September 23, 2026, (Inside AI) — A public commitment of 200 million euros over five years is not enough to build Europe's proposed AI gigafactories, according to Octave Klaba, CEO of French cloud provider OVHcloud. He estimates operators would need annual sales of 300 million to 400 million euros just to break even on facilities built around fast-depreciating graphics chips.
The EuroHPC program is seeking industry consortia to build and operate up to seven AI gigafactories, large computing facilities for training, fine-tuning and running AI models. Klaba said the public sector would buy 40 million euros of computing services a year, while an operator would spend at least 600 million euros for chips alone. "The numbers don't add up," he said.
Klaba's warning lands as Europe tries to close a widening AI infrastructure gap with the United States and China. The continent hosts only a handful of large-scale AI compute clusters, and most frontier models are trained on American hardware. Brussels has positioned gigafactories as a strategic answer, but Klaba's math suggests the current funding model may not attract the private capital needed to make them viable.
National Demand Is Too Weak
Klaba also criticized the program's focus on national markets, which he said prevented OVHcloud from joining existing French consortia. "Any national strategy is doomed to failure: there are simply not enough customers to make investments of this size profitable. In our view, national demand is too weak and we must operate on a European scale," he said.
Read: Finland Risks Strained Power Supply After Google AI Deal, Opposition Warns
His argument rests on the economics of AI hardware. Graphics chips, or GPUs, lose value quickly as newer generations arrive. Operators must recover their investment within a few years, which requires steady demand from many customers across borders. A single country, even one as large as France, may not generate enough demand to fill a gigafactory.
Klaba estimated the public order would cover only 10% to 15% of the turnover a gigafactory needs. That leaves a large gap for private operators to fill, and it is unclear whether they will step in under current terms.
The CEO's comments echo a broader debate in Europe about how to fund AI infrastructure. Some policymakers favor public-private partnerships, while others argue that the market should decide. Klaba's stance suggests he sees a role for public money but believes the scale must match the ambition.
OVHcloud, which went public in 2021, operates data centers across Europe and has positioned itself as a sovereign cloud alternative to U.S. hyperscalers. Its inability to join French consortia, as Klaba described, highlights the tension between national industrial policies and the cross-border cooperation that large AI projects may require.
EuroHPC has not yet announced the final list of gigafactory sites or consortia. The program's success will depend on whether it can attract operators willing to bet on European AI demand. Klaba's numbers suggest that bet may be harder to justify than officials expect.
Read: TCS Unit to Invest Up to $7.4 Billion in AI Data Center Campus
For now, the gigafactory plan remains a work in progress. Klaba's public critique adds a prominent industry voice to the debate, and it may push EuroHPC to reconsider the size and structure of its public commitment. Whether that leads to a larger envelope or a different approach remains to be seen.
