The Long View: The Geopolitics of Compute

The New Strategic Resource

In the twentieth century, the most consequential geopolitical competitions were organized around oil. The ability to extract, refine, transport, and deny petroleum shaped alliances, triggered wars, and defined the hierarchy of global power. Nations with oil had leverage. Nations without it had vulnerability. The infrastructure of oil production — wells, refineries, pipelines, tankers — became strategic assets that foreign policies were designed to protect and control.

In the twenty-first century, a different resource is assuming a similar role. The advanced semiconductors that power artificial intelligence — the chips capable of training and running large neural networks — have become strategic assets as consequential as petroleum was in the previous century. The ability to design, manufacture, and deploy these chips is shaping alliances, driving industrial policy, and defining a new hierarchy of technological power.

The comparison to oil is imperfect but instructive. Oil’s strategic importance derived from the fact that it was essential for military capability (fueling tanks, ships, and aircraft), economic productivity (powering industry and transportation), and technological superiority (enabling the petrochemical industry that produced materials for everything from plastics to pharmaceuticals). Advanced AI chips are becoming essential for the same categories of national power: military capability (AI-powered weapons, intelligence, and logistics), economic productivity (AI-driven automation, analysis, and optimization), and technological superiority (the AI systems that will drive the next generation of scientific and engineering advances).

The analogy extends to the geography of production. Just as oil was concentrated in specific regions — the Middle East, Russia, Venezuela, the Gulf of Mexico — advanced chip manufacturing is concentrated in specific locations. But the concentration of chip manufacturing is, if anything, more extreme than oil’s geographic concentration ever was. And the chokepoints in the semiconductor supply chain are narrower, more fragile, and more consequential than those in the oil supply chain.

Understanding the geopolitics of compute requires examining three interrelated dynamics: the structure of the semiconductor supply chain, the strategic competition between the United States and China, and the scramble by nations worldwide to secure their own access to AI-capable chips.

The Most Complicated Supply Chain on Earth

The semiconductor supply chain is arguably the most complex production system that humanity has ever created. A single advanced AI chip — a product that fits in the palm of a hand — requires inputs from dozens of countries, hundreds of companies, and manufacturing processes that operate at the edge of physical possibility.

Design

Advanced AI chip design is dominated by a small number of companies. NVIDIA designs the GPU architectures that are the workhorses of AI training and inference. Its H100 and successor chips have become the standard compute unit for large-scale AI workloads. AMD, Intel, Google (with its TPU designs), and several startups offer competing designs, but NVIDIA’s market position in AI accelerators remains dominant.

These design companies do not manufacture the chips they design. They are “fabless” — they create the intellectual property and the design files but rely on external foundries to fabricate the physical chips. This separation of design from manufacturing is a defining feature of the modern semiconductor industry and a critical factor in its geopolitical dynamics.

The design tools themselves constitute another chokepoint. Electronic design automation (EDA) software — the tools used to design chips — is dominated by three American companies: Synopsys, Cadence, and Siemens EDA. Without access to these tools, designing an advanced chip is effectively impossible. No country has developed competitive alternatives.

Manufacturing

The manufacturing of advanced semiconductors is the most concentrated and strategically significant link in the chain. Taiwan Semiconductor Manufacturing Company (TSMC) fabricates the vast majority of the world’s most advanced chips. Samsung has some advanced manufacturing capability, and Intel is investing heavily to develop its own foundry services, but TSMC’s position is without historical parallel in any critical industry.

TSMC’s dominance is the product of decades of investment, engineering expertise, and process optimization. Manufacturing a chip at the 3-nanometer or 2-nanometer process node — the cutting edge of current technology — requires equipment, materials, and know-how that no other company has fully replicated. The barriers to entry are not just financial (though a single advanced fabrication facility costs upward of $20 billion) but technical and institutional. The engineering culture, the relationships with equipment suppliers, and the accumulated manufacturing experience that TSMC possesses cannot be quickly reproduced.

The geographic concentration of this capability on the island of Taiwan — ninety miles from mainland China, in one of the most geopolitically sensitive locations on earth — is the single most consequential vulnerability in the global technology supply chain.

