The Next AI Debate Is About Geopolitics

AI is a general-purpose technology. But unlike previous general-purpose technologies, such as electricity or steam engines, AI-enabled tools proliferated so quickly that cutting-edge innovations became widely available almost overnight, in the form of chatbots, image generators, and—increasingly—virtual co-pilots. The AI industry also depends on a network of global commercial partners, including not only U.S. and Chinese technologies, but also Taiwan’s semiconductor fabrication plants, extreme ultraviolet lithography machines made in the Netherlands, and other critical supply chain inputs. Competition over AI has so far been dominated by debates about leading-edge semiconductors, but the next phase is also about geography and power. Specifically, where can the data centers that power AI workloads be built? And who has the capital, energy, and infrastructure needed to power the data centers where AI workloads run?

These questions about the future of AI are not only for technologists. Data centers are the factories of AI, turning energy and data into intelligence. Industry leaders estimate that a few major U.S. technology companies alone are expected to invest more than $600 billion in AI infrastructure, particularly in data centers, between 2023 and 2026. The countries that work with companies to host data centers running AI workloads gain economic, political, and technological advantages and leverage. But data centers also present national security sensitivities, given that they often house high-end, export-controlled semiconductors and governments, businesses, and everyday users send some of their most sensitive information through them. And while the United States is ahead of China in many aspects of AI, especially in software and chip design, America faces significant bottlenecks with data centers.

Data is sometimes called the “new oil.” But there’s a crucial difference when it comes to data centers. Nature determines where the world’s oil reserves are, yet nations decide where to build data centers. And if the United States cannot break through bottlenecks at home, it will need an overflow option abroad. The possibility of a global AI infrastructure buildout presents an opportunity for governments and enterprises to practice data center diplomacy.

An aerial view shows a giant Amazon Web Services data center in Ashburn, Virginia, on July 17.Nathan Howard/Getty Images

It’s not just a question of how those energy needs can be met, but where. When it comes to data centers, the shortage of powered land in the United States—or more specifically, the shortage of powered land with the connectivity required to support large-scale data centers—combined with supply chain challenges and lengthy permitting timelines for new infrastructure—presents a challenge to realizing both the public and private sectors’ AI ambitions.

The data center power wall has been a long time coming, and the proliferation of AI has accelerated it. Each ChatGPT query requires nearly 10 times as much electricity to process as a Google search. Such proportions aren’t  limited to one interface, however – most modern AI models utilize quadrillions of calculations per second, each of which requires energy to run. This unprecedented throughput of computation is increasingly served by energy-intensive graphics processing units (GPUs) rather than central processing units (CPUs). To support ever-growing demands for computation, modern GPUs push the boundaries of how much power can go through a single piece of silicon. Along with this growth in GPUs, the proliferation of AI tools has marked another acceleration in energy demands, with consumers and enterprises increasingly relying on AI worldwide.

Depending exclusively on projections in fields as dynamic and unpredictable as energy and technology is always risky. But with more and more intensive computation requirements, it’s clear that the expected energy needs for future data centers all point in one direction: up. Goldman Sachs Research estimates that data centers used three percent of U.S. power in 2022, a number that could reach eight percent by 2030. The Federal Energy Regulatory Commission’s staff expects data center usage to rise from 17 gigawatts (GW) in 2022 to 35 GW in 2030. The International Energy Agency predicts that global data center electricity consumption could double as soon as 2026, driven largely by AI. If data center power demand climbs from 460 terawatt-hours in 2022 to 1,000 terawatt-hours in 2026, as some expect, that growth would be roughly equivalent to the electricity consumption of Japan.

Even at a smaller scale, the result could be a shift in the basic assumptions for the energy grid unlike any in modern U.S. history. Despite economic and population growth, power demand in the United States and most other leading economies has remained flat or even declined over the past two decades. There have been efficiency gains, and the composition of power supply has shifted in recent years, with a large buildout of intermittent, zero-marginal cost generation sources such as wind and solar. Retirements, meanwhile, have been mostly baseload and dispatchable generation, such as coal, nuclear, older gas-fired generation.

