Summary
- Gdańsk University of Technology’s Tricity Academic Computer Network IT Center has submitted a bid to host the Gaia AI Factory.
- The proposal includes at least 25,000 AI accelerators and sits within Europe’s publicly backed AI infrastructure programme.
- The bid turns AI sovereignty into a power, cooling, security, and operations challenge for Polish infrastructure.
Gdańsk University of Technology has submitted a bid to host the Gaia AI Factory, placing Poland inside Europe’s expanding programme of publicly backed AI infrastructure.
The bid has been submitted by the university’s Tricity Academic Computer Network IT Center. The proposed facility would include at least 25,000 AI accelerators and form part of the wider EuroHPC Joint Undertaking effort to build AI Factories across Europe.
Poland has pledged PLN400 million to the programme. The Gaia proposal would build on the country’s existing supercomputing base while giving Gdańsk a role in Europe’s effort to provide AI compute capacity for research, public-sector use, start-ups, and industry.
A factory in name, a data centre in practice
The AI Factory label can obscure the physical work required. A facility equipped with tens of thousands of accelerators is a dense data centre environment, shaped by electrical distribution, thermal management, high-speed networking, physical security, software orchestration, and operational resilience.
The accelerator count gives the project its headline scale, but the surrounding infrastructure determines whether the system can run. Dense AI clusters require reliable power feeds, robust UPS and backup strategies, careful airflow or liquid cooling design, resilient internal networks, and controls that can keep equipment within operating limits. The more expensive the compute layer becomes, the less tolerance there is for weak facility design.
Cooling may become one of the hardest design choices. Depending on the accelerators selected and the target rack densities, the facility may need liquid cooling or a hybrid approach combining liquid-cooled IT with air-cooled support environments. That would add pipework, coolant distribution, leak detection, maintenance regimes, and commissioning requirements to the technical programme.
Power quality is equally important. AI training and high-performance computing workloads can create concentrated and sustained demand, and public compute infrastructure cannot rely on availability targets that look acceptable only on paper. The electrical system must be designed for uptime, maintainability, and equipment protection from the start.
Public AI capacity moves into the grid debate
The Gaia bid reflects a wider European shift. AI sovereignty is no longer only a question of models, datasets, regulation, or cloud contracts. It is also a question of who owns and operates the buildings, where the power comes from, and whether European organisations can access large compute resources without relying entirely on overseas hyperscale platforms.
Poland has advantages in that contest. It has a sizeable domestic market, strong engineering talent, a growing technology sector, and a Central European position that can support regional demand. Gdańsk adds academic depth and an established computing institution, while the Baltic region has emerging interest in large digital infrastructure projects.
The constraints are just as physical. Large AI infrastructure needs grid capacity, land, cooling options, secure operations, procurement capability, and long-term funding for refresh cycles. The building may be publicly backed, but it will compete for some of the same equipment, contractors, and energy resources as private data centre developments.
Poland’s power mix and grid development will also sit in the background. European AI infrastructure is being built at the same time as sustainability reporting becomes more demanding, and future public compute projects will be judged against energy use, water use, carbon intensity, and local grid effects.
Selection under the EuroHPC process would only start the harder phase. The bid would need to become a deliverable site, with power, cooling, security, networking, procurement, and operations aligned behind a working machine. Europe’s AI Factory programme will be judged not by the number of accelerators promised, but by how many facilities reach dependable service and stay upgradeable as AI hardware changes.