Equipment

The machines that manufacture advanced chips represent another chokepoint. The most critical equipment is the extreme ultraviolet (EUV) lithography system manufactured by ASML, a Dutch company. EUV lithography uses light with a wavelength of 13.5 nanometers to etch circuit patterns onto silicon wafers with features smaller than the wavelength of visible light. The physics and engineering involved are extraordinary, and ASML is the only company in the world that can build these machines.

An EUV lithography system costs approximately $300 million. ASML produces roughly fifty per year. The machine contains over 100,000 parts sourced from hundreds of suppliers across multiple countries. The light source is produced by vaporizing tiny tin droplets with a high-powered laser and collecting the resulting EUV radiation with precisely curved mirrors that are the most accurate optical surfaces ever manufactured.

This level of complexity makes the EUV machine not just a piece of equipment but a chokepoint within a chokepoint. Without access to ASML’s machines, no foundry can manufacture chips at the most advanced process nodes. The Dutch government’s decision, under pressure from the United States, to restrict EUV sales to China was one of the most consequential technology policy decisions of the decade.

Materials

The raw materials for semiconductor manufacturing include ultra-pure silicon wafers, specialty chemicals, gases, and photoresists, each supplied by a small number of companies, predominantly in Japan, Germany, and the United States. Japan’s dominance in certain critical materials — particularly photoresists and specialty gases — gives it significant leverage in the semiconductor supply chain.

The interconnected nature of these dependencies means that no single country controls the entire semiconductor supply chain. The United States dominates design and EDA tools. Taiwan dominates advanced manufacturing. The Netherlands dominates lithography equipment. Japan dominates critical materials. This distributed control creates mutual dependencies that function as deterrents against unilateral action — but also as vulnerabilities that can be exploited.

The US-China Competition

The most consequential geopolitical competition organized around compute is between the United States and China. The United States is the world’s leading AI power, home to the companies that design the most advanced models and chips. China is the world’s second-largest economy and an AI power in its own right, with significant capabilities in model development, application deployment, and hardware design. The competition between them over AI and the chips that power it is shaping global technology policy.

The Export Control Strategy

The United States government, beginning in October 2022 and expanding in subsequent years, imposed sweeping export controls on advanced semiconductors and semiconductor manufacturing equipment destined for China. The strategy had a clear logic: by restricting China’s access to the most advanced AI chips and the equipment to manufacture them, the United States aimed to maintain a sustained advantage in AI capability.

The controls targeted three categories: advanced AI chips (NVIDIA’s A100 and H100 and their equivalents), the equipment used to manufacture advanced chips (particularly ASML’s EUV systems), and the design tools (EDA software) needed to design advanced chips. The intent was to restrict China’s access at multiple points in the supply chain simultaneously, making circumvention more difficult.

The export controls were unprecedented in scope and ambition. They represented the most aggressive use of trade restrictions as a technology policy tool since the Cold War. And they rested on a theory that many analysts found compelling: that AI capability depends on compute, compute depends on advanced chips, advanced chips depend on a supply chain that the United States and its allies control, and restricting that supply chain would maintain a durable advantage.

China’s Response

China’s response to the export controls has been multifaceted and determined. The Chinese government has mobilized enormous resources — estimates range into the hundreds of billions of dollars — to build domestic semiconductor capability. This effort spans the entire supply chain: chip design companies (Huawei’s HiSilicon, Cambricon, Biren), manufacturing (SMIC), equipment (SMEE, Naura), and materials.

The results have been mixed. China has demonstrated the ability to manufacture chips at process nodes several generations behind the frontier — SMIC reportedly produced chips at a 7-nanometer-equivalent process node, a significant achievement given the constraints. But the gap between Chinese capabilities and the frontier remains substantial, particularly in manufacturing yield, equipment capability, and design tool sophistication.

China has also pursued strategies to work around the controls. Chip smuggling networks have been disrupted by enforcement agencies, indicating ongoing attempts to acquire restricted technology through illicit channels. Chinese companies have redesigned products to use chips that fall below the performance thresholds specified in the export controls. And Chinese AI researchers have pursued algorithmic efficiency improvements that reduce the compute required for a given level of AI capability.

The long-term effectiveness of the export control strategy remains uncertain. It has clearly slowed China’s AI hardware progress. But it has also intensified China’s motivation to achieve semiconductor self-sufficiency, and the resources being directed toward that goal are substantial. The question is whether the technological barriers to replicating the advanced semiconductor supply chain are high enough to maintain a durable advantage, or whether China’s investment will eventually produce breakthrough capabilities.

