Google's €13B Finland Bet Buys Time, Not New Reactors • The Weighted Average

European AI buyers should add Finland to their capacity diligence, not rewrite deployment budgets after Google announced at least €13 billion of infrastructure investment for 2027–2028. The accompanying nuclear agreement adds a proposed 10MW to an earlier 38MW uprate—a 26.3% increase in planned uprates, not a new reactor or a comparable jump in the plant’s total output.

The valuable power is mostly already there

The September 9 package joins digital infrastructure, energy contracts, and local development. Google names Hamina, Kajaani, Muhos, and Vaala as investment locations. The BBC’s reporting describes three new data centers and an expansion of the existing Hamina site. These are meaningful supply-side commitments. They do not identify a customer’s accelerator allocation, model endpoint, reservation price, or service commencement date.

The nuclear component is unusually concrete. Fortum’s inside-information release describes a 22-year power purchase agreement beginning at smaller capacity in 2028 and reaching half of Loviisa’s capacity during 2030–2049. The contract gives Fortum revenue certainty to continue its lifetime-extension investments through the plant’s operating-license horizon in 2050. A long commitment to buying output helps make maintenance and modernization financeable.

That is not the same transaction as financing a new reactor. Fortum says the plant could not continue beyond 2030 without its lifetime-extension investment program. Google’s agreement supports that program, including work still requiring investment decisions. Preserving an existing source against retirement can materially strengthen a power system even when most of the electricity is not new. The honest description is retention plus an incremental uprate, with a separate option to explore future generation.

Fortum’s operating record gives the distinction scale. Its Loviisa plant page reports 7.9TWh of production and 89.4% overall availability in 2025. Those are historical results, not guaranteed future delivery. They nevertheless show why preserving the asset matters: the economic value is not confined to the marginal capacity increase that makes the newest announcement easy to headline.

For an AI operator, the consequence is a change in the questions asked of a supplier. Does the claimed energy advantage come from keeping existing output available, adding new generation, contracting a share of a plant, or moving demand closer to supply? Each may be useful. None automatically proves that the customer’s inference service will be cheaper, more resilient, or available sooner. The contract needs to connect the infrastructure story to the service being purchased.

This extends the archive’s distinction between infrastructure capital and deliverable AI capacity. A large commitment can justify deeper diligence without justifying an immediate migration. Finland becomes a more credible long-term option; a current deployment remains a current deployment, with its own latency, availability, and cost requirements.

Ten megawatts change the increment, not the category

The cleanest original comparison comes from reading the predecessor announcement. In May 2024, Fortum announced approximately 38MW of additional capacity from low-pressure turbine modernization. That project improved electricity-production efficiency without increasing reactor thermal output. The September 2026 release says Google’s agreement is expected to enable a further 10MW increase, in addition to the already planned 38MW expected to be live in 2028.

Combining those two disclosures gives 48MW of planned uprates. The incremental change is (48 − 38) ÷ 38 × 100 = 26.3%, rounded. Both inputs measure additional generating capacity, so the comparison avoids mixing annual electricity production with a power rating. It also avoids crediting the entire earlier modernization to the new contract. The percentage describes the uprate program, not the output of the whole plant.

Google's deal lifts planned Loviisa uprates by 26.3%

Additional generating capacity, MW · plans, not delivered output

Fortum turbine and PPA disclosures · May 2024 / September 2026

The financing detail is more important than the impressive percentage. Fortum says its approximately €1 billion lifetime-extension program still has €700M of capital expenditure pending investment decisions, alongside approximately 80% of the projects. Those figures measure different things: money and project count. They should not be collapsed into one completion percentage. A signed revenue agreement improves the investment case; individual work packages still have to be approved and executed.

Google’s clean-energy blueprint describes additional layers: a total of 629MW of contracted new-to-grid onshore wind and a 94MW battery near Kajaani expected to operate in late 2027. Wind capacity, nuclear output, and battery discharge power are not interchangeable units of firm supply. The published battery power rating alone does not establish its energy duration, so it cannot support a claim about how long it would carry a data center through a shortage.

Location is another lever. Google says it worked with Fingrid and local partners to place northern and central demand near existing grid infrastructure and carbon-free supply. Its blueprint cites a commissioned study of a hypothetical 1GW demand shift toward the Oulu–Kajaani region, estimating €520 million in consumer savings over twenty years. That is a modeled scenario, not a measured reduction in household bills or the disclosed load of Google’s announced sites. The mechanism is worth investigating without converting the modeled benefit into a guarantee.

The local integration story also predates this investment. Google’s earlier Hamina heat-recovery project describes supplying recovered heat to the community. That is a distinct use of data-center waste heat, not evidence that the new sites already have customers connected to a heating network. Separating the mechanisms keeps a useful infrastructure argument from becoming an untestable bundle of environmental claims.

