Geopolitics, AI, and Chile's Strategic Autonomy

1. INTRODUCTION

The central historical question of geopolitics is who controls space. Strategic pathways, maritime routes, and natural resources are some of the elements that define the distribution of power and the freedom of action of states. The current level of advancement in artificial intelligence (AI) adds a new variable to this equation. It is now important to know which state actors control data centers, semiconductors, models, and standards upon which other states depend.

The main powers in AI development, the United States and China, compete in this area with different emphases. To some extent, each replicates in this new dimension the type of power it already exerts over geographic space. Washington seeks to control critical points in the technological chain, while Beijing expands through diffusion and mass adoption.

This article argues that AI expands the geography of power into a new space of competition, where data, models, computing, standards, and user networks become sources of influence. For countries like Chile, this could affect their strategic autonomy, as seemingly technical decisions may ultimately condition their national security. It also presents opportunities, as the new geography still requires energy, minerals, territory, and routes—areas where Chile has relevant assets.

This analysis first describes the layers of the AI ecosystem; then, the approaches of China and the U.S. in this area; subsequently, the vulnerabilities of our country; and finally, the options for Chile in this new area of geopolitics.

2. THE LAYERS OF THE NEW VARIABLE OF POWER

The AI ecosystem can be understood as a set of overlapping layers, each composed of nodes that represent sources of influence. At the base are energy, water, and critical minerals that support digital infrastructure. Above them is a second layer, composed of data centers; transmission media, such as submarine cables and semiconductors; and computing capacity. In this layer, influence comes from who produces and operates that infrastructure, and in the case of advanced chips, from who controls their export.

The third layer consists of data, where influence comes from the jurisdiction under which it is stored and the access that other states can claim over it. Next are the models, whether closed or open, which create dependency on the provider and its updates. Higher up are the standards and regulations, where influence is exerted by those who set the rules of the game. At the final layer are the users and applications, where network effects increase the cost of switching providers or platforms.

This reading allows us to observe that power in the realm of AI is exercised from specific nodes within these layers and that no state actor controls them all. It also demonstrates that geography still matters, as data centers require territory, energy, and water; and submarine cables traverse specific routes. Therefore, states that lack the capacity to compete at the technological frontier can occupy relevant positions in the lower layers, a point that will be further developed for Chile.

3. TWO APPROACHES TO AI

The approaches of the United States and China regarding AI translate into this dimension the ways in which both powers already exercise their power in the classic geographic realm; an imperfect but useful analogy for understanding each actor’s approach in the previously outlined ecosystem.

United States: Control of Critical Points

The United States maintains its global power through a network of alliances and bases located in regions important to its strategy. Regarding AI, it replicates this logic by dominating the chokepoints of the technological chain. Its companies concentrate on the design of advanced chips, the main clouds, and frontier models. Additionally, since 2022, Washington has used export controls to prevent China from accessing the most advanced semiconductors and equipment. The AI Action Plan of July 2025 formalized this approach in three pillars, aimed at accelerating innovation, building infrastructure, and leading AI diplomacy. Furthermore, this plan seeks to prevent adversaries from accessing computing, as it is a matter of geostrategic competition and national security.[1]

This U.S. containment policy also has an expansion dimension, aimed at having partners adopt U.S. technology before Chinese technology. The aforementioned plan states that the United States must meet global demand by exporting its complete technology package. In this regard, the executive order of July 2025 that created the American AI Exports Program offers partners packages of hardware, data, models, cybersecurity, and applications.[2]

This expansion is also supported by selective alliances around the semiconductor chain, which reproduce the logic of the U.S. security alliance network in the technological realm. The declaration Pax Silica, signed by Japan, South Korea, Singapore, the Netherlands, Israel, the United Arab Emirates, the United Kingdom, and Australia, among other states, seeks to secure the supply chains of chips used in AI models.[3] This initiative, to which Chile recently joined, establishes a hierarchy among allies integrated into U.S. technological infrastructure and clients who only access the market.

Access to U.S. technology thus operates as a bargaining tool, just as the security guarantees it provides to allies under certain conditions. The risk of this approach is that the conditional use of access to that technology may erode the trust of its partners, as they may perceive U.S. policy as more transactional than a stable commitment.

