Google Backs $15B Nordic Expansion with Nuclear and Grid Storage Deals

Google announced a massive €13 billion ($15 billion) capital expenditure program across Finland for 2027–2028, funding three new data center campuses in Kajaani, Muhos, and Vaala alongside an expansion of its flagship Hamina facility. Crucially, the buildout is tied directly to new energy infrastructure: a 22-year power purchase agreement (PPA) with Fortum to support life extension for the Loviisa nuclear power plant, 629 MW in onshore wind commitments, and a contracted 94 MW battery storage facility designed to stabilize the regional grid in Kajaani. For DevOps leaders and enterprise cloud architects, this development reflects a decisive turning point in how compute availability is negotiated. Raw GPU silicon availability is no longer the primary bottleneck in hyperscale expansion; access to continuous, decarbonized power is now the hard constraint defining cloud region roadmaps. By co-locating multi-facility campuses near high-voltage corridors in northern Finland rather than congested metropolitan centers, hyperscalers are actively redirecting high-density training clusters to areas with underutilized transmission capacity and natural thermal advantages. This expansion fits into a broader macro trend across 2026 where cloud providers are directly intervening in national energy generation to satisfy the soaring power appetites of large foundation models. With data center power consumption surging and corporate net-zero targets looming, hyperscalers are moving past standard virtual renewable energy certificates (RECs) toward 24/7 carbon-free energy (CFE) architectures. Sponsoring nuclear life extensions and deploying co-located battery energy storage systems (BESS) allows operators to smooth intermittent renewable generation and run AI training clusters around the clock without drawing punitive grid surcharges. In practice, engineering organizations should expect regional cloud deployment strategies to increasingly diverge between latency-sensitive inference nodes and compute-heavy batch training zones. Workload orchestration pipelines will increasingly need to support carbon-aware scheduling and regional tiering, dispatching non-latency-critical training jobs to power-resilient Nordic regions while keeping inference endpoints closer to population centers. Platform teams should assess their multi-region topologies to leverage capacity in emerging high-efficiency Nordic nodes as they come online.