EU’s €20bn AI gigafactory plan faces ‘cathedrals in the desert’ criticism

TL;DR:

• The European Commission’s plan to back four-to-five 100,000-GPU “AI gigafactories” with a €20bn fund is drawing criticism from MEPs and analysts who say the demand case is unproven.
• A formal call for proposals has been delayed twice and is now expected this spring; 76 indicative bids covering 60 sites in 16 countries were submitted in last year’s informal sense-check.
• Greens MEP Sergey Lagodinsky said no one could explain the business case to him; CEPS researcher Nicoleta Kyosovska’s report headline asks whether the gigafactories are “sanctuaries of innovation, or cathedrals in the desert?”.

The Politico investigation lands ahead of the formal launch and surfaces three structural concerns: weak European demand for frontier-LLM-scale compute outside Mistral, deepening reliance on Nvidia GPUs, and the much larger scale of competing US programmes — OpenAI’s $500bn Stargate plan and Anthropic’s $50bn infrastructure commitment.

Context

The gigafactory programme follows the Commission’s 2024 announcement of 19 AI Factories — smaller AI-focused supercomputers spread across 16 countries — and is intended to be four times their size. Mistral, Europe’s leading frontier-model developer, is not waiting for the gigafactory infrastructure; it announced a €1.2bn investment in Swedish data centres in February and raised $830m in March to build a near-Paris facility with around 14,000 GPUs.

Bulgarian MEP Eva Maydell argued the focus on LLM-scale compute misses Europe’s industrial strengths: “There’s an entire world beyond LLMs, where our strategic advances as Europeans lie.” A separate concern raised by 18 MEPs concerns Nvidia’s market dominance: the gigafactories may amplify Europe’s reliance on a single US chip supplier, the inverse of the sovereignty rationale the Commission cites.

For UK readers, the parallels with domestic policy are sharp. UK industry analysts have repeatedly warned that the AI Growth Zones programme and the broader 1.6→6GW data centre target risk similar mistakes if the demand-side strategy is left unspecified — a point Steve Rigby made forcefully in The Times this morning. The UK is constrained by power and grid capacity; the EU, on Politico’s reporting, is constrained by uncertainty about who will use the infrastructure once built. Both governments are committing capital before resolving the matching demand strategy. The Politico piece quotes Cisco’s Jeff Campbell pointing out that EU spend volumes are an order of magnitude smaller than US programmes — a problem the UK shares.

Looking forward

The formal call for gigafactory proposals is expected this spring, and the practical question is whether the Commission narrows its specification toward industrial-AI use cases — battery research, materials, simulation — that play to European strengths, or maintains an LLM-scale general-purpose framing that struggles to attract anchor tenants. UK firms watching the programme will read the eligibility criteria carefully: cross-border participation in any EU-anchored AI compute build is the most concrete UK-EU AI cooperation lever currently on the table.