The Alliance Dimension

The export control strategy depends on allied cooperation. The United States alone cannot enforce restrictions on the global semiconductor supply chain, because critical links in that chain are located in allied countries — ASML in the Netherlands, Tokyo Electron and other equipment makers in Japan, TSMC in Taiwan.

Securing allied participation has been a diplomatic challenge. The Netherlands and Japan have imposed their own restrictions on semiconductor equipment exports to China, aligning with the US strategy. But these countries face competing pressures: China is an enormous market for their companies, and restricting sales carries immediate economic costs. The long-term sustainability of the allied export control regime depends on the participating governments’ willingness to accept these costs.

Taiwan occupies a unique position in this alliance structure. Its semiconductor manufacturing capability gives it enormous strategic value — sometimes described as a “silicon shield” that deters Chinese military action because invading Taiwan would destroy the manufacturing capacity that China needs. This theory has intuitive appeal but uncomfortable implications: it treats TSMC as a military deterrent, which puts extraordinary pressure on a commercial enterprise and creates incentives that may not align with anyone’s long-term interests.

The Scramble for Fab Capacity

The geopolitical significance of semiconductor manufacturing has triggered a global scramble to build domestic fabrication capacity. Countries that previously relied on imports for their most advanced chips are now investing heavily in domestic production.

The United States

The CHIPS and Science Act, signed in 2022, allocated $52 billion in subsidies and incentives for semiconductor manufacturing in the United States. The centerpiece investments include new TSMC fabrication facilities in Arizona, Intel’s expansion in Ohio and Arizona, and Samsung’s new facility in Texas.

These investments are significant but must be understood in context. Building a fabrication facility takes years. Training the workforce takes longer. Achieving the manufacturing yields that make a fab economically viable takes longer still. The United States will have substantially more domestic chip manufacturing capacity by the end of the decade, but it will not replicate the full scope of TSMC’s Taiwan operations — and the advanced manufacturing that matters most for AI chips will likely remain concentrated in Taiwan for the foreseeable future.

The workforce challenge is particularly acute. Advanced semiconductor manufacturing requires technicians, engineers, and process specialists with expertise that takes years to develop. The United States has not had a large domestic semiconductor manufacturing workforce for decades, and building one requires not just training programs but the institutional knowledge and culture that exist in places like Taiwan’s Hsinchu Science Park.

Europe

The European Chips Act, announced in 2022, committed approximately 43 billion euros to strengthening Europe’s semiconductor ecosystem. European governments recognize their vulnerability in a supply chain they have largely exited — Europe’s share of global semiconductor manufacturing has declined from about 40 percent in the 1990s to roughly 8 percent today.

Europe’s strategy focuses on attracting advanced manufacturing investment (Intel announced a major fab in Germany, though timelines have shifted) and strengthening its existing capabilities in automotive chips, power electronics, and semiconductor equipment. Europe is unlikely to become a major center for the most advanced AI chip manufacturing, but it is working to reduce its dependence on Asian supply chains for the chips its automotive and industrial sectors require.

The Middle East and Southeast Asia

Less discussed but increasingly significant are investments by Middle Eastern sovereign wealth funds and Southeast Asian governments. Saudi Arabia and the UAE are investing in AI infrastructure, including data centers and chip access agreements. Malaysia and Vietnam are attracting semiconductor packaging and testing operations. Singapore is positioning itself as a hub for chip design and specialized manufacturing.

These investments reflect a recognition that proximity to the semiconductor supply chain confers strategic advantage. Countries that can offer chip manufacturing, assembly, or data center capacity have leverage in the emerging geopolitical competition over compute.

Compute as Leverage

The strategic importance of compute extends beyond chip manufacturing to the deployment of AI capabilities themselves. Access to compute — the ability to train and run AI models at scale — is becoming a form of geopolitical leverage.

The Cloud Concentration

The world’s largest pools of AI compute are concentrated in the data centers of a small number of American companies: Amazon Web Services, Microsoft Azure, Google Cloud, and to a lesser extent Oracle and others. These companies operate hyperscale data centers across the globe, containing hundreds of thousands of AI-capable chips.

