Finland AI Power Timeline 1977–2026: How Nuclear Energy and Cold Climate Turned Finland Into an AI Data Center Hub

Hamina matters because it is not a plan — it is fifteen years of operating history. Google bought the 60-year-old former Summa paper mill and surrounding land from Stora Enso in 2009 for about €200 million, and opened the data center there in 2011. Instead of building a conventional chiller plant, Google reused the mill’s original industrial intake pipes to draw cold seawater from the Gulf of Finland for cooling — an early, real-world example of adapting existing industrial infrastructure rather than building cooling capacity from scratch.

Google has since invested roughly €3.5 billion in the Hamina region, adding a seventh data-center building in 2024 alongside a further €1 billion sustainability and AI pledge — investment that predates, and is separate from, the 2026 €13 billion announcement. Google also reports 98% carbon-free electricity use across its Finnish operations as of 2023, achieved partly through direct renewable power-purchase agreements layered on top of Finland’s already low-carbon grid.

Hamina also shows how a data center can give something back to a local energy system: an offsite heat-recovery partnership with utility Haminan Energia is projected to help supply roughly 80% of the heating demand of nearby homes and businesses — a genuinely large share, though it depends on the specific size of Hamina’s local district-heating network, not a claim that scales automatically to every new site.

Google’s own framing of the announcement describes “data center and other supporting infrastructure investments and partnerships” across all four locations, chosen after consulting Business Finland and grid operator Fingrid specifically for existing grid infrastructure and carbon-free power access. As of September 2026, the northern sites are announced and land-acquired, not yet built — construction is scheduled across 2027 and 2028.

Megawatts (MW) and gigawatts (GW) measure power — a rate of electricity demand or generating capacity at any given instant, the same units used for power plants and data-center capacity targets. Megawatt-hours, gigawatt-hours and terawatt-hours (MWh/GWh/TWh) measure energy — how much electricity is actually consumed or generated over a period of time. One gigawatt running continuously for one hour equals one gigawatt-hour; running for a full year, it equals roughly 8.76 terawatt-hours. Confusing the two is one of the most common errors in data-center reporting: an announced “1 GW” facility does not automatically consume 8.76 TWh a year, because it may not run at full load continuously from day one.

Loviisa → Finnish electricity grid → long-term PPA → Google’s Finnish operations. That is the real chain — not individual electrons traveling directly from a Loviisa reactor to one specific Google server. A PPA is a commercial and financial contract: Google commits to buy a defined share of Loviisa’s output at agreed terms over the contract’s life, and that electricity is delivered through the same shared Nordic grid everyone else uses. This is standard practice for how large companies procure long-term, low-carbon power — it doesn’t require, and doesn’t create, a dedicated physical line.

Why would Google sign a 22-year nuclear agreement?

• Long-term price and supply certainty — a two-decade contract locks in electricity terms well beyond typical corporate planning horizons, useful for a company whose AI compute demand is expected to keep growing.
• Firm, low-carbon baseload — unlike wind or solar, nuclear output doesn’t depend on weather, which matters for facilities that need to run continuously.
• Supporting plant lifetime extension — Fortum has said the PPA’s predictable revenue helps fund the roughly €1 billion investment program keeping Loviisa operating through 2050.
• Matching a large, growing energy demand — company-stated reasoning; independent analysts have also noted it as a template other AI infrastructure operators may follow for financing nuclear life extensions elsewhere.

Loviisa is a two-unit nuclear power plant on Finland’s south coast, owned and operated by Fortum. Its two Soviet-designed VVER-440 pressurised-water reactors entered commercial operation on 9 May 1977 and 5 January 1981, respectively, with a combined net capacity of 1,014 MW (2 × 507 MW). Loviisa has supplied roughly 10% of Finland’s electricity in recent years. Originally licensed to operate to 2027 (Unit 1) and 2030 (Unit 2), both units’ licenses were extended by Finnish authorities in February 2023 to run through the end of 2050, backed by a roughly €1 billion modernization program.

