The rise of Gigafactories
The European Union (EU) has recently escalated its commitment to artificial intelligence (AI) competitiveness by announcing a comprehensive plan to fund AI Gigafactories (AIGFs). This initiative marks a strategic shift away from a primary focus on regulation towards boosting industrial capacity and technological leadership. The Commission, acknowledging Europe’s competitiveness delay in the domain of AI, particularly concerning compute infrastructure, is investing heavily in building these large-scale facilities across the continent. The core strategy, embedded within the broader ‘AI Continent Action Plan,’ is to provide a “strong response to American and Chinese strengths” in AI, enhancing the EU’s technological autonomy, resilience, competitiveness, and security. This monumental effort, backed by a dedicated €20 billion fund under the InvestAI initiative, aims to establish up to five AIGFs, each projected to host at least 100,000 advanced chips.
The Strategic Rationale and Strength of EU Gigafactories
The establishment of AI Gigafactories is driven by the imperative to supply European AI communities with the massive computational resources necessary to compete globally in the development of frontier AI models. The AIGF model is explicitly designed to train next-generation models, requiring compute scales far beyond what existing infrastructure offers.
A key strength lies in the strategic scope and scale of the investment. AIGFs, defined as large-scale facilities combining high-performance computing (HPC), energy-efficient data centres, and AI-driven automation, are expected to provide world-class AI computing infrastructure. With each facility intended to host a minimum of 100,000 high-end accelerators, the initiative brings a “respectable scale” to the EU. Crucially, this move is designed to remove the structural barrier that currently forces European research organisations and start-ups to rely heavily on foreign cloud providers for training state-of-the-art models.
The financial model supporting the AIGFs represents a significant innovation. They rely on Public–Private Partnerships (PPP), where the EU and member states cover up to 35% of capital expenditure (CapEx), mobilizing mostly private funds for the remainder, including all operational expenditure. This hybrid model is expected to accelerate deployment, attract necessary capital, and ensure that infrastructure operates partly on commercial principles.
Moreover, the Gigafactories are positioned to integrate with and complement the network of smaller AI Factories (which typically house up to 25,000 chips each). While the AI Factories prioritize flexible access for SMEs, focusing on training and finetuning, the AIGFs target massive computing power for developing more complex frontier models and large-scale inference. This layered ecosystem is intended to drive breakthroughs in strategic sectors, including medicine, cleantech, space, health, manufacturing, and climate, providing secure infrastructure for trustworthy AI. Indeed, the Commission has cited the “overwhelming response” of 76 expressions of interest for the AIGF projects as a “powerful signal” of market demand.
Critical Challenges and Expert Concerns
Despite an ambitious vision, the AIGF plan has met considerable criticism from technical experts and analysts, who raise serious concerns regarding feasibility, strategy, and technological focus.
1. Scale and Speed Deficit: While the AIGFs are large for the EU, the planned total capacity (approximately 400,000 advanced chips) is dwarfed by investment levels seen in the US private sector. For instance, experts note that Meta planned to have 1.3 million GPUs operating by the end of 2025 alone, demonstrating that the European investment, while significant, is unlikely to create a quick “catching up” effect against global leaders.
2. Overreliance on Generative AI and Vendor Lock-in: A primary critique is that the EU is “adhering to the dominant, possibly over-hyped wave of generative AI”. This focus risks making the infrastructure inflexible, especially as the current generation of large language models (LLMs) exhibits well-publicised weaknesses like hallucination, and scaling them up is yielding diminishing performance returns. Experts contend that rather than solely emulating the US approach, the EU should launch moonshots on alternative, more trustworthy AI solutions, such as neuromorphic AI or neuro-symbolic AI, which promise higher energy efficiency and adherence to EU values.
Relatedly, the overwhelming reliance on Nvidia for graphics processing units (GPUs) introduces a significant vulnerability. Nvidia’s proprietary software stack creates interoperability problems and the risk of vendor lock-in, threatening the EU’s aspiration to build a truly sovereign ‘EuroStack’.
3. Demand Uncertainty and Competition: The stated primary goal—to train and deploy frontier AI models—is deemed “never very realistic” in the short term, given that Europe currently hosts very few leading AI labs capable of generating the necessary anchor demand. If AIGFs cannot secure anchor customers, they must pivot to a multi-client model focused on aggregating low-to-moderate industrial and academic workloads. This scenario pits the AIGFs directly against agile, often cheaper, neocloud providers, whose services already offer high flexibility, raising fears of idle capacity.
4. Location and Geographic Dispersion: Finally, critiques point out that the political motivation for locating AI factories (and likely AIGFs) across multiple member states risks dispersing resources in areas not optimally suited for AI ecosystems. Analysis shows that deployment often occurs far from established AI “hubs of excellence”. Crucially, many locations lack the favourable energy conditions of Nordic countries (Sweden and Finland), which offer low-carbon energy and high-efficiency cooling—factors necessary for competitiveness and achieving Green Deal objectives.
MINERVA: Supporting the Ecosystem and Bridging Gaps
Against the backdrop of the AIGF strategy and its inherent weaknesses, a project like MINERVA, acting as the European Support Centre for Scalable AI Research and Deployment, can play a vital role in underpinning the EU’s infrastructure investment and filling critical ecosystem gaps.
MINERVA is designed to bridge the existing gap in HPC utilisation by AI/ML communities by accelerating knowledge transfer. A major challenge facing the EU is the inability to attract and retain world-class talent to operate and exploit these massive infrastructures. MINERVA directly addresses this through Advanced Training and Capacity Building activities, providing specialized expertise on HPC architectures, software stacks, and scaling workloads. By offering a “Train-the-Trainers” approach targeting entities like National Competence Centres (NCCs) and European Digital Innovation Hubs (EDIHs), MINERVA ensures a cascading effect in skill dissemination.
Strategically, MINERVA focuses on the open-source AI domain, placing special emphasis on the research and development of open-source foundation models. This directly supports the alternative AI pathway promoted by critics, offering a European vision of AI that prioritizes transparency and reuse over proprietary solutions. Furthermore, MINERVA integrates expert support on ethical and responsible AI regulations, assisting users in addressing compliance issues—a core requirement for Europe’s trustworthy AI agenda that is generally absent from a pure compute focus.
Finally, MINERVA is essential for operationalizing the multi-client user setup deemed most feasible for AIGFs. It increases SME and start-up uptake of HPC resources, aggregating the precise type of low-to-moderate demand that AIGFs will rely upon if anchor customers fail to materialize. By creating a Community Hub linking research, industry, and HPC specialists, and establishing a single entry point for multi-level technical support (L1, L2, L3), MINERVA helps overcome the fragmentation and interoperability challenges inherent in the distributed EuroHPC infrastructure landscape. By focusing on competence, ethical alignment, and broad industrial adoption, MINERVA ensures that the massive hardware investment embodied in the Gigafactories does not become a mere monument, but a functional, competitive ecosystem.
References
- EU Plans for AI (Giga) Factories: Sanctuaries of Innovation, or Cathedrals in the Desert?: https://cdn.ceps.eu/2025/11/251027-Sanctuaries-or-Cathedrals.pdf
- Built for Purpose? Demand-Led Scenarios for Europe’s AI Gigafactories: https://www.interface-eu.org/publications/ai-gigafactories
- What if generative AI is reaching its limits?: https://www.europarl.europa.eu/RegData/etudes/ATAG/2025/774701/EPRS_ATA(2025)774701_EN.p
No responses yet