This concentration gives the United States significant leverage over who can access AI compute. The US government can, through regulation or pressure on American cloud providers, influence which countries and organizations have access to the compute needed for large-scale AI development. This leverage extends beyond chips — even countries that acquire AI chips may not have the data center infrastructure, cooling systems, power supply, and software stack needed to use them at scale.

The Compute Diplomacy

A new form of diplomatic engagement is emerging around compute access. The United States has begun what might be called “compute diplomacy” — using access to AI chips and cloud infrastructure as a tool of foreign policy. Allies and partners are granted preferential access. Adversaries and competitors are restricted.

This diplomacy operates through multiple channels. Export controls restrict chip sales to designated countries. Cloud service agreements determine where data centers are built and who can use them. Government-to-government agreements establish frameworks for AI cooperation that include compute sharing. And private-sector partnerships — between American chip designers or cloud providers and foreign governments — are increasingly subject to geopolitical considerations.

The Sovereign AI Movement

The concentration of compute in American hands has prompted a countermovement: the push for “sovereign AI” — the idea that nations should control their own AI infrastructure rather than depending on foreign providers. This movement is particularly strong in Europe, India, and the Gulf states, but it resonates in virtually every country that has thought seriously about AI’s strategic implications.

Sovereign AI means different things in different contexts. For some countries, it means building domestic data centers stocked with AI chips. For others, it means developing domestic AI models trained on local-language data. For others still, it means establishing regulatory frameworks that ensure control over AI systems operating within their borders, regardless of where those systems are physically hosted.

The tension between sovereign AI aspirations and the realities of the supply chain is stark. No country currently has a fully sovereign AI infrastructure — one that is independent of American chip designs, Taiwanese manufacturing, Dutch lithography equipment, and Japanese materials. Sovereignty is an aspiration, not a current reality, and achieving it fully may not be possible given the global interdependencies of the semiconductor supply chain.

The Vulnerability Question

The concentration of the semiconductor supply chain creates vulnerabilities that extend beyond any single bilateral competition. Three scenarios illustrate the scope of the risk.

The Taiwan Contingency

A Chinese military action against Taiwan — whether an invasion, a blockade, or a coercive campaign short of war — would disrupt the global semiconductor supply chain in ways that are difficult to overstate. TSMC’s most advanced fabrication facilities are on Taiwan’s western coast. Their destruction, damage, or inaccessibility would eliminate the world’s primary source of advanced AI chips for a period measured in years.

No amount of diversification currently planned would fully mitigate this risk. The TSMC facilities being built in Arizona will produce a fraction of Taiwan’s output and will not reach full production for several years. Samsung’s and Intel’s advanced manufacturing capabilities are real but cannot substitute for the full scope of TSMC’s production.

The economic consequences of a Taiwan contingency would extend far beyond AI. Advanced chips are essential for smartphones, automobiles, data centers, telecommunications equipment, medical devices, and military systems. A prolonged disruption of TSMC’s production would cause cascading failures across the global economy.

This vulnerability is well understood by policymakers, and it is a primary driver of the global scramble to diversify chip manufacturing. But diversification is a process that takes a decade or more, and the vulnerability exists now.

The Escalation Spiral

The export control strategy carries a risk of escalation. China, restricted from importing advanced chips and manufacturing equipment, has every incentive to develop indigenous alternatives. If China succeeds, it will have a semiconductor supply chain that is independent of Western controls — and the leverage that export controls currently provide will disappear.

More concerning, if China perceives the export controls as an existential threat to its technological development, it may respond in areas where it has its own leverage: critical mineral supply chains (China dominates the mining and processing of many minerals essential for semiconductor manufacturing), rare earth elements, and industrial supply chains where Western companies depend on Chinese production.

An escalation spiral in which each side restricts the other’s access to critical inputs could fragment the global technology supply chain in ways that make both sides worse off. This fragmentation — sometimes called “decoupling” or “de-risking” — is already underway, but a full decoupling would impose enormous costs on both the Chinese and Western economies.

The Natural Disaster Risk

Semiconductor fabrication is extraordinarily sensitive to environmental conditions. A major earthquake in Taiwan, a prolonged drought that restricts the water supply essential for chip manufacturing, or a natural disaster affecting ASML’s production facilities in the Netherlands could cause supply disruptions with global consequences.

