French artificial intelligence infrastructure developer Sesterce is planning to invest more than €10 billion in a new AI campus at the former Kaipola paper mill in Jämsä, Finland, beginning with a 200 MW first phase and eventually targeting 600 MW of capacity. The company says construction is planned to begin in 2026, approximately 2,000 people could work on the project during construction, and around 300 permanent positions could remain once the campus is fully operating.
What makes the project more significant than another large European data-centre announcement is the industrial site underneath it. Kaipola once produced hundreds of thousands of tonnes of paper annually before UPM-Kymmene Corporation permanently shut the mill after years of declining graphic-paper demand. Sesterce now wants to reuse the site’s electricity infrastructure, water systems, buildings and logistics connections to support one of Europe’s largest proposed AI computing campuses.
That conversion captures a broader change in the economics of artificial intelligence infrastructure. As access to grid capacity, land, water, substations and permitting becomes increasingly difficult, former factories and industrial complexes can carry something more valuable than obsolete buildings: infrastructure that may already have been designed to support energy-intensive operations.
The unanswered question is whether Kaipola’s industrial inheritance can support the extraordinary scale Sesterce is promising. A 600 MW campus would be a major new load even by Finnish standards, while the project’s financing partners remain undisclosed and Sesterce has not yet completed the purchase of the site.
Why does the former Kaipola paper mill give Sesterce a potential infrastructure advantage?
UPM permanently closed the Kaipola paper mill in early 2021 after concluding that deteriorating graphic-paper demand and the site’s cost structure no longer supported economically viable production. The closure affected 448 positions and removed approximately 720,000 tonnes of annual graphic-paper production capacity, including 450,000 tonnes of newsprint and 270,000 tonnes of coated mechanical paper.
UPM subsequently sold the property to Kaipola Green Port Oy in 2021 for redevelopment into industrial and logistics uses. The site covers approximately 80 hectares and contains more than 200,000 square metres of industrial buildings along with electricity, water and transportation infrastructure created for heavy industry.
Those characteristics are precisely what large AI infrastructure developers are increasingly seeking. A greenfield data-centre development can require years of preparation before computing equipment arrives because developers must secure land, build substations, negotiate grid connections, construct water and cooling systems and obtain environmental permits. A brownfield industrial property cannot eliminate those requirements, but it may provide a significant head start.
Sesterce’s strategy is therefore not simply to demolish an old factory and build a server warehouse. The company is attempting to reuse infrastructure created for one energy-intensive industrial era and adapt it to another.
Finland already has a powerful precedent. Google converted a former paper mill in Hamina into a data centre after acquiring the site in 2009, and the location has since become one of the foundations of Google’s Finnish infrastructure footprint. Google’s latest plans involve at least €13 billion of additional digital infrastructure investment across Finland during 2027 and 2028.
Kaipola could represent a second generation of the same industrial conversion strategy, but at a scale increasingly determined by artificial intelligence rather than conventional cloud computing.
How large would a 600MW Sesterce AI campus actually be for Finland?
The 600 MW headline becomes more revealing when compared with Finland’s existing electricity system.
Fingrid reported that total Finnish electricity consumption reached 84.6 terawatt-hours in 2025, while the national consumption peak reached approximately 13.6 GW. On a simple capacity-times-hours calculation, a continuously utilised 200 MW load would correspond to roughly 1.75 TWh of annual electricity consumption. A continuously utilised 600 MW load would correspond to approximately 5.26 TWh.
Those figures should not be treated as forecasts of actual Sesterce consumption because utilisation, commissioning schedules, cooling requirements and the definition of the disclosed MW capacity will determine the real electricity draw. They nevertheless demonstrate the scale involved.
The first 200 MW phase alone would be equivalent to roughly 2.1% of Finland’s entire 2025 electricity consumption if operated continuously at the stated capacity. At 600 MW, the equivalent rises to approximately 6.2%.
Measured against Finland’s 2025 peak electricity demand, 200 MW represents about 1.5%, while 600 MW represents approximately 4.4%.
The comparison becomes even more striking against Finland’s wider data-centre pipeline. Fingrid said in August that data-centre projects with signed electricity connection agreements had planned combined final capacity of nearly 5 GW. Sesterce’s proposed 600 MW campus would equal about 12% of that total, although Sesterce has not disclosed whether Kaipola is already included within Fingrid’s figure.
Sesterce’s longer-term objective of more than 1 GW of AI infrastructure across Finland would be equivalent to roughly one-fifth of the nearly 5 GW data-centre pipeline covered by those connection agreements.
Does an existing industrial grid connection guarantee Sesterce can obtain 600MW?
No, and this is one of the most important distinctions in evaluating the announcement.
Sesterce says Kaipola’s existing infrastructure includes an electricity grid, but the existence of industrial electrical infrastructure does not establish that 600 MW of additional continuous capacity is already available. A former paper mill and a high-density AI campus can have materially different load profiles, redundancy requirements and connection specifications.
