AI's Hidden Geography Shapes Global Tech Landscape

By signing a 22-year power agreement with a local nuclear plant, Google has committed 13 billion dollars to build three data centers in northern Finland, where colder climates lower operational costs.

The company signed a 22-year agreement to secure power from a local nuclear reactor, highlighting how modern data centers now resemble heavy industry rather than traditional tech hubs, while colder climates like Finland offer natural cooling advantages that reduce operational costs.

A country of 5.6 million people, Finland is attracting significant investment as developers build three large data centers in its northern reaches to leverage natural cooling.

Google secured a 22-year agreement to access roughly half of a Finnish nuclear power plant's output, underscoring the critical importance of stable and affordable energy sources for AI operations.

The country's cold climate provides a natural advantage for cooling the immense heat generated by these facilities, a key factor in reducing the operational expenses of large-scale data centers.

Rather than clustering near major hubs like London, Amsterdam, Frankfurt, and Paris to satisfy latency needs, AI training infrastructure is relocating to more remote areas since physical proximity to users is now unnecessary.

This relocation is driven by the fact that AI training processes do not require physical proximity to end-users, unlike traditional data centers that needed urban proximity to minimize latency for businesses and consumers; consequently, data centers planned between 2026 and 2028 are expected to be located an average of 175 km from major cities, compared to a previous average of 46 km.

Developers are opting for regions like the Spanish countryside instead of urban markets like Amsterdam, where land with secured electricity costs €2.7 million per megawatt.

For instance, land with secured electricity in Amsterdam costs approximately €2.7 million per megawatt, whereas in smaller European markets, these costs can drop below €500,000 per megawatt, driving AI infrastructure development toward regions like the Spanish countryside and northern Sweden.

The electricity demand for AI is escalating rapidly, with global data center consumption reaching roughly 415 terawatt-hours in 2024, and the International Energy Agency (IEA) projects this to nearly triple to 1,200 terawatt-hours by 2035.

This surging demand is threatening to overwhelm existing grids, with slow transmission line deployment potentially delaying one-fifth of planned data center capacity through 2030, forcing tech giants to compete directly with factories and residential sectors for limited power capacity.

Google and Microsoft are reviving nuclear stations in the U.S. and Finland to meet surging energy demands and secure stable operations for their AI infrastructure.

Google is not only involved in Finnish nuclear operations but also reopening nuclear stations in the U.S. state of Iowa, while Microsoft supports the revival of the Three Mile Island reactor; this AI boom also forces a physical integration with traditional mining, refining, and manufacturing sectors, creating downstream demand for materials such as copper, silicon, aluminum, steel, and rare earths.

Beijing launched the 'Eastern Data Western Compute' project to strategically bridge the gap between its wealthy eastern population centers and the resource availability in its western regions.

Under this initiative, data generated around China's eastern cities is sent westward for processing in large computing hubs located closer to cheaper electricity and land, mirroring a strategy seen in industrial production where manufacturing moves closer to raw material sources.

Chips designed in the US, produced by TSMC in Taiwan using Dutch ASML lithography, and powered by Finnish energy form the complex, fragile physical chain of modern AI.

Advanced chips, primarily designed in the US by companies like Nvidia, are predominantly fabricated in Taiwan by TSMC, a process that depends on decades of specialized infrastructure.

Furthermore, ASML in the Netherlands holds a near-monopoly on the extreme ultraviolet lithography systems essential for this advanced chip fabrication, illustrating a supply chain where an AI model may depend on US design, Dutch machinery, Taiwanese fabrication, and Finnish energy-powered data centers.

JP Morgan identified electricity capacity, semiconductors, and critical materials like gallium as emerging choke points; the IEA projects data centers alone could consume over 10% of global gallium production by 2030, with almost all of it sourced from China, driving governments to aggressively localize industrial capabilities for AI security.

Regional incidents have prompted the UAE to disperse its multi-gigawatt AI campus to mitigate the physical vulnerabilities inherent in specific geographical locations.

For example, the UAE is dispersing its multi-gigawatt AI campus to protect against physical attacks following regional incidents, while Taiwan's critical chip industry faces significant proximity risks due to being only 100 kilometers from mainland China.

Data centers are also inherently limited by local grid capacity, the locations of undersea cables, and dependencies on fiber optic networks; thus, political stability and favorable geography provide long-term advantages that software improvements alone cannot replicate.

Future technological advancements, particularly in cooling systems and chip efficiency, could diminish the current necessity for cold-weather locations like Finland, potentially reshaping the geographic advantages for AI infrastructure.

This evolution might make regions with abundant capital and energy, such as the Middle East, more suitable hosts for AI operations; however, regardless of these technological leaps, AI will remain fundamentally rooted in physical constraints and geographical considerations.