China: Expansion through Diffusion

China has projected its influence in recent decades through trade and infrastructure, with the Belt and Road Initiative as its most visible expression. In AI, it follows a similar path, seeking to expand through mass adoption. In July 2025, it presented its Action Plan for Global AI Governance, aimed at the Global South through digital infrastructure, technology transfer, and participation in international standards.[4] Additionally, in July 2026, it launched the World Cooperation Organization on AI in Shanghai, with 29 founding countries.[5]

The central instrument of the Chinese approach is open models that any company or government can download, adapt, and deploy without restrictive licenses. The results are visible, as during 2025, Chinese models surpassed U.S. models in downloads on Hugging Face, accounting for 41% of the total; and the Alibaba model family has accumulated over 113,000 derived models.[6] In addition, there is a circuit in the manufacturing realm, as open models allow for low-cost AI deployment in factories, logistics, and robotics, whose operational data feed back into the models. This physical circuit largely escapes U.S. export controls, which focus on access to training chips.[7]

The dependency generated by Chinese open models is different from that created by closed services. An open-weight model running on its own infrastructure cannot be suspended from Beijing, reducing the risk of access denial. Its effect is more indirect, as those who build on Chinese models adopt their updates, tools, and the technical ecosystem that develops around them. In this way, China gains influence through use, without the need for states to sign agreements.

The limit of this approach is trust, as China’s influence depends on whether governments, companies, and developers are willing to integrate models designed under Beijing’s interests into their systems. For example, Chinese models incorporate content restrictions required by Beijing’s regulations, resulting in censored or biased responses on sensitive issues for the Communist Party. For this reason, several governments restricted the use of applications like DeepSeek on official devices in 2025, which hinders Chinese penetration precisely in state and sensitive uses.[8]

4. CHILE’S EXPOSURE

In the face of competition between both powers, middle countries like Chile face a dilemma between achieving a degree of technological sovereignty and the dependency that results from integrating into external ecosystems. Developing its own capabilities offers greater autonomy but requires resources, scale, and time that cannot be sustained, while adopting external technologies accelerates digital transformation at the cost of new forms of dependency. Therefore, in decisions regarding AI models, clouds, or infrastructure providers, certain aspects must be considered that could affect our strategic autonomy, which can be identified in at least four areas. It can be argued that Chile already faces each of these risks, which could manifest in contexts of crisis or tension between powers, when reversing the decision would be costly.

– Risk of Alignment

Referring to the risk that the choice of technology providers becomes a condition of the bilateral relationship, such that a technical or commercial decision ultimately affects areas such as security cooperation, trade, or access to technology. In Chile, this risk materialized in February 2026, when the United States revoked the visas of three Chilean officials for their role in the Chile-China Express cable from China Mobile, whose concession had been approved and revoked in January.[9]

– Control of Critical Infrastructure

States that build or operate cables and data centers gain surveillance or disruption capacity over essential flows. In June 2026, the Humboldt cable was approved, which will connect Chile with Australia starting in 2027 through a partnership in which Desarrollo País and Google each hold 50%.[10] The project reinforces Chile’s position as a node to the Asia-Pacific, but also concentrates half of the control of a strategic infrastructure in a U.S. company. Furthermore, if Chile had opted for the Chile-China Express cable, that infrastructure would have been owned by China Mobile, a state-owned enterprise subject to Chinese National Intelligence Law, which requires organizations to cooperate with the country’s intelligence services.

– Sovereignty over Sensitive Information

Critical state data hosted by foreign providers is subject to extraterritorial laws and intelligence services. Chile addresses this risk only from the privacy perspective through Law 21719, which regulates international transfers of personal data and whose entry into force is still pending. However, there is no explicit national security criterion for state and defense data hosted in foreign clouds.

– Supply Chain

A state actor can suspend supply or restrict chips, computing, updates, or cloud services precisely when they are most needed. Latam-GPT, the regional model promoted by CENIA and presented in February 2026, illustrates this exposure across multiple layers simultaneously. It was built on Llama 3.1, an open-weight model from the U.S. company Meta, and its first version was trained in the AWS cloud.[11] Additionally, the supercomputer planned at the University of Tarapacá depends on chips whose export is controlled by the United States. Thus, the main regional effort to reduce external dependencies relies on technology from a single country.

5. CONCLUSION: CHILE’S OPTIONS

The question of who controls space remains relevant; although today it has expanded and coexists with another question, which is who controls the layers of the AI ecosystem that other actors depend on. Chile is exposed in the upper layers but possesses relevant assets in the lower layers, which depend on geology, climate, and geography. Our country is the world’s leading producer of copper and the second-largest producer of lithium, and the northern region records some of the highest levels of solar radiation in the world, allowing it to offer clean and low-cost energy to an industry whose electricity consumption is rapidly growing. These advantages have already translated into over USD 7.5 billion in committed investments in data centers by AWS, Microsoft, and Google.[12] Moreover, the Humboldt cable could make the country the first direct digital link between South America and Asia-Pacific.

