Summary
- Sesterce announced a planned investment exceeding €10bn in an AI campus at the former Kaipola paper mill in Jämsä, Finland.
- The proposed first phase targets 200MW, with a subsequent expansion to 600MW rather than a completed operating facility.
- Sesterce forecasts about 2,000 construction jobs and 300 operational jobs and has outlined renewable power and cooling commitments.
French AI infrastructure developer Sesterce has outlined plans to invest more than €10bn in a computing campus at a former paper mill in Jämsä, Finland, potentially turning a disused industrial site into a major European AI infrastructure location. The proposal targets 200MW in its first phase and a subsequent expansion to 600MW.
The company announced the scheme on 8 October and identified the former Kaipola mill as its intended development site. It describes the project as an AI campus for supercomputing and related workloads, although the investment total is a planned commitment rather than evidence that the entire amount has already been deployed.
Sesterce estimates approximately 2,000 jobs during construction and 300 once the site is operating at full scale. It says Finnish contractors and subcontractors will be prioritised, but the eventual split of contracts, duration of construction employment and number of local permanent hires remain to be demonstrated.
The announcement places a large proposed computing load on a site with an industrial history. That can offer advantages in land availability, established transport access and the prospect of reusing parts of an existing industrial footprint, although former paper manufacturing infrastructure is not automatically suitable for modern high-density computing.
From announcement to operational delivery
The first phase alone would require substantial electrical capacity and specialised cooling equipment. A 200MW development represents a different infrastructure undertaking from a conventional enterprise server room; reaching 600MW would introduce additional requirements for transmission, substations, grid flexibility and backup resilience.
Sesterce has tied its plans to renewable generation and stated an ambition to enable new renewable capacity in proportion to the campus’s electricity use within ten years of operation. That is a forward-looking commitment and should not be treated as proof that the proposed power generation already exists.
The company also points to a closed-loop liquid cooling approach supplemented by collected rainwater and recycled water. Liquid cooling can help manage densely packed accelerator hardware, but actual water use depends on the full cooling design, external heat rejection, operating temperature and local climate.
Jämsä is positioned within a Finnish technology infrastructure market attracting interest from developers seeking a combination of low-carbon power, technical skills and a stable European jurisdiction. Access to an attractive national electricity mix does not by itself guarantee a grid connection large enough for every proposed campus.
Computing projects of this size must coordinate their commercial construction plans with network operators and authorities overseeing power system development. The timing and cost of network reinforcements can influence whether a campus reaches its intended capacity in one sequence or through staged additions.
The former Kaipola paper mill’s industrial legacy introduces a second set of due-diligence questions. Reuse may reduce greenfield land conversion, yet developers still need to understand ground conditions, remediation liabilities, demolition work and whether existing services can be repurposed economically.
Commercial and technical constraints
An AI campus must also compete for equipment whose procurement cycles may be long, including power transformers, cooling equipment and specialised computing hardware. The pace at which those assets can be supplied will influence the practical value of any headline capacity target.
While data centre investment announcements often stress the prospect of new jobs, construction demand tends to be concentrated during the build period. Operating facilities can employ fewer people than the scale of the capital investment suggests because much of the expenditure goes into electrical, mechanical and computing equipment.
The commercial outcome will depend on whether sufficient customers commit to the proposed capacity and on the terms of financing, electricity supply and tenancy agreements. A preliminary anchor customer or expression of interest can reduce marketing risk, but does not establish completed financing for every expansion phase.
For local authorities, the project brings possible changes to the tax base, contractor demand and workforce requirements. These benefits have to be assessed alongside the effects of a large new electricity consumer and the demands placed on land, roads and local services.
Sesterce’s announcement provides a concrete location and ambitious phased capacity figures, rather than a completed construction or commissioning milestone. Future evidence will include planning decisions, grid agreements, financing commitments and the start of physical construction.
Its ultimate contribution to Europe’s AI computing capacity will depend on delivery over several years, not solely on the €10bn headline. The proposal nevertheless sets out a substantial potential industrial reuse project with considerable infrastructure requirements.
The electricity system will also have to balance a large industrial customer with the needs of households and other employers. Developers can finance dedicated connection infrastructure, but the availability of network capacity depends on decisions made across the wider grid and the timing of generation investments.
Cooling design deserves equal scrutiny because accelerator generations change their power density and acceptable operating temperatures. A liquid cooling specification is not a final energy-efficiency measure; useful comparisons require defined workloads, annual operating conditions and transparent information about how heat is rejected or reused.
