Paweł Rożyński: An AI Gigafactory Could Transform Poland, but Three Traps Await

Poland’s bid to host one of Europe’s most powerful computing centres could help the country escape its role as a technology subcontractor. But the AI gigafactory also carries three serious traps: energy constraints, rapidly aging equipment and a lack of customers.

Poland is seeking the location of a major computing facility being developed through the European Union’s EuroHPC JU programme. The stakes go beyond the infrastructure itself. American cloud giants such as Microsoft, AWS and Google currently control much of the AI market. Relying entirely on their infrastructure increases Poland’s dependence on external providers.

A powerful domestic cluster could give Poland greater control over where and under what conditions AI models are trained and data are processed. The benefits could extend beyond home-grown language models such as PLLuM. Large-scale computing capacity could attract capital and support start-ups working in digital medicine, biotechnology, industry and defence. Amid rising geopolitical tensions, the ability to process sensitive data in Europe and develop dual-use technologies also has strategic importance.

The first obstacle is electricity. A facility using tens of thousands of advanced AI accelerators could require hundreds of megawatts, while access to sufficiently powerful grid connections already limits data-centre growth. Energy prices and the source of that power also matter. Poland’s electricity is expensive, and infrastructure of this scale must meet demanding standards for efficiency, water use and environmental impact. The country would therefore need parallel investment in power networks and new energy sources.

The technology itself creates another risk. New generations of AI chips appear quickly, so delays could leave the facility operating with hardware that already trails the latest systems. Yet building the facility alone would not guarantee success. If much of its capacity went unused, the project would fail to meet its promise. The state could reduce that risk by guaranteeing purchases of some computing time, but sustained demand from companies, researchers and public agencies would still be necessary. Large data centres could also face local protests over energy and water consumption or noise, as has happened in the United States and already with planned data centres in Poland.