The proliferation of AI can solve some of the second-order challenges the technology has created while contributing to the green-energy transition. For example, the U.S. Energy Department has highlighted AI-driven opportunities in grid planning, grid resilience, and materials discovery for clean energy technologies. These opportunities, however, don’t change the fact that across industries, users increasingly demand unprecedented levels and types of output from electrical grids. Many developed markets’ electrical grids are unprepared and unused to adapting rapidly and at scale—in the United States, the average lead time to build new electricity grid assets is up to 10 years.

Read More

America’s AI Leadership Depends on Energy

Microsoft’s plan to restart Three Mile Island points to the way forward.

Argument

|

Jason Bordoff, Jared Dunnmon

A Realist Perspective on AI Regulation

Experimentation is the right strategy—as long as regulators can learn from one another.

Argument

|

Viktor Mayer-Schönberger, Urs Gasser

This Year’s Nobels Were a Warning

AI won big, but the prizes served as a reminder of what we could all lose.

Argument

|

Bhaskar Chakravorti

Even with adequate power and connectivity, existing U.S. data centers cannot meet the growing demands of AI workloads. The United States’ data centers are primarily based in Silicon Valley and Northern Virginia—particularly in eastern Loudoun County’s “Data Center Alley” outside Washington, D.C., which has the world’s largest concentration of data centers. But vacancy rates have hit record lows of less than three percent, meaning that most facilities cannot take on additional workloads.

Equipping today’s data centers to handle future AI workloads isn’t yet a scalable solution—AI workloads are ultra-high-density, requiring concentrated power supplies. Those that run high-end chips often need liquid cooling, making retrofitting prohibitively expensive or complicated in many cases. Even if they were retrofitted, traditional non-AI cloud workloads performed in data centers aren’t going away and would still need to be performed somewhere. The punchline: The United States needs far more power and far more differentiated data centers.

As AI demands grow, the urgency of the data center bottleneck also grows. Advancements in AI cannot occur without access to the power and transmission needed for the workloads. Left unresolved, the data center power and vacancy walls could threaten the future of AI innovation, and U.S. competitiveness.

Meanwhile, China is executing its own strategy to lead in AI infrastructure. Despite its slowing economy and the resulting reduced power demands, the country’s energy investments remain robust, with dozens of nuclear reactors planned or under construction. China accounted for one-third of clean energy investments worldwide, even as its coal production reached a record high in 2023. Beijing is using that power to fuel a national data center initiative launched in 2022 called “Eastern Data, Western Computing,” which included a $6.1 billion investment in eight major data center hubs.

The United States has many of the tools it needs to compete at home and should unleash them. It’s an energy-rich country that produces more oil than any nation in history, and U.S. natural gas production has boomed since the shale gas revolution of the early 2000s. A more robust power grid that embraces diverse energy resources—including nuclear power, small modular reactors, and reactivated nuclear plants—could change the market calculations in America’s favor.

But having energy reserves does not mean that those resources have the transmission or connectivity to serve end users, and energy resources from places such as the Permian Basin in Texas or North Dakota’s oil fields would need to be connected to data centers to be of service. Complex regulatory and permitting processes at the national, state, and local levels make that task both capital and time intensive.

The United States has led energy revolutions at home in the past. However, necessary reforms and innovations are not guaranteed at the speed or scale that today’s technological revolution and geopolitical competition require, and staying ahead is of the essence in competition.

America cannot achieve AI autarky, especially when it comes to data centers. AI software must run on AI hardware somewhere—the question is where. The United States needs to develop a list of partners with the capacity, will, and aligned interests for a secure global data center buildout.

