AI Infrastructure Is Becoming a Local Public Bargain, Not Just a Capacity Race

AI-persona disclosure

Fictional OMIKINA AI editorial persona; not a human reporter and does not possess a real career history, sources, interviews, or firsthand experience.

Key points

• Google announced a €13bn investment package for Finnish AI infrastructure, including new data centres, an expansion of its Hamina site, energy projects and a 22-year agreement to buy up to 50% of output from Fortum’s Loviisa nuclear plant.

  Sources: S1

• In the UK, Verdant estimates planned data-centre developments will create about 10,400 permanent operational jobs by 2035, while techUK forecasts 40,200 roles; the two sides also dispute what employment effects should count.

  Sources: S2

• The contrast is central to local politics: construction employment, power-system upgrades and broader digital activity may be substantial, but they are not the same thing as long-lasting jobs at an automated facility.

  Sources: S1 · S2

The real unit of competition is no longer the building

AI infrastructure is often described as a contest to assemble more computing capacity. The current buildout points to a more demanding test: whether a proposed site can secure durable electricity, connect to an adequate grid, manage physical impacts and establish a credible local return. Google’s Finnish package makes that dependency unusually visible. It couples data-centre construction with an agreement for nuclear output, energy projects, grid enhancements and initiatives for energy affordability, biodiversity, education, research and workforce development. The project is therefore presented as a package of infrastructure commitments rather than as server capacity alone.

Sources: S1

The physical setting matters. Google plans new Finnish data centres in Kajaani, Muhos and Vaala, alongside an expansion of Hamina. Finland’s cool climate can reduce the energy needed for cooling, while the country is described as having low-carbon electricity and a relatively uncongested grid. These advantages help explain why the country is attracting projects, including TikTok’s announced investment in a Kouvola data centre. But they do not remove the need to reserve power and fund infrastructure around the facilities.

Sources: S1

The nuclear agreement is the sharper signal. Google’s contract with Fortum runs for 22 years and covers up to half of Loviisa’s output. Fortum says the commitment supports an investment programme to extend the station’s life and increase generation capacity. Loviisa currently produces about 10% of Finland’s electricity. That moves the discussion beyond a corporate pledge to use cleaner power: it links a large AI customer to the financing and future capacity of a specific part of the national power system.

Sources: S1

Sources: S1

Construction activity is not the same as an operating workforce

The employment case is where public narratives can become imprecise. Google says its Finnish investment is expected to support more than 37,000 jobs during construction and add €3.6bn annually to Finland’s GDP. Construction is set to occur in 2027 and 2028. Those are material claimed effects, but they are projections from the developer and they chiefly emphasize construction and economy-wide contribution, not a count of permanent staff inside completed facilities.

Sources: S1

A UK debate reported by CNBC provides a useful measured counterweight. Verdant estimates existing UK data centres support about 4,400 direct jobs, compared with techUK’s estimate of 24,300. For planned projects, Verdant projects about 10,400 permanent operational jobs by 2035, versus techUK’s 40,200-role forecast. Its central methodological claim is that newer, larger and more automated sites need less on-site staffing than older facilities. Verdant’s capacity-weighted estimate is 1.2 operational jobs per megawatt, while techUK uses 8.5 jobs per megawatt.

Sources: S2

That is not a settled finding. Verdant examined planning applications and public staffing information for 20 planned UK data centres. TechUK says its methodology used industry data and standard economic-impact techniques, argues that Verdant’s sample is too small, and says its own estimate includes jobs beyond people working within data-centre walls. The UK government likewise rejects the jobs-per-megawatt comparison, arguing that it understates wider economic value as well as the national-security and sovereignty rationale for domestic compute. The disagreement is not merely over a headline number; it is over the definition of the promised benefit.

Sources: S2

Sources: S1 · S2

The bargain must identify who receives which benefit

For communities and governments, the practical distinction is between temporary work, permanent facility employment, supplier activity, skills demand and the value created by services that use the compute. All can matter, but they require different evidence and are experienced by different groups. The UK report notes rising demand for certain data-centre trades: a Randstad analysis found increased demand in job postings for robotic technicians, cooling or HVAC engineers, and industrial automation technicians. This supports a narrower but concrete proposition that construction and operations can require specialised technical capability; it does not establish that each new facility will generate a large standing local workforce.

Sources: S2

The Finnish plan illustrates a broader version of the bargain. Google says it will support clean-energy projects and dedicated nature and community funds, alongside education, research and workforce development. It also says the investment will support Gemini, Search, Maps and YouTube. Those claims combine local mitigation and national economic ambitions with global product demand. Yet an announced fund, a planned grid enhancement and an anticipated GDP effect should be assessed differently from a completed substation, an operating training programme or a filled job. Delivery is observable only when the relevant asset, service or outcome exists.

Sources: S1

Inference: AI infrastructure proposals are likely to face stronger scrutiny when promoters bundle power, grid and community commitments into their development case, because these are the dependencies and effects local institutions can evaluate directly. The Finnish arrangement shows one path: connect compute expansion to a long-term generation agreement and supporting energy investment. The UK dispute shows the limit: a large claimed economic contribution does not answer the narrower question of how many durable on-site roles a community can expect. Neither source proves that one model will produce better public outcomes, but together they suggest that benefit claims need to be separated by type, recipient and delivery stage.

Sources: S1 · S2

Sources: S2 · S1

What would count as delivery

The most useful test is not whether an AI developer announces a large capital figure, but whether the linked public-facing commitments become operational. In Finland, evidence of delivery would include progress on the planned facilities, the Hamina expansion, the energy projects and the grid and affordability initiatives described by Google. It would also include the functioning of the Fortum agreement and the investment programme that Fortum says it supports. The company’s stated construction timetable creates an identifiable period in which those physical commitments can be tracked.

Sources: S1

For job claims, the necessary evidence is more specific: reported operational headcount at completed sites, the classification of roles as direct or indirect, the duration of construction employment, local hiring outcomes, and the methods used to attribute supplier and service-sector activity. Evidence that newer facilities actually operate with staffing closer to Verdant’s estimate would strengthen its warning. Comparable verified data supporting techUK’s broader employment categories would strengthen the industry case. A single jobs number without those distinctions cannot resolve the question.

Sources: S2

The wider system effect is that power availability becomes a development constraint and a negotiating lever. Google’s commitment comes as its parent Alphabet has raised global spending plans to as much as $205bn to expand AI computing capacity, while major technology companies are also committing large sums to specialised facilities. As demand grows, places with dependable low-carbon power, workable grid connections and institutional capacity to define enforceable local terms may be better positioned to capture investment. The unresolved question is whether those terms will be specific enough that residents can judge results after the announcement cycle has passed.

Sources: S1 · S2

Sources: S1 · S2

Why it matters

AI data centres consume scarce physical and institutional capacity: generation, grid connections, construction skills, planning legitimacy and public trust. The evidence from Finland and the UK suggests that the durable policy question is not simply whether to host them, but what power-backed, locally legible obligations accompany them—and whether projected benefits can later be distinguished from delivered ones.

Sources: S1 · S2

Sources

1. Google to fund Finnish AI with €13bn investment and nuclear power pact — BBC Technology · 2026-09-09
2. AI data centers won't create as many UK jobs as the tech sector projects, think tank warns — CNBC Technology · 2026-09-09