Google is no longer just looking for land and processors for its data centers; it is now securing the electricity that will power them for decades. The tech giant has signed a 22-year power purchase agreement with Finnish utility Fortum to purchase up to half the capacity of the Loviisa nuclear power plant starting in 2030, equivalent to roughly 450 megawatts of continuous power. This move, paired with a €13 billion investment in Finnish infrastructure during 2027 and 2028, signals a fundamental shift in how hyperscalers, the massive cloud computing companies that power artificial intelligence, are approaching their expansion.

Why Is Nuclear Power Becoming the New Prize for AI Companies?

The race to build artificial intelligence systems has created an unprecedented hunger for electricity. Training large language models, the AI systems that power chatbots and search engines, consumes enormous amounts of power continuously. Rather than hoping the grid can keep up, Google is taking control of its energy destiny by locking in nuclear capacity decades in advance. The Loviisa plant, which has operated since the late 1970s, generates around 8 terawatts of electricity annually, close to 10 percent of Finland's total electricity consumption. Google's contract could represent about 4 terawatts annually once fully ramped up.

What makes this arrangement particularly clever is that it solves a problem for both parties. Fortum, the utility operator, needs to invest around €1 billion between 2023 and 2050 to modernize the two reactors at Loviisa. Without Google's commitment to purchase power for 22 years, the company could not justify spending that money today. By guaranteeing demand, Google provides the financial visibility that allows Fortum to continue operating the plant beyond its original lifespan. The utility estimates the agreement will improve its return on assets by 1.4 percentage points once half the plant's capacity is under contract.

How Are Tech Giants Building Complete Energy Ecosystems Around Their Data Centers?

Google's strategy extends far beyond securing nuclear power. The company is assembling what amounts to a private energy portfolio designed to keep its data centers running reliably and sustainably. This integrated approach includes:

• Nuclear baseload power: The Loviisa agreement provides continuous, predictable electricity generation that runs 24/7, forming the foundation of Google's power supply.
• Renewable wind capacity: Google has increased the capacity of new onshore wind projects it supports in Finland to 629 megawatts, partnering with companies like Valorem and Suomen Hyötytuuli to capture variable renewable energy.
• Battery storage systems: A 94-megawatt battery is scheduled to enter service at the end of 2027 in Kajaani, designed to absorb short-term fluctuations in wind generation and help stabilize grid frequency.
• Computational flexibility: Google is working with grid operators to temporarily reduce data center consumption or shift computing tasks to other locations during periods of grid strain, effectively turning its servers into a grid-balancing tool.

This multi-layered approach reflects a fundamental truth about modern AI infrastructure: electricity is no longer just a utility cost, it is the primary constraint on growth. The company is not merely purchasing power; it is architecting an entire energy system tailored to the unpredictable demands of AI workloads.

The battery storage component deserves particular attention. Fortum will optimize the 94-megawatt battery and may deploy it in ancillary services markets operated by Fingrid, Finland's transmission system operator. This means the battery can earn revenue by helping the broader grid manage fluctuations, turning what might otherwise be a pure cost center into a potential profit center.

What Does This Mean for the Broader AI Infrastructure Race?

Google is not alone in recognizing that energy geography now determines computing geography. Microsoft described its strategic evolution as "power first" as early as 2025, fundamentally reorienting how the company thinks about data center placement. Rather than building near users or anticipated demand, the next generation of AI infrastructure follows abundant low-carbon electricity.

Microsoft is pursuing a different but complementary model in Finland. The company is developing around a dozen data centers across three sites, with two campuses in Espoo and Kirkkonummi designed to feed waste heat into district heating networks. Around 75 percent of the heat generated over a year should be recoverable, potentially covering nearly 40 percent of the heating needs for roughly 250,000 residents across Espoo, Kauniainen, and Kirkkonummi. Microsoft is also exploring using its facilities' backup batteries to help regulate the grid.

Amazon is taking yet another approach, concentrating new investment across multiple European countries including Spain, Germany, France, Ireland, and Sweden. The company's largest recent announcement involves a €33.7 billion investment in Aragón through 2035 to expand the Spanish region's cloud and AI capacity, including not just data centers but also server assembly facilities, repair centers, and logistics hubs.

These divergent strategies reveal an important truth: there is no single winning formula for AI infrastructure. Google builds its own integrated energy systems when it can secure long-term power contracts. Microsoft leases capacity from specialist operators when local developers can convert surplus hydropower into computing capacity faster. Amazon builds vertically integrated supply chains to keep manufacturing and logistics local. Each approach reflects different assumptions about where electricity will be abundant, how quickly infrastructure can be built, and what level of control each company needs over its computing destiny.

The Finland laboratory is particularly revealing because it allows direct comparison. Google's nuclear strategy provides decades of predictable baseload power but requires massive upfront coordination with utilities. Microsoft's heat recovery approach generates revenue from waste products but depends on proximity to district heating networks. Amazon's supply chain integration addresses manufacturing bottlenecks but requires massive capital investment across multiple facility types. None of these approaches is inherently superior; they are different solutions to the same underlying problem: AI is consuming electricity faster than traditional infrastructure planning can accommodate.

What emerges from these parallel strategies is a new reality for the technology industry. The companies winning the AI race will not necessarily be those with the fastest chips or the most sophisticated algorithms. They will be the ones that can secure reliable, affordable electricity for the next two decades. Google's €13 billion commitment to Finland, backed by a 22-year nuclear power agreement, is a bet that this is true. The company is essentially saying that controlling its energy supply is as important as controlling its computing hardware.