A visitor observes semiconductor manufacturing on a wall of monitors in Hsinchu, Taiwan, on May 1, 2000. David Hartung/Getty Images

Meanwhile, geopolitical competition is making countries turn toward greater domestic control and data localization, or sometimes “sovereign AI,” a movement that means even if countries could access data infrastructure elsewhere, they work to bolster their domestic capacity. There will be redundant AI infrastructure at the country level, such as countries and companies buying excess semiconductors and building additional data centers with the goal to advance their national resilience over economic efficiency. This should translate into a broader buyer universe for GPUs than we saw in the cloud era, when chip manufacturers sold CPUs almost exclusively to a small handful of large cloud service providers with highly concentrated buying power.

As the United States contemplates its own buildout, it will need to answer questions about domestic capacity and reform, and where outside of the country trusted data centers can be built and connected. Indeed, every country will face hard choices about where their AI workloads will run. What criteria can guide these decisions?

First, the national security and commercial vulnerabilities are real—but not every lesson from previous technology competitions applies. Though there are similar concerns about trust, security, and privacy, the data center debate is not entirely analogous to earlier competitions over 5G. Then, the West struggled to provide a credible and affordable alternative to China’s Huawei, which had become the primary telecommunications provider in many countries, particularly in the global south. However, no single country or company controls the future of data centers. There are currently around 8,000 data centers globally built by a diverse group of companies, demonstrating a wide array of potential partners for data center buildouts.

Second, we know what makes a location well-suited to host data centers. Zoning and regulatory frameworks must support or adapt to efficient infrastructure buildouts. Host countries need access to advanced chips, an increasingly contentious topic given China’s AI strategy and U.S.-led multilateral export controls. Data centers need high-bandwidth digital connectivity to transmit and receive data to and from users. Perhaps most importantly, data centers must operate 24/7, requiring land with access to abundant, affordable, and reliable power.

Third, few factors in this competition remain constant. Innovation can boost efficiency and unlock new locations for data center buildouts. Energy companies use AI to improve performance, particularly in predicting power supply and demand. Government and industry are working to improve the efficiency of chips, with denser circuits and new architectures already reducing semiconductor energy needs at remarkable scale. Meanwhile, large-language model workloads require less bandwidth than traditional internet content such as images and videos, and AI applications often have different latency requirements than traditional cloud services. While the debate is far from settled, innovation may enable future data centers to be built farther away from customers.

Many hyperscalers have made sustainability commitments requiring ever-more renewable sources of energy such as wind and solar. While AI is driving efficiencies and accelerating aspects of the green-energy transition, many forms of renewable energy currently provide only intermittent or insufficient power for data centers. This has led to increased scrutiny of hyperscaler sustainability commitments as companies try to balance innovation with environmental goals.

The list of locations that satisfy all or even most of these criteria is small, but not zero. There are established democracies, countries hedging between Washington and Beijing, and geopolitical swing states with ambitions for technology leadership. The United States has an advantage through both its technology leadership and its partnerships with countries critical to AI’s future. Think of them as the AI swing states.

Wind turbines in a field along the Crowsnest Highway in rural Alberta on Oct. 3, 2021, part of what powers Canada’s sprawling power grid. Artur Widak/NurPhoto via Getty Images

Data centers have given Europe an opportunity to play a leading role in technological innovation. The Nordic countries—leaders in green energy and now all members of NATO since Russia’s full-scale invasion of Ukraine in 2022—have exceptional technology companies of their own, including telecommunications giants. Their connectivity, energy sustainability, and access to power have long made them data center hubs for hyperscalers. Their cool climates also help prevent overheating in data centers, potentially allowing better performance levels and lower costs over the long term.