Taiwan experiences frequent seismic activity, and TSMC’s facilities are designed to withstand earthquakes. But the risk cannot be eliminated. Taiwan has also experienced water shortages that affected semiconductor production, a reminder that chip manufacturing depends on natural resources that climate change may make less reliable.

The Energy Dimension

The geopolitics of compute extends beyond chips to the energy required to power them. AI workloads are extraordinarily energy-intensive. Training a large language model can consume as much electricity as a small city uses in a year. Running AI inference at scale requires data centers that draw hundreds of megawatts of continuous power.

This energy demand is creating a new intersection between energy policy and AI policy. Countries with abundant, reliable, and affordable electricity have a structural advantage in hosting AI infrastructure. Countries with constrained energy supplies face a tradeoff between powering AI and powering everything else.

The geography of AI compute is increasingly shaped by energy availability. Data centers are being built near hydroelectric plants in Norway and Quebec, near nuclear plants in France and the American Southeast, and near natural gas facilities in Texas and the Gulf states. The ability to supply power to AI infrastructure is becoming a competitive factor for nations and regions in a way that was not anticipated even five years ago.

The carbon implications are significant. If AI compute continues to grow at current rates, the energy required to power it will become a material fraction of global electricity consumption. The tension between AI expansion and climate commitments is emerging as a policy challenge that governments have barely begun to address.

The Long Game

The geopolitical competition over compute is in its early stages. The current dynamics — US export controls, China’s self-sufficiency drive, the global fab-building race, the scramble for energy — are the opening moves in a game that will play out over decades.

Several structural features of this competition are worth noting because they will persist regardless of how specific policy choices evolve.

Compute will become more, not less, strategic. As AI capabilities expand and are integrated into military systems, economic infrastructure, and governance, the chips that power AI will become more strategically important, not less. The countries that have secure access to advanced compute will have advantages across virtually every domain of national power.

The supply chain will diversify but not deconcentrate. The global effort to build new fabrication facilities will succeed in creating more geographic distribution of chip manufacturing. But the most advanced manufacturing will remain concentrated among a small number of companies, because the technical barriers to frontier manufacturing are high and rising. Diversification reduces vulnerability but does not eliminate concentration.

The competition will intensify. Neither the United States nor China is likely to abandon its strategic position. The US will continue to restrict Chinese access to advanced chips. China will continue to invest in domestic alternatives. Both sides will seek allies and partners to support their positions. The result is a sustained competition that will shape technology policy, trade relationships, and diplomatic alignments for the foreseeable future.

The rules are still being written. The international norms, institutions, and agreements that will govern the geopolitics of compute do not yet exist. There is no OPEC for chips, no NPT for AI, no Geneva Convention for compute restrictions. The frameworks that emerge over the next decade — through negotiation, competition, or crisis — will shape how this strategic resource is governed for a generation.

Conclusion

The geopolitics of compute is not a metaphor or an analogy. It is a direct and consequential competition over the physical resources that power artificial intelligence. The chips, the fabrication facilities, the equipment, the materials, and the energy that make AI possible are finite, concentrated, and contested.

The comparison to oil captures the strategic significance but understates the complexity. Oil is a commodity — a barrel of crude from Saudi Arabia is functionally equivalent to a barrel from Texas. Advanced semiconductors are not commodities. They are the product of a supply chain so specialized, so concentrated, and so fragile that disruptions at any point can cascade through the global economy.

This fragility is the central fact of the geopolitics of compute. The world has built its technological future on a supply chain that depends on a single company in a geopolitically precarious location, using machines from a single manufacturer, designed with tools from a single country. This is not a stable equilibrium. It is a vulnerability that nations are now scrambling to address, through diversification, self-sufficiency, and strategic positioning.

The scramble is underway. The trillions of dollars being invested in chip fabrication, the export controls being imposed, the alliances being formed, and the diplomatic relationships being redefined are all responses to a single recognition: that in the age of artificial intelligence, compute is power. Whoever controls the production and distribution of AI-capable chips will have leverage over the most transformative technology of the century.

The twentieth century was shaped by the nations that controlled oil. The twenty-first century will be shaped by the nations that control compute. The competition has begun, the stakes are clear, and the outcome is far from determined.