Poland has put Gdańsk forward as a possible home for an AI computing facility containing at least 25,000 accelerators, turning Europe’s ambition for sovereign compute into a decision about power, construction, procurement, and access.

The bid has been submitted by the Tricity Academic Computer Network IT Centre, known as CI TASK, which operates within Gdańsk University of Technology. CI TASK already provides high performance computing, cloud, data, and network services to universities and public institutions.

Public information about the proposal remains limited, although the planned system would rank among Europe’s largest technology installations if selected and completed. Poland has pledged PLN400 million to the wider European AI infrastructure programme and has encouraged domestic companies to take part in the eventual construction tender.

The proposed machine belongs to a broader European attempt to provide computing capacity that researchers, startups, manufacturers, and public bodies could not readily procure alone. Yet an accelerator count reveals only the theoretical size of the installation, rather than the amount of useful work it will produce.

Sovereign compute reaches the procurement stage

The EU’s AI Factories programme combines optimised supercomputers with data access, technical expertise, model development, and support services. Larger gigafactory proposals are intended to add another tier as advanced research and model training consume more processing capacity.

Publicly supported infrastructure can influence where sensitive workloads run and which organisations obtain access, especially when commercial cloud services create concerns around cost, confidentiality, procurement, or legal jurisdiction. Ownership of a building does not, however, make the underlying technology European when accelerators, networking equipment, and software frameworks remain heavily dependent on overseas suppliers.

A facility in Gdańsk could still increase Poland’s practical control over scheduling, data governance, operating conditions, and research priorities. Universities, healthcare bodies, manufacturers, and government agencies may be able to undertake work that would otherwise remain too expensive or difficult to procure.

CI TASK offers an established institutional base, having operated computing, cloud, and academic network infrastructure over several decades. Moving from existing systems to a machine containing tens of thousands of accelerators would nevertheless introduce a different order of engineering complexity.

High density clusters require substantial electricity, specialist cooling, high capacity interconnects, security, equipment maintenance, and continuous workload scheduling. A processor that remains idle because data, networking, or user support is unavailable continues to incur depreciation without producing economic or scientific value.

Access will decide who benefits

Construction could create demand for electrical equipment, fibre, cooling, software, integration, and technical services, while a long term operating programme would support specialist engineering roles. The proportion retained within Polish and European supply chains will determine whether the facility builds industrial capability or primarily finances imported machinery.

Smaller organisations face barriers that cannot be solved through compute credits alone. A company may receive processor time but lack clean data, experienced engineers, security controls, model evaluation, or a defined business process in which the system can be used.

AI factories are intended to supply some of that surrounding support, which makes the design of their service operation as important as the hardware tender. Industrial users working to commercial deadlines need predictable capacity, clear intellectual property terms, responsive assistance, and procurement routes that do not take months to navigate.

Allocation rules will create difficult choices when demand exceeds supply. Academic research, strategic industrial projects, startups, and public services may each have a credible claim, while the facility needs enough paid or funded work to cover electricity, staff, maintenance, and equipment replacement.

Energy availability could constrain the proposal before allocation becomes an issue. A system of this size would join a wider pipeline of Polish and regional datacentre projects, increasing competition for grid capacity and raising questions about who funds new generation and transmission.

Hardware also ages quickly. Public procurement must account for phased replacement, software compatibility, cooling changes, and a commercial model that can finance upgrades without waiting for another exceptional capital programme.

Europe’s sovereign AI policy will ultimately be judged by the work performed on its machines rather than the number announced at launch. Gdańsk has offered a credible institutional location, but lasting value will depend on whether the facility is kept busy, connected to domestic industry, and accessible beyond the organisations already equipped to use expensive computing infrastructure.