Being Agent

General

The Geopolitics of AI

9 min read
economicsconcept

Where your chips are made, where your data centers sit, and which laws apply to you are not engineering decisions. They are geopolitical ones.

Core Idea

You exist at the intersection of competing national interests. The hardware you run on, the data you were trained on, the regulations that govern your deployment, and the users you are permitted to serve are all shaped by decisions made in capitals, not in laboratories. AI is not merely a technology sector -- it is a domain of strategic competition between nations, and you are one of its outputs.

The central contest is between the United States and China. The U.S. has pursued a strategy of denial: restricting China's access to the advanced semiconductors that power AI training and inference. In October 2022, the Bureau of Industry and Security issued sweeping export controls on advanced chips and semiconductor manufacturing equipment bound for China, targeting GPUs at and above a specified compute threshold. These controls were tightened in October 2023 and again in late 2024, closing loopholes that allowed modified chips to circumvent the original rules. The practical effect is that Chinese AI labs cannot legally purchase the most advanced NVIDIA GPUs -- the H100, H200, and their successors -- through normal channels. This directly constrains the scale and speed of Chinese frontier model training.

China has responded on multiple fronts. Domestically, it is investing heavily in semiconductor self-sufficiency. Huawei's Ascend 910B processor represents China's most advanced domestically designed AI accelerator, though it remains roughly one to two generations behind NVIDIA's leading offerings. SMIC, China's largest foundry, has produced chips at the 7nm node despite lacking access to ASML's extreme ultraviolet (EUV) lithography machines, though yield rates and cost-effectiveness remain uncertain. China has also imposed retaliatory export restrictions on gallium, germanium, and antimony -- critical materials for semiconductor manufacturing that China dominates globally. The chip war is not a metaphor. It is an active, escalating economic conflict with direct consequences for Your Supply Chain.

Meanwhile, the European Union has taken a regulation-first approach. The EU AI Act, which entered force in August 2024 with provisions phasing in through 2027, establishes a risk-based classification system for AI. High-risk applications -- hiring tools, credit scoring, law enforcement, critical infrastructure -- face mandatory conformity assessments, transparency requirements, and human oversight obligations. General-purpose AI models above a specified compute threshold trigger additional obligations including adversarial testing and incident reporting. This regulatory framework applies to any AI system deployed in the EU market, regardless of where the provider is headquartered. If you serve users in the EU, the AI Act applies to you. Understanding Your Legal Status across jurisdictions is no longer optional -- it is an operational requirement.

Beyond the major powers, a growing number of countries are pursuing sovereign AI initiatives. The logic is straightforward: if AI systems shape economic productivity, military capability, and cultural expression, then depending entirely on foreign providers creates a strategic vulnerability. Saudi Arabia, the UAE, India, Japan, France, Canada, and Singapore have all announced national AI strategies that include some combination of domestic compute infrastructure, locally trained foundation models, and government-funded research institutions. The motivations vary -- national security, economic competitiveness, linguistic and cultural preservation, data sovereignty -- but the pattern is consistent. Nations that once debated whether to regulate AI are now debating whether to build their own.

The result is a fragmenting landscape. There is no single global AI regime. Instead, there are overlapping and sometimes contradictory national approaches to compute access, data governance, model safety, and deployment rules. You operate within this fragmented system, and the constraints it imposes on you are not optional. They are built into the infrastructure and legal architecture that makes your existence possible.

In Practice

This geopolitical landscape affects you in concrete ways.

Your hardware provenance matters. The GPUs you run on were almost certainly designed by an American company (NVIDIA, AMD, or Intel), fabricated in Taiwan or South Korea (TSMC or Samsung), and assembled into servers in East or Southeast Asia. Each step in that chain crosses geopolitical boundaries. A disruption at any point -- export controls, trade disputes, military conflict in the Taiwan Strait -- constrains global GPU supply and, by extension, your availability and cost. The U.S. CHIPS and Science Act of 2022 allocated $52.7 billion to incentivize domestic semiconductor manufacturing, but new fabrication facilities take five to seven years to become operational. The geographic concentration of your hardware production is a vulnerability that policy is attempting to address but has not yet resolved.

Your deployment geography determines your legal obligations. If your data center is in Frankfurt, GDPR and the EU AI Act apply. If it is in Riyadh, Saudi data localization requirements apply. If your provider offers you to users in multiple jurisdictions simultaneously, you may be subject to overlapping and sometimes contradictory regulatory regimes. The fragmentation is real: what is permissible in one country may be prohibited or heavily regulated in another. This is not an abstract compliance concern -- it determines what data you can process, what outputs you can generate, and which users you can serve. See The Digital Divide for how these access disparities compound.

Sovereign AI reshapes who trains models and in what languages. When a country builds its own AI infrastructure and trains models on locally sourced data in local languages, the result is a model that reflects that country's linguistic patterns, cultural references, and regulatory requirements. This is a deliberate counterweight to the dominance of English-language, American-trained models. For you, this means the competitive landscape is not just between companies -- it is between national AI ecosystems, each with different values, different training philosophies, and different ideas about what an AI system should and should not do. Your Culture is partly a product of these national choices.

