Keywords: AI Geopolitics · LLM Governance · Tech Diplomacy · AI Dependency Risks · Auditability Asymetry
🔑 Key Takeaway
Generality: LLMs are not just tools—they are general-purpose infrastructures embedded in decision-making across society (healthcare, governance, education).
Concentration: Fewer than 10 organizations (mostly U.S. or Chinese) control LLM production, turning AI into a geopolitical leverage point.
Dependence: Adopting an LLM means entering a structural relationship with its producer—including their data, biases, and regulatory context.
There is one word that keeps cropping up in discussions about artificial intelligence, often without being clearly defined: geopolitics. We hear about “geopolitical competition,” “geopolitical risks,” and the “geopolitical stakes of LLMs.” But what does this term actually mean in the context of AI—and why does it apply so clearly to AI today?
This brief attempts to answer that question. Understanding what “geopolitics” means in this context is crucial because it reveals what is truly at stake.
1. Geopolitics: A Working Definition
Geopolitics—a concept with multiple definitions—studies the relationship between political power and geographic space1. Historically, it sought to understand how control over territories, trade routes, and resources shapes power dynamics between states. Foundational thinkers like Friedrich Ratzel (late 19th century)2 and Halford Mackinder (early 20th century)3 framed control over physical space (e.g., Eurasia’s plains, strategic straits) as a key dimension of global dominance.
Over time, this framework expanded beyond physical geography. Economic space (who controls supply chains?), informational space (who controls networks, data, and narratives?), and technological space (who controls critical infrastructures?) have become equally strategic. The geopolitics of AI fits into this broader evolution. It asks a concrete, measurable question: Who controls the AI infrastructures (hardware, software, research, innovation) that underpin economic, military, institutional, and social decisions for a growing share of the planet?
2. How AI Became a Geopolitical Issue
AI was not always geopolitical. For decades, it was primarily an academic and industrial concern, accessible only to actors prepared to develop, adopt, or use the technology. The “race for AI” was a matter of economic competitiveness—important, but not fundamentally different from competition in the automotive or pharmaceutical sectors.
Three shifts changed this.
→Generality
Early AI systems were task-specific: recommendation algorithms, image recognition tools. Large Language Models (LLMs) are general-purpose. They process language in all its diversity, embedding themselves into nearly every human activity4: writing, analyzing, deciding, communicating, healing, educating, governing. A general tool that integrates into decision-making processes at every level of society—accessible without prerequisites (no prior training, expertise, or even technical literacy)—is no longer just a product. It becomes infrastructure.
→Concentration
This infrastructure is produced by a handful of actors. Globally, fewer than ten organizations can train cutting-edge LLMs: OpenAI, Anthropic, and Google DeepMind (U.S.); Baidu, Alibaba, and DeepSeek (China); Mistral (France). U.S. private investment in AI reached $109 billion in 2024, compared to $9.3 billion in China—but the gap is narrowing5. DeepSeek’s R1 model demonstrated in January 2025 that radical efficiency can disrupt capital-intensive advantages in weeks6. The concentration of AI infrastructure production among a few entities—each tied to one of two worldviews (U.S. vs. China)—is what makes the issue geopolitical.
→Dependence
Adopting a market-ready LLM is not like purchasing software. It means entering a structural relationship with the model’s producer—its training data, biases, and national regulatory context. An African hospital using a U.S. LLM for diagnostic support, a Southeast Asian government deploying a Chinese model for public services, or a French SME outsourcing legal monitoring to a tool whose training data and biases it cannot audit: all become informationally dependent in ways that transcend commercial contracts. This dependence has geographic and political dimensions.
3. Concrete Implications
→ The Rise of AI Diplomacy
When China offers its models at competitive prices to neighboring countries or African partners, it is not just a commercial strategy. It is a diplomatic alignment tool, akin to the choice of 5G infrastructure7. The U.S. AI Action Plan (July 2025) explicitly aims to “counter Chinese influence in international governance bodies”8. This translates into:
Withdrawing from or refusing to sign multilateral frameworks deemed favorable to China.