Finland itself is entering a much more demanding phase of electricity growth. Fingrid now expects national electricity consumption, currently below 90 TWh annually, to rise to between 106 TWh and 123 TWh by 2030. By 2035, the range widens to between 110 TWh and 162 TWh, reflecting uncertainty around data centres, industrial electrification, hydrogen production, electric heating and transport.
Fingrid has warned that if every project already holding a consumption connection agreement were built at full planned capacity, Finland’s electricity consumption could rise by almost 40% compared with 2025.
That means Kaipola may possess a valuable starting point without possessing a finished 600 MW power solution.
Sesterce has attempted to address the wider system impact by committing to enable one megawatt of new renewable generating capacity in Finland for every megawatt consumed by the campus, with the additional renewable capacity to be delivered within ten years of operations beginning. If fully implemented against a 600 MW campus, that commitment would imply 600 MW of additional renewable generation capacity.
The timing matters. A data centre can begin consuming electricity before the corresponding renewable projects are operating, while wind and solar generation do not necessarily produce electricity during the same hours that a continuously operating AI cluster needs it. Finland will therefore require additional balancing capability and grid investment even if developers successfully contract large volumes of renewable energy.
Why does Finland still attract AI data centres after removing its electricity-tax advantage?
Sesterce’s announcement arrives at an important moment for Finnish data-centre economics.
Until June 30, 2026, qualifying data centres benefited from Finland’s lower electricity tax category II. From July 1, electricity consumed by data centres moved into the general category I as part of a government tax reform.
The difference is substantial. The energy-tax component increased from 0.05 euro cents per kilowatt-hour to 2.24 cents, a rise of 2.19 cents per kilowatt-hour before considering the strategic stockpile fee, which applies to both categories.
Theoretical calculations show why the change matters at Sesterce’s proposed scale. Applying the 2.19-cent difference to a continuously utilised 200 MW load would represent roughly €38 million of additional annual electricity tax compared with the former preferential regime. At 600 MW, the equivalent difference would exceed €115 million annually.
Those numbers are not forecasts because actual taxable electricity consumption will depend on utilisation, commissioning and the final project structure. They do demonstrate that Finland is no longer competing for hyperscale projects primarily through exceptionally low data-centre electricity taxation.
Instead, developers appear to be valuing a different combination of advantages: abundant low-carbon electricity, relatively competitive wholesale prices, political stability, cool climate conditions, strong fibre infrastructure and industrial land with existing power connections.
Statistics Finland reported that 95% of domestic electricity generation in 2024 was fossil-free, including nuclear and renewable generation. Wind power alone supplied approximately one-quarter of domestic electricity production.
For Sesterce, the Kaipola thesis therefore appears to depend more heavily on physical infrastructure and energy availability than on tax arbitrage.
Why is the undisclosed anchor customer so important to the €10bn investment?
Sesterce says the Kaipola project is supported by real demand and committed workload capacity, and that an anchor customer has already been secured.
Chief Executive Officer Youssef El Manssouri told Finnish broadcaster Yle that a contract had been signed and described the customer as one of the world’s two leading AI laboratories. He did not identify the company, and there is currently no public evidence establishing which organisation has committed the workload.
That distinction matters enormously because the credibility of a €10 billion compute project depends heavily on who is contractually obligated to use the infrastructure and under what terms.
A long-duration agreement with a highly creditworthy AI company could make project financing substantially easier. An agreement containing flexible expansion options, termination provisions or conditional capacity commitments could create a different risk profile.
No public information currently establishes the length of the contract, committed revenue, amount of compute reserved, pricing structure or credit protections.
The unidentified customer is therefore simultaneously one of the project’s strongest validation points and one of its largest information gaps.
Can a relatively young company finance a project exceeding €10bn?
The scale difference between Sesterce and the proposed investment deserves particular scrutiny.
Yle reported, based on French commercial registry information, that Sesterce generated approximately €11.95 million of turnover in 2024. The planned Kaipola investment therefore exceeds that historical revenue figure by more than 800 times.
That comparison should not be interpreted as evidence that Sesterce needs to finance €10 billion from its own balance sheet. Large data-centre developments are commonly funded through joint ventures, infrastructure investors, customer commitments, project finance and other forms of external capital.
Sesterce has already demonstrated access to a much larger strategic partner elsewhere in Europe. In May, SoftBank Group Corp. and Sesterce announced a majority SoftBank-owned joint venture selected to develop and operate a 1 GW AI data-centre campus in Bosquel, France. That project forms part of SoftBank Group’s broader commitment to build 5 GW of AI infrastructure capacity in France.
The relationship strengthens Sesterce’s credibility as an operator capable of attracting institutional partners, but it does not establish that SoftBank is financing Kaipola. Sesterce has not disclosed the financial backers for the Finnish project.
The capital structure is therefore one of the most important missing pieces. Investors, contractors and local stakeholders will need to distinguish between the project’s headline capital requirement and the funding that has already been committed.