However, these opportunities would only translate into bargaining power if the state uses them with strategic foresight. Currently, decisions seem to be made in a fragmented manner, project by project, resulting in a drift towards the U.S. ecosystem, reinforced by its recent incorporation into Pax Silica, with clouds, cables, and chips from that origin, in a country whose main trading partner is China. To convert this position into strategic autonomy, four lines of action are proposed, each aimed at mitigating one of the exposure mechanisms described.

– The first line of action addresses the alignment pressure and consists of creating a body that jointly evaluates digital infrastructure decisions. This task could be housed in a National Security Council that would allow for the permanent integration of technological, diplomatic, and defense dimensions. Within this framework, a committee on critical technologies, involving the ministries of Science, Foreign Affairs, Defense, Public Security, and Finance, as well as ANCI, could apply national security criteria to these decisions before the next crisis arises and provide coherence to the country’s position.

– To reduce the risk of control over critical infrastructure, the proposed committee should apply a prior and transparent security review procedure to submarine cables, data centers, and cloud services serving the state, with known criteria for ownership, jurisdiction, and access. This would have allowed the case of the Chinese cable to be resolved with clear rules, rather than in the midst of sanctions.

– In response to the exposure of sensitive information, the state needs a national security criterion for its own data, complementing the privacy approach of Law 21719. This entails classifying public data according to its sensitivity level and requiring that defense, intelligence, and critical infrastructure data be hosted in infrastructure under Chilean jurisdiction or with verifiable contractual guarantees.

– Finally, to decrease the risk of access denial, Chile must diversify its providers while using its position to obtain better conditions. In public procurement, this means requiring portability and multi-cloud architectures; preferring for sensitive uses models that can be deployed on its own infrastructure. Simultaneously, the country can link incentives for data centers, such as access to energy and expedited permits, to countermeasures like reserved computing capacity for national research and state organizations.

ALEJANDRO AMIGO
Senior Researcher, AthenaLab

---

[1] White House, Winning the Race. America’s AI Action Plan, July 2025. https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf

[2] White House, Promoting the Export of the American AI Technology Stack, July 2025. https://www.whitehouse.gov/presidential-actions/2025/07/promoting-the-export-of-the-american-ai-technology-stack/

[3] Real Instituto Elcano, Pax Silica. Alliances, Frontier and Markets in the Geopolitics of the Chip. https://www.realinstitutoelcano.org/en/analyses/pax-silica-alliances-frontier-and-markets-in-the-geopolitics-of-the-chip/

[4] ANSI, China Announces Action Plan for Global AI Governance, August 2025. https://www.ansi.org/standards-news/all-news/8-1-25-china-announces-action-plan-for-global-ai-governance

[5] Government of the People’s Republic of China, Xi Calls for Equitable Global AI Governance, Unveils New Cooperation Body, July 2026. https://english.www.gov.cn/news/202607/18/content_WS6a5c27e1c6d00ca5f9a0c4d8.html

[6] Hugging Face, State of Open Source on Hugging Face. Spring 2026, 2026. https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026

[7] Ngor Luong, USCC, Two Loops. How China’s Open AI Strategy Reinforces Its Industrial Dominance, March 2026. https://www.uscc.gov/sites/default/files/2026-03/Two_Loops–How_Chinas_Open_AI_Strategy_Reinforces_Its_Industrial_Dominance.pdf

[8] Al Jazeera, Which Countries Have Banned DeepSeek and Why?, February 6, 2025. https://www.aljazeera.com/news/2025/2/6/which-countries-have-banned-deepseek-and-why

[9] UPI, Chinese Cable Project Causes Rift Before Chile Presidential Transition, March 2026. https://www.upi.com/Top_News/World-News/2026/03/03/latam-chile-kast-boric-transition-dispute/5301772565134/

[10] Ex-Ante, How the Google Submarine Cable Advances with the State, June 2026. https://www.ex-ante.cl/como-avanza-el-cable-submarino-de-google-con-el-estado-luego-de-que-el-gobierno-dejara-el-proyecto-chino-en-el-congelador/

[11] CENIA, Latam-GPT: The First Open Regional AI Created with Latin American Data, February 10, 2026. https://cenia.cl/2026/02/10/latam-gpt-la-primera-ia-regional-abierta-creada-con-datos-latinoamericanos/

[12] CNN Chile, Amazon, Microsoft, and Google Among Them. Chile Adds Over USD 7.5 Billion in Committed Investments in Data Centers, 2026. https://www.cnnchile.com/negocios/amazon-microsoft-google-chile-suma-mas-us7500-millones-inversiones-comprometidas-data-centers/