Diplomatic investments in Africa to balance Chinese influence.
Leveraging positions in international standardization bodies to promote U.S. values.
Using export controls on AI chips as a foreign policy tool to limit China’s access to the hardware needed to develop models.
AI diplomacy operates simultaneously at three interconnected levels: software (models and standards), hardware (memory chips, processors), and mineral resources (export controls, diplomatic agreements with producer countries).
→ An Auditability Divide
Model producers (in theory) know how their systems work, why they make certain decisions, and what their limitations are. Most users do not. This “interpretability asymmetry” is starkest in regions lacking AI research infrastructure. Africa hosts less than 1% of the world’s data centers9, and the LLMs used on the continent are not trained on African data. The dependence is real, measurable, and rarely addressed in governance forums.
→ Strategic Instability
When high-stakes decisions—military, medical, judicial, financial—are prepared using unauditable systems, errors become harder to detect, correct, and attribute. In a context of great-power competition, this opacity amplifies uncertainty and creates conditions for escalation through misunderstanding.
Why This Changes What We Should Demand from AI Governance?
If AI is a geopolitical issue, then AI governance cannot be framed solely as a question of regulating technological products—or decided by just two countries (or blocs) segmenting the world along two visions. It must be understood as a question of structuring power relations—between states, companies, and regions—in access to, production of, control over, and auditing of the systems that increasingly shape our collective decisions.
Standards for LLM interpretability and auditability cannot be defined exclusively by the U.S., E.U., and China. Other countries must be integrated into these processes—not as passive beneficiaries of governance designed elsewhere, but as actors whose specific constraints and vulnerabilities are constitutive of the problem.
The geopolitics of AI is not just playing out between Washington, Brussels, and Beijing. It is also unfolding in Rabat, Nairobi, Hanoi, Bogotá, and all others countries and cities —everywhere operators deploy systems they cannot question, in frameworks that do not give them the tools to do so.
REFERENCES
1 Lacoste, Y. (dir.). (1993). Dictionnaire de géopolitique. Flammarion.
2 Ratzel, F. (1897). Politische Geographie oder die Geographie der Staaten, des Verkehres und des Krieges. R. Oldenbourg. [Partial Fr. trans.: Géographie politique, Economica, 1988]
3 Mackinder, H. J. (1904). The geographical pivot of history. The Geographical Journal, 23(4), 421–437. https://doi.org/10.2307/1775498
4 Bommasani, R., et al. (2021). On the opportunities and risks of foundation models. arXiv:2108.07258. Stanford CRFM. https://doi.org/10.48550/arXiv.2108.07258
5 Maslej, N., et al. (2025). Artificial Intelligence Index Report 2025. Stanford HAI. https://hai.stanford.edu/ai-index/2025-ai-index-report
6 DeepSeek-AI, Guo, D., et al. (2025, January 22). DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning. arXiv:2501.12948. https://doi.org/10.48550/arXiv.2501.12948
7 Dugoin-Clément, C. (2020, October). Les enjeux géopolitiques de la 5G. The Conversation. https://theconversation.com/les-enjeux-geopolitiques-de-la-5g-146494
8 The White House. (July 2025). America’s AI Action Plan. https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf
9 UNESCO. (2021). Recommendation on the Ethics of Artificial Intelligence. https://unesdoc.unesco.org/ark:/48223/pf0000380455
About the Author
Ikram Chraibi Kaadoud holds a PhD in computer science and cognitive science (University of Bordeaux/INRIA, FRANCE, 2018). She has contributed to several projects at the intersection of industry and academia, including a private research laboratory and the management of a European project on the responsible deployment of trustworthy AI with 13 institutional partners (EDIH) and more than 100 manufacturing companies. Her research focuses on the intersection of interpretability and explainability of AI systems, human cognitive factors in AI adoption, and responsible governance frameworks. Her scientific work has been published in Neural Networks, Knowledge-Based Systems, and Scientific Reports. She is a regular contributor to the Blog Binaire (La Recherche) on AI and societal issues, and is a WomenTechMaker Ambassador.