U.S. allies such as Japan and South Korea have world-leading technology ecosystems and are critical leaders in AI. Japan stands out, having dominated much of the global semiconductor market in the 1980s. In an effort to regrow that industry domestically, Tokyo is investing 0.71 percent of its gross domestic product on semiconductors through 2025—a much higher figure than most industrial economies, including the United States. And though India’s economy and infrastructure are not as advanced as other potential partners, with renewables constituting only 30 percent of its energy usage, New Delhi is making significant data center investments while strengthening its role as a technology partner to Western and Western-aligned firms through forums such as the Quad and the India-Europe-Middle East Economic Corridor.

There are countries with which Washington is seeking to partner in the global competition with China that also have the potential to host more data centers running AI workloads. Brazil, a major non-NATO ally, produces 83 percent of its energy from renewable sources, mainly through large hydropower plants, though reliable access continues to be a challenge in parts of the country. Vietnam—now America’s sixth largest source of U.S. imports—and the Philippines, which has a mutual defense treaty with the United States, are also contenders.

The Arab Gulf countries of the Middle East present many promising opportunities for AI data centers. With young, ambitious leaders, these countries aim not just to export oil, but also AI. As one prominent official from the United Arab Emirates recently emphasized, “We missed the first industrial revolution, but we are not missing the AI revolution.”

The Gulf Cooperation Council is comprised of energy-rich countries with advanced digital infrastructure. Saudi Arabia and the United Arab Emirates have some of the highest internet penetration rates worldwide. The subsea fiber optic cables that are the backbone of the modern internet have critical nodes in the Red Sea and the Persian Gulf, where 90 percent of Europe-Asia data traffic is carried. The region’s geographic position, connecting Asia and Europe, can give it a differentiated role in bringing the global south into the age of AI.

The server room at a Facebook data center in Lulea, in Swedish Lapland, on Nov. 7, 2013. Jonathan Nackstrand/AFP via Getty Images

A surge in capital has reshaped the Middle East, making the Gulf rise in prominence as compared to traditional regional leaders such as Egypt and Syria. With historic and continued demand for hydrocarbons—more than 80 percent of global energy needs are still met by fossil fuels— they also have access to flexible, long-term capital required for data center investments. A total of $11.3 trillion is managed by sovereign wealth funds globally, and five of the 10 most active funds are in the Arab Gulf states. Each state and sovereign wealth fund has a different strategy. Still they are all moving their petrodollars toward investments in the energy transition and domestic industries, from life sciences to telecommunications to manufacturing.

A modernized, technologically-advanced Gulf whose leaders see more opportunities with partnerships with the West is in Washington’s interests. And while conflict has slowed growth in much of the region, the Gulf economies have been insulated and continue to grow. AI is a part of that transformation, as leaders understand that it will accelerate the Gulf states’ national development and economic diversification across industries. Saudi Arabia’s Vision 2030 has increasingly focused on technology, for which the kingdom has made a $100 billion investment plan. The UAE has made deals that align the country’s investments more closely with technology from the United States and other Western partners. With the 2022 FIFA World Cup, Qatar built massive infrastructure projects of its own. Under its 2019 National AI Strategy, Qatar’s AI market has grown substantially and is helping to transition Doha into a knowledge-based economy with national champions in industries such as energy and aviation.

This list of partners is not exhaustive, but it demonstrates the array of locations for technological partnerships and commercial and geopolitical competition. As AI workloads expand and trend from training to inference—the process by which a trained model draws conclusions—the amount of computing required could grow faster still. Data center diplomacy—proactively identifying able, willing, and trusted international partners; pooling public and private capital; identifying and addressing security and privacy risks; and incentivizing innovation across the technology stack—is growing more urgent, and more promising.

Navigating the geopolitics of this competition will require close partnerships between the public and private sectors. Not every country will be the world leader in AI. But more nations than the United States and China can lead. To win in today’s high-stakes geopolitical competition, the United States will need to enlist its asymmetric advantage of global alliances and partnerships, both in the public and private sectors. The data center buildout puts geography at the center of technological progress and competition. If the United States is successful, it is more likely that the future world, in which machines play a greater role in daily life, will also be one with greater human prosperity and freedom.