Why does the unfinished Kaipola property transaction matter?
Sesterce has described the investment decision as made, and its chief executive has said preparatory clearing and interior work have begun. Yet the company does not currently own the former Kaipola mill site.
Kaipola Green Port remains the property owner. Its chairman told Yle that negotiations to sell the entire site to Sesterce were at an advanced stage and that he expected the transaction to succeed, but the sale had not been completed when the project was announced.
This does not invalidate the development. Large infrastructure projects frequently proceed through stages in which land transactions, permits, financing and customer contracts advance simultaneously.
It does, however, demonstrate why the €10 billion figure should be treated as planned investment rather than committed operating infrastructure.
Jämsä’s municipal leadership has taken a similarly cautious position. City officials have welcomed the potential economic significance while stressing that permits and other implementation steps remain outstanding.
That caution is appropriate given the scale. Moving from a signed customer agreement and advanced property negotiations to a functioning 200 MW AI campus will require progress across land ownership, planning, environmental approvals, electricity supply, construction, financing and equipment procurement.
Could brownfield industrial sites reduce the environmental conflict around AI infrastructure?
The timing of the Kaipola announcement makes Sesterce’s brownfield strategy particularly relevant.
Finland recently ordered environmentally significant preparatory work to stop at Google’s proposed Muhos and Kajaani data-centre sites until required environmental impact assessment procedures are completed. Hundreds of hectares had already undergone preparatory changes including tree removal and other ground works.
The Google projects are separate from Sesterce and involve different sites, companies and permitting circumstances. They nevertheless demonstrate that Finland’s enthusiasm for data-centre investment does not remove environmental requirements.
Kaipola offers an alternative development model. Rather than converting undeveloped land, Sesterce proposes reusing an 80-hectare industrial complex whose landscape has already supported heavy manufacturing for decades.
That does not mean environmental impacts disappear. Electricity generation, construction, backup systems, water handling, transport and the eventual disposal or replacement of computing equipment all carry environmental consequences.
Sesterce has nevertheless structured several commitments around those concerns. The company says drinking water will not be used for cooling, with a closed-loop liquid system supplemented by rainwater and recycled water. It has also promised a €10 million community fund once the campus is commissioned and says its environmental commitments will be independently audited and reported annually.
The commercial significance could extend beyond corporate sustainability. If permitting becomes one of the constraints on European AI infrastructure, brownfield locations that reduce land disturbance could become faster and more valuable development assets.
Would the new AI campus replace the economic role of the old Kaipola paper mill?
Not in employment terms alone.
The paper mill closure affected 448 positions. Sesterce expects approximately 300 permanent jobs once the full campus is operating, equivalent to roughly two-thirds of the number of positions affected by the mill closure.
The comparison is imperfect because the occupations, salary structures, required skills and wider supplier ecosystems differ materially. The project could also generate indirect employment through maintenance, security, energy infrastructure and local services.
During construction, the impact would be considerably larger, with Sesterce forecasting around 2,000 jobs and saying it intends to prioritise Finnish contractors and subcontractors.
The bigger transformation lies in what the site produces. Kaipola once converted wood fibre, water, electricity and industrial machinery into physical paper distributed around the world. Sesterce wants the same location to convert electricity, semiconductors and digital networks into artificial-intelligence computing capacity.
That is a remarkably direct illustration of Europe’s industrial transition.
Could former factories become one of Europe’s most valuable AI infrastructure assets?
The Kaipola project points toward a potentially durable investment theme.
AI infrastructure increasingly rewards locations that combine electricity, industrial land, cooling potential, fibre, logistics and community acceptance. Sites built for pulp, paper, chemicals, steel and other energy-intensive industries can possess several of those characteristics before a data-centre developer arrives.
They also carry limitations. Electrical connections may need major upgrades, buildings may not suit high-density computing, remediation can be expensive and industrial sites are not automatically located near the fibre routes or customers required by modern AI systems.
What matters is the time that existing infrastructure can save.
Google’s long-running Hamina operation demonstrates that a Finnish paper mill can successfully become a major data-centre site. Sesterce is now proposing to push that model much further into the AI era, with high-density liquid-cooled computing and a potential 600 MW campus.
The €10 billion headline should therefore be viewed as the beginning of the investment case rather than its conclusion. The anchor customer provides important demand validation, while Finland’s power system and Kaipola’s infrastructure give the concept a credible industrial foundation.
But the most consequential milestones remain ahead: closing the property acquisition, disclosing or securing financing, establishing sufficient grid capacity, progressing permits, beginning the 200 MW first phase and eventually proving that the economics justify a 600 MW expansion.
If those pieces fall into place, Kaipola could become more than another large European data centre. It could demonstrate that some of the most valuable real estate in the AI infrastructure boom is already occupied by factories that the previous industrial era no longer needs.
Discover more from Business-News-Today.com
Subscribe to get the latest posts sent to your email.