Olkiluoto 3, operated by TVO, began regular commercial electricity production on 16 April 2023 after roughly 14 years of construction delays — a 1,600 MW European Pressurised Reactor (EPR), Europe’s largest single reactor unit and the first Finnish reactor to enter service since 1980. In 2025, OL3 alone generated 10.38 TWh, while all three Olkiluoto units together supplied roughly a quarter of Finland’s total electricity consumption. Combined with the country’s other nuclear plants, Finland’s nuclear fleet produced around 38–39% of national electricity in recent years, materially improving the country’s energy security and reducing electricity imports.

Olkiluoto 3 was not built to power AI data centers — construction began in 2005, years before generative AI created today’s compute demand. It matters to this story because it demonstrates the scale of low-carbon baseload capacity Finland can bring online, and it helped stabilize the country’s overall electricity supply-demand balance in the years just before Google’s 2026 investment decision.

Generation capacity and local grid connection capacity are different things. Finland’s overall electricity system has run a domestic-production surplus in recent years — domestic generation covered about 93% of national demand in 2025, and the country has been a net electricity exporter at times since Olkiluoto 3 came online. But that national picture doesn’t automatically mean every specific town has enough local transmission capacity to connect a large new industrial-scale load. This is precisely the distinction at the center of the political debate: opposition politicians argue that individual data-center projects, each large enough to consume as much power as a small city, are currently assessed one at a time, making it hard for grid operator Fingrid and the government to see their combined effect on transmission and connection capacity nationally.

Google has said its site selection specifically involved consulting Fingrid for locations with existing grid infrastructure and access to carbon-free electricity — and paired the announcement with its own new generation (629 MW of wind PPAs) and flexibility investment (a 94 MW grid-balancing battery near Kajaani), both of which are aimed at supporting grid stability rather than only drawing from it.

No one can responsibly answer this with a simple yes or no. Multiple real mechanisms push in different directions at once, and as of September 2026 neither side has been settled by data.

Centre Party leader Antti Kaikkonen told Reuters a national permitting system for data-center investments is needed because, currently, “no one is really looking after the overall picture” as individual projects are evaluated separately. Social Democratic lawmaker Niina Malm argued electricity availability should be treated as a broader internal-security issue, so that ordinary households have enough energy at a price they can afford. Prime Minister Petteri Orpo pushed back directly, telling reporters there is enough electricity and that Google is committed to keeping supply sufficient and prices under control. As of this update, no permitting legislation has been introduced — this remains a live political argument, not a resolved policy outcome.

Two real, separate examples show both the promise and the limits. Google’s Hamina heat-recovery partnership with Haminan Energia is projected to help meet roughly 80% of nearby homes’ and businesses’ heating demand. In Kajaani, the CSC-operated LUMI supercomputer’s waste heat supplies about 20% of the city’s district heating through an agreement with Loiste Lämpö. (A separate Microsoft-Fortum district-heating project in the Espoo area is a distinct company and location from Google’s Finnish sites, and should not be conflated with either.) Not all AI data-center heat can simply warm homes: feasibility depends on a district-heating network already existing nearby, the temperature of the recovered heat, distance from the facility, and the underlying economics of connecting the two systems — none of which are guaranteed at every new site.

Not uniformly, and not for every workload. Data-center geography depends on many factors beyond climate: electricity price, grid connection availability, latency to end users, fiber connectivity, land, taxation, permitting speed, water access, cooling design, political stability, workforce and proximity to customers. AI training workloads — large, batch-style jobs less sensitive to split-second latency — can often tolerate a more remote location than latency-sensitive consumer services like search or video calls, which is part of why northern sites suit hyperscale AI training and cloud capacity specifically. Finland, Sweden, Norway and, to a lesser extent, Iceland have each attracted large hyperscale projects for this reason — but Ireland, Germany and Spain continue to attract different classes of data-center investment for their own, different combinations of advantages. “Northern Europe is becoming increasingly attractive for certain large-scale data-center projects” is the accurate framing — not that all AI or all European data centers are relocating north.

Several tracks remain open, none of them decided as of this update: whether Finland adopts a national data-center permitting system as the opposition has demanded; how construction actually proceeds at Muhos, Vaala and Kajaani through 2027–2028; whether the Google-Fortum MoU on new nuclear and additional renewable generation converts into a firm investment; how the 94 MW Kajaani battery and 629 MW of new wind capacity perform once operational in late 2027; and whether Finnish household electricity prices move measurably as the new load comes online. This article will be updated as each of these resolves.