The best counterargument is that preservation counts

A skeptical reading can go too far. Calling the deal mostly existing power does not make it empty. If modernization keeps substantial low-carbon generation operating when it otherwise would retire, preservation has system value. The relevant comparison is not only new plant versus old plant; it is a grid with the retained asset versus one without it. Fortum’s disclosure supports taking that counterfactual seriously, while leaving the precise customer benefit to be demonstrated.

Nor is new nuclear absent from the partnership. The companies signed a memorandum to explore additional generation and flexibility, including potential new reactors at Loviisa. But exploration and an executed generation project are different evidence states. Fortum’s March 2025 feasibility conclusion placed possible new Nordic nuclear supply in the second half of the 2030s at the earliest, conditional on suitable market and regulatory conditions. The new announcement does not supply a replacement construction timetable.

The strongest bullish case is therefore not immediate abundance. It is that a large buyer is giving an experienced operator a sufficiently durable demand signal to undertake long-lived work. The arrangement could also help develop future projects. That is a credible market-structure argument, but it remains weaker as a claim about this quarter’s GPU rental prices. A procurement team should preserve the distinction even when a vendor’s sales presentation does not.

Execution can break the thesis in ordinary ways. The remaining investment decisions may take longer, power upgrades may arrive on different schedules, or demand may outpace the combination of retained generation and new supply. Those are risks to test, not events reported as having occurred. A useful diligence pack would show which milestones are contracted, which depend on third parties, and which remain aspirations. Ask who bears the consequences if the customer-facing start date slips.

Even successful infrastructure delivery need not translate into the customer’s preferred service. Google Cloud’s regional product documentation explicitly distinguishes baseline services from additional products that arrive later and evolve with demand. The geography of a data-center announcement is not a complete product catalog. Verify the required service, hardware configuration, data location, and quota rather than assuming that a regional footprint makes every capability locally obtainable.

The same distinction appeared in yesterday’s analysis of sovereign AI deployment boundaries: locality must solve a real operating requirement. A Finnish site may be valuable for a particular workload, but the jurisdictional label alone does not settle latency, support, portability, or the path taken by every related service. Evidence that would strengthen the case is a documented regional configuration with usable capacity, clear terms, and a measured advantage for the workload being moved.

Negotiate the bridge from electrons to service

The rational response is staged commitment. Put Finland on the supplier-comparison sheet for European work, especially where location and long-term energy resilience matter. Keep a current provider until a tested alternative clears the actual operating requirements. The new investment improves the reason to ask for a proposal; it does not itself supply the proposal’s price or prove that migration work pays back.

Google’s own investment summary also emphasizes a broader local ecosystem, including suppliers, training, and community funding. That can matter to sustained operation, but it should not be represented as free implementation capacity for a cloud customer. Ask who will perform the customer’s migration, what support is included, and which expenses remain outside the infrastructure commitment.

The price comparison needs a complete boundary. Request a quote for the required compute service, storage, network movement, support, and transition period. Then compare the delivered configuration with the existing one. Neither Google nor Fortum publishes a customer inference tariff in these announcements. Dividing €13B by an unrelated power figure would manufacture a cost-per-megawatt statistic whose numerator and denominator do not describe the same project scope. The absence of that shortcut is a feature of the analysis, not a missing calculation.

Today’s Nvidia sports-AI brief shows how training-framework choices affect a compute budget, while Apple’s camera entitlement ladder shows how a bundle can limit usable AI access. Both reinforce the relevant question here: which exact unit of service does the buyer receive? Capital, energy, hardware, software access, and completed work need separate entries in the ledger.

Before switching, require a small workload trial on the proposed configuration, a documented failure path, and a migration estimate that includes rollback. Keep the evidence from that trial independent of the supplier’s national investment narrative. If the service meets the requirement at a better total cost, the energy partnership is useful supporting context. If it does not, the size of the announcement should not rescue the business case.

Cloud procurement leads: request regional availability and committed-capacity terms now; switch only after a workload-level comparison, with migration and parallel-running costs included.
Infrastructure and energy buyers: distinguish retained output, contracted plant share, the earlier 38MW uprate, the new 10MW increment, and prospective generation. Ask for delivery milestones rather than summing unlike capacity claims.
Engineering and finance teams: preserve the incumbent plan until price, quota, locality, and recovery behavior are verified. Watch remaining Loviisa investment decisions, the battery’s delivery, and customer-facing service availability for evidence that improves the verdict.

The transaction deserves attention precisely because it is less magical than an AI power headline. A long-term customer is helping an existing plant remain useful while pursuing additional supply. That can be commercially and environmentally important without being new-reactor construction. For this quarter’s buyer, the right move is disciplined diligence: negotiate the bridge from a credible power commitment to a service the team can actually run.

The valuable power is mostly already there

Ten megawatts change the increment, not the category

The best counterargument is that preservation counts

Negotiate the bridge from electrons to service

Sources