The race for artificial intelligence dominance is not just about who builds the smartest algorithms; it is fundamentally about who controls the physical infrastructure that powers them. Hyperscale data centers, the massive industrial facilities that train and run AI models, are increasingly central to geopolitical strategy, constrained by raw materials, manufacturing capacity, and data laws that remain stubbornly tied to geography. As the US and China compete for AI supremacy, control over the infrastructure underpinning digital power is reshaping how nations approach technology competition.

What Are Hyperscale Data Centers and Why Do They Matter for AI?

Hyperscale data centers are industrial-scale computing facilities that consume more than 10 megawatts of continuous power, with some reaching over 200 megawatts, equivalent to powering a medium-sized city. These facilities house tens of thousands of servers and high-performance networking equipment, enabling cloud services, scientific computing, and critically, the training and deployment of advanced AI models. A single hyperscale facility can cost anywhere from hundreds of millions to more than a billion dollars to build and operate, requiring extensive power infrastructure, backup generators, advanced cooling systems, and both cyber and physical security.

For AI specifically, these data centers are where the computational heavy lifting happens. Training a frontier large language model, or LLM (a type of AI that processes and generates human language), requires enormous amounts of computing power. The location of these facilities depends on three critical dimensions: access to raw materials and specialized manufacturing, existing geographical attributes like water and energy availability, and data sovereignty laws governing where data can be stored and processed.

How Does Geography Create Chokepoints in the AI Supply Chain?

The supply chain for advanced AI chips is highly concentrated and geographically fragmented, creating multiple points where one country or company can exert control. At the raw materials level, China dominates upstream supply of several critical minerals essential to semiconductors and data centers. China accounts for approximately 60 to 70 percent of global rare-earth mining and nearly 90 percent of rare-earth processing capacity. These rare-earth elements are used in semiconductor manufacturing, and some are also used in data transmission between servers within data centers. Additionally, China holds up to 94 percent of global production of rare-earth magnets, key components in semiconductor manufacturing processes like precision wafer polishing and extreme ultraviolet lithography, as well as in data center operations like data storage and cooling systems.

Beyond raw materials, the manufacturing of advanced AI chips is dominated by a small number of foundries concentrated in East Asia. Taiwan's Taiwan Semiconductor Manufacturing Company Limited, or TSMC, supplies the majority of the highest-performance chips used in AI accelerators, accounting for roughly two-thirds to over 70 percent of the global pure-play foundry market, which consists of manufacturers that fabricate chips designed by other companies. South Korea's Samsung Foundry represents the only other supplier operating at comparable scale.

Even within chip fabrication, another critical chokepoint exists in lithography capabilities. Extreme ultraviolet, or EUV, lithography tools are essential for producing the most advanced chips at scale, and supply is effectively dominated by a single vendor. The Dutch company ASML holds a monopoly on EUV lithography and approximately 90 percent of the broader lithography market, making the Netherlands a critical node in the frontier chip pipeline.

Why Physical Geography Still Matters in a Digital World

While digital technologies often seem borderless, the physical infrastructure supporting them remains deeply dependent on geography and national data laws. Hyperscale data centers require specific geographical conditions to operate efficiently. They need abundant water for cooling systems, reliable and affordable energy sources, open space for construction, and proximity to cities where users and talent clusters exist. These requirements mean that not every country can host hyperscale facilities, and those that can gain significant economic and strategic advantage.

Data sovereignty regimes, which govern where data can be stored and processed within national borders, further constrain where hyperscale capacity can be developed. Various states are increasingly localizing their subjects' data to protect against unauthorized foreign access, creating regulatory barriers that shape where companies can build data centers. This intersection of physical constraints and legal requirements means that the geography of hyperscale infrastructure directly influences where digital power can and cannot be built across the geopolitical landscape.

Steps to Understanding the Geopolitical Stakes of Data Center Control

• Raw Material Leverage: China's dominance in rare-earth mining, processing, and magnet production gives it control over upstream supply chains that feed into semiconductor manufacturing and data center operations, creating potential leverage over downstream technology producers.
• Manufacturing Concentration: TSMC's control of roughly two-thirds to over 70 percent of advanced chip fabrication, combined with ASML's monopoly on EUV lithography, means that Taiwan and the Netherlands are critical nodes in the global AI infrastructure pipeline.
• Physical Infrastructure Requirements: Hyperscale data centers require abundant water, reliable energy, open space, and proximity to talent hubs, meaning only certain geographies can host them, creating natural advantages for countries with these resources.
• Data Sovereignty Barriers: National data localization laws restrict where companies can build data centers and store data, fragmenting what appears to be a borderless digital infrastructure into geographically bounded systems.

The concentration of these capabilities across different countries and companies creates what scholars call "technopolitics," where control over technology infrastructure becomes as important as traditional geopolitical assets like territory or energy reserves. The US has already begun using semiconductor export controls as instruments of strategic leverage against geopolitical rivals, recognizing that controlling access to advanced chips is equivalent to controlling access to AI capability itself.

For countries competing in the AI race, the implications are profound. Building hyperscale data center capacity requires not just capital but also access to rare materials, advanced manufacturing, energy infrastructure, and favorable data laws. China's dominance in rare-earth supply and processing gives it leverage over the entire downstream supply chain. Meanwhile, Taiwan's role as the primary supplier of advanced chips and the Netherlands' control of lithography equipment make these countries critical nodes that cannot be easily bypassed. The US, despite its technological leadership in AI software and algorithms, remains dependent on these global supply chains for the physical infrastructure needed to train and deploy frontier AI models.

As geopolitical tensions rise and export controls tighten, the geography of hyperscale data centers is becoming a central battleground in the US-China AI race. Nations are increasingly recognizing that digital power, like traditional power, is rooted in physical geography and cannot be entirely virtualized away. The country or coalition that can secure reliable access to rare materials, advanced manufacturing, energy, and favorable regulatory environments will have a decisive advantage in building the hyperscale infrastructure needed to lead in AI.