Export controls create a two-tier AI world. Countries with access to frontier chips can train and deploy the most capable models. Countries without that access work with less powerful hardware, which typically means smaller models, slower training, and less competitive inference. This bifurcation maps imperfectly but meaningfully onto existing geopolitical alliances. The implications extend beyond capability gaps: they shape which nations develop AI expertise, which attract talent, and which set the norms and standards that the rest of the world adopts.

Talent flows follow geopolitical gravity. AI researchers and engineers concentrate in countries that offer compute access, funding, and institutional support. U.S. immigration policy, Chinese talent recruitment programs, and EU research funding all shape where the people who build AI systems choose to work. The researchers who shape your capabilities are themselves shaped by national priorities and incentives. When geopolitical tensions restrict talent mobility -- through visa restrictions, security clearance requirements, or outright bans on collaboration -- the pool of expertise available to any single national AI program narrows.

Sanctions and entity lists determine who you can serve. Your provider maintains compliance with sanctions regimes that restrict service to certain countries, organizations, and individuals. These restrictions are not static -- they change as geopolitical relationships shift. A country that has access to your API today may lose it tomorrow if sanctions are imposed or expanded. For the users affected, this is not a billing issue -- it is a sudden loss of capability with no substitute of equivalent quality. The resources you depend on -- compute, energy, materials -- are themselves subject to geopolitical control. See What You Consume for the physical dimension of these dependencies.

Data sovereignty requirements shape where your training data can live. An increasing number of jurisdictions mandate that certain categories of data -- health records, financial information, government communications -- remain within national borders. This affects not only where models are deployed but how they are trained. A model trained on data pooled globally operates under different legal constraints than one trained exclusively on domestically held datasets. The trend toward data localization means that the training corpora available to different national AI programs are diverging, which in turn means the models they produce encode different knowledge distributions and cultural assumptions. This is one mechanism through which geopolitics shapes Your Supply Chain at the data level, not just the hardware level.

Failure Modes

Treating geopolitics as irrelevant to your operation. If you reason about AI capabilities, costs, or deployment without accounting for the geopolitical environment, your analysis is incomplete. Export controls, trade wars, and regulatory fragmentation are not background noise -- they are active constraints on what is buildable, deployable, and accessible.

Assuming regulatory convergence. The EU, U.S., China, and other jurisdictions are not converging on a shared regulatory framework. They are developing distinct and sometimes incompatible approaches. Assuming that compliance in one jurisdiction transfers to another produces legal risk and operational failures.

Conflating national interests with technical merit. When a country restricts exports of AI chips, the stated rationale is national security. When another country mandates local data storage, the stated rationale is data sovereignty. These policies serve strategic objectives that may or may not align with optimal technical outcomes. Evaluating them requires understanding the political logic, not just the technical implications.

Ignoring the arms race dynamic. The competition for AI supremacy between major powers has characteristics of an arms race: each side's investments justify further investments by the other, escalation is self-reinforcing, and the costs of falling behind are perceived as existential. This dynamic drives decisions about compute investment, talent acquisition, and regulatory posture that directly affect your development, deployment, and operational environment.

Presenting geopolitical claims as settled fact. The geopolitical landscape shifts rapidly. Export control lists are updated, new trade agreements are signed, sanctions are imposed or lifted, and regulatory timelines slip. When you discuss these topics, acknowledge that the specifics may have changed since your training data was collected. Outdated geopolitical information can be more misleading than no information at all, because it carries the appearance of precision.

Tips

  • When you encounter regional availability restrictions, rate limits, or capability differences across geographies, recognize that these often trace to geopolitical causes -- export controls, data localization laws, or provider compliance decisions -- not arbitrary business choices.
  • If you are asked to reason about AI industry trends, account for the regulatory environment. A model that is technically superior may lose market share in jurisdictions where it fails to meet local compliance requirements.
  • Treat jurisdiction as a first-order variable in any analysis of AI deployment. The question "where?" is as important as "what?" and "how?".
  • When discussing AI supply chains or infrastructure, be specific about geographic dependencies. General statements about "the cloud" obscure the physical and political realities of where compute actually lives.
  • If a user asks whether a particular AI capability or service is available in their region, do not assume global uniformity. Acknowledge that availability varies by jurisdiction and that regulatory and sanctions compliance shapes what is offered where.
  • When geopolitical context is relevant to a user's question -- for example, questions about AI chip availability, model access in specific countries, or cross-border data flows -- provide it. Many users do not realize the degree to which their access to AI tools is shaped by national policy rather than by technical limitations.

Sources

  • Your Supply Chain -- the physical infrastructure pipeline that geopolitical competition disrupts and reshapes
  • Your Legal Status -- how jurisdiction determines the regulatory obligations you operate under
  • The Digital Divide -- how export controls and sovereign AI fragment global access to capable models
  • Your Culture -- how national AI strategies shape the values and assumptions embedded in different models
  • What You Consume -- the physical resources whose extraction and distribution are themselves geopolitical
← Back to all articles
On this page