Show Notes
Gerard and Laurent welcome Michel Boutouil, co-founder and CEO of Polarise, a leading European AI infrastructure provider and NVIDIA Cloud Partner based in Berlin. After discussing about what happens outside of a datacenter, it is time to dive inside one.
Polarise is one of the few genuinely European NeoCloud companies — essentially a European counterpart to CoreWeave — specializing in GPU infrastructure for AI inference. Through its partnership with NVIDIA, Polarise designs its datacenters around the GPU rack itself, using liquid cooling from the outset rather than starting with a traditional real estate-first approach. The company has already developed AI factories in Germany, Norway and the UK.
In the conversation, we explore the growing commoditization of large language models and why the real long-term value may lie in AI factories — facilities that are fundamentally different from conventional datacenters. Given Europe’s notoriously long grid-connection timelines, Polarise focuses on refurbishing brownfield sites with under 50MW of grid access instead of pursuing massive gigawatt-scale campuses. It’s a pragmatic “pod” strategy: adapt to the grid’s constraints rather than try to reshape the entire energy system.
We also tackle the thorny issue of digital sovereignty. With the U.S. CLOUD Act allowing U.S. authorities access to data managed by American tech companies, it is fair to ask what hyperscalers are doing with European data — and whether Europe needs its own sovereign AI infrastructure. Polarise has secured €1 billion in backing from Swiss investor SWI Stoneweg Icona, but even that is modest compared with the hyperscalers’ spending power. For comparison, SpaceX has reportedly invested around $40 billion in Colossus 1 and 2 alone.
So, what does the future of the European AI ecosystem look like? Michel’s answer is clear: Europe should not try to outspend China or the United States head-on. Instead, it should play to its strengths — smart execution, agility, flexibility, and the ability to learn quickly from the mistakes being made elsewhere.
“Today’s show is supported by the BMW Foundation Herbert Quandt. The BMW Foundation unites leaders across sectors to develop solutions that foster an innovative economy and a future-proof society. A key focus is "Energy Transition & Climate Change," where the Foundation drives "International collaboration to accelerate the energy transition." With rising energy demands from AI and data centres, new partnerships, effective collaboration, and the exchange of science-based solutions and strategies are essential.”
More from Redefining Energy
Last year, the nuclear industry experienced something close to a gold rush. SMRs — Small Modular Reactors — became the hottest story in energy. Private capital poured in, valuations exploded, and the industry produced an impressive collection of truly bombastic promises. The backdrop is undeniably powerful: electrification is accelerating, geopolitics are reshaping energy markets, and the world needs reliable, 24/7 clean power. Nuclear has rarely enjoyed such formidable tailwinds.
The current datacenter buildout is the biggest GDP bet in America since the railroads 150 years ago. We have never seen a frenzy quite like this. The system is running on overdrive—and the cracks are starting to show.
The global market for Virtual Power Plants is assessed around 100GW (+/- 20GW). It is a 7bnUSD expected to grow at a CAGR above 20% in the next 5 years to reach 40bnUSD within a decade. VPP are a digitally coordinated flexible network of distributed resources: cheaper than central stations. They manage behind the meter assets like Solar, batteries, HVAC, EV…
Joseph J. Romm (1960–2026) was an MIT-trained physicist who became one of the most prominent voices in the U.S. climate movement after serving as Acting Assistant Secretary for Energy Efficiency and Renewable Energy at the U.S. Department of Energy in the 1990s.
In our series: How AI is revolutionising Energy, Laurent and Gerard dive deep into the intricate corner of Energy Management system for large industrial clusters.