, whether through orchestration layers such as the emerging capabilities of Mistral AI, specialised infrastructure like ASML, the world’s only manufacturer of advanced lithography systems critical to chip production, or through regulatory frameworks that actively shape market dynamics.
In parallel, exploring more cost-efficient models, such as neocloud providers like Nscale and Nebius, may offer a pragmatic way to build on Europe’s current position.
Access is not sovereignty
Europe is not falling behind in the AI race because it lacks access to computing, but because it does not control it. GPUaaS lowers the barrier to entry and accelerates AI development and adoption, but it does not change the structure of power.
While both the EU and its member states are moving towards a more competitive posture, Europe’s current trajectory risks deepening dependency rather than reducing it.
This is not to say the EU’s efforts are without value, but capacity building and sovereignty are not the same thing.
The real question is whether Europe can convert its current position of managed dependence into genuine strategic autonomy, rather than simply locking in reliance on external suppliers.
Sovereignty will not arrive with the next gigafactory, but it might when Europe’s own chips finally do.
And compute dependency isn’t the only current issue for Europe; it is increasingly becoming a question of data sovereignty and political autonomy.
