By Daniel Dobos and Prathit Singh
In a rapidly evolving digital landscape, the race for artificial intelligence has become a defining feature of global power politics. Once measured in territorial control and military might, competition between nations is now unfolding in algorithms, computing power, and all components of their fragile global supply chains. As governments scramble to assert “AI sovereignty”, questions of trust, transparency, and control over critical technologies are shaping new geopolitical alignments. Amid this ‘AI race’, which is now in many ways complementary to the arms race, Switzerland has positioned itself differently. By developing Apertus, a fully open-source large language model developed by Swiss research institutions, Switzerland has offered an alternative vision where transparency, ethical governance, and inclusivity form the foundation of a trusted digital future. For International Geneva, such an approach presents an opportunity to redefine how multilateralism adapts to AI.
AI Sovereignty as the New Arms Race
Technology, instead of territory, has become the new domain of geopolitical competition, with great powers competing to win the race for future-defining tools like AI, quantum computing, and biotechnology.
Artificial intelligence has become the latest frontier of geopolitical competition. From Washington to Beijing, world leaders are increasingly framing AI and technology as a strategic resource that could determine global influence in the decades ahead. The rhetoric used by leaders across the world, from Xi Jinping and Donald Trump to Vladimir Putin and Narendra Modi, often echoes the rhetoric of a nuclear arms race, claiming that whoever leads in AI will rule the geopolitical future. Technology, instead of territory, has become the new domain of geopolitical competition, with great powers competing to win the race for future-defining tools like AI, quantum computing, and biotechnology. Many of these leaders also view the advent of artificial general intelligence (AGI) or even “Superintelligence” as a “magic line” that once crossed would grant an overwhelming, unassailable advantage. This all-or-nothing mindset has spurred billions in investment from governments in developing AI models nationally amid fears of falling behind, and is used to justify breaking copyrights, privacy, and other legal or ethical concerns.
The race for AI, therefore, should be less about reaching a singular finish line than shaping the norms that govern it. In this race for developing faster AI models, trust, transparency, and ethical governance should emerge as the real measures of leadership.
Yet this framing misconstrues the nature of technological progress. Unlike nuclear capability, there is no single “magic line” that once crossed guarantees technological dominance. AI development is often incremental and interdependent, built on shared research, global data, and collective innovation. No single “AGI moment” could guarantee countries a perpetual dominance in a field as fast-evolving as AI. And making a step before others does not lead to being able to prevent others from doing the same or very similar steps. Instead, advances in machine learning come as incremental milestones, more akin to the space race than the nuclear race. Being first to launch a satellite or put a human on the moon yielded prestige and temporary leverage, but it never permanently set back other nations. The race for AI, therefore, should be less about reaching a singular finish line than shaping the norms that govern it. In this race for developing faster AI models, trust, transparency, and ethical governance should emerge as the real measures of leadership.
The Race for AI and the Swiss Context
In a landscape dominated by major powers and corporate laboratories, Switzerland has taken a distinct path towards digital sovereignty. Rather than competing for speed or scale, it has focused on values of trust and transparency as the foundations of technological independence. The decision to develop its own large language model grew out of a national reflection on how small but innovative states can retain control over critical digital infrastructure while contributing to global progress.
The initiative emerged from a collaboration between ETH Zurich, École Polytechnique Fédérale de Lausanne (EPFL), and the Swiss National Supercomputing Centre (CSCS), supported by a strategic private-public partnership. Underlying the initiative were discussions towards a unified national effort, recognising that Switzerland’s strength lay in pooling expertise rather than fostering competition between institutions. Timing proved equally strategic. Switzerland began investing in computing infrastructure before the global surge in AI demand, acquiring hardware and expertise at a fraction of the later cost. Yet it launched the project only after the field had matured, integrating lessons from earlier models. This balance of foresight and restraint enabled Switzerland to develop a model that is both advanced and efficient. The goal is not to be the first, the goal is to be the first who do it right.
At the heart of this effort lies a renewed vision of what “open source” means. In the Swiss conception, openness extends beyond access to code. It encompasses transparency, reproducibility, and accountability, and enables anyone to understand how a system is built, what data it was trained on, and how it can be improved.
Apertus: An AI Model With Swiss Values
Launched in September 2025, the Swiss AI model Apertus marks a significant milestone for Switzerland’s digital sovereignty. Developed together with ETH Zurich, EPFL, and the Swiss National Supercomputing Centre (CSCS), it represents an alternative to the dominant approaches to artificial intelligence by incorporating the Swiss values of transparency, neutrality, and inclusivity in its very design.
The model’s defining feature is its transparency: all training data, filtering methods, and architectural choices are publicly documented. Unlike many other “open” models that release only their parameters, Apertus makes its entire development process accessible, allowing others to understand, reproduce, and improve it. This transparency addresses the opacity of AI decision-making systems, one of AI’s core challenges, and offers a foundation for greater accountability in their use.
Apertus was trained on 15 trillion tokens drawn from across the public internet, with careful respect for copyright and data protection. Websites that had opted out of data collection were excluded, and any later retractions were honored, reflecting the model’s alignment with European regulatory standards such as the GDPR and the forthcoming EU AI Act. Underlining Switzerland’s commitment to sustainability in digital innovation, the model was trained entirely on renewable hydroelectric power at the CSCS supercomputer in Lugano, one of the world’s most energy-efficient facilities.
Its technical design also reflects Switzerland’s international outlook. Apertus is multilingual by design, incorporating text from more than a thousand languages, from globally spoken ones to minority and regional rare-spoken languages such as Romansh (the fourth national language of Switzerland, besides German, French, and Italian). This feature enhances accessibility for humanitarian and diplomatic actors operating in multilingual environments, providing an inclusive platform that mirrors the linguistic diversity of International Geneva itself.
The model’s guiding philosophy is not to be the fastest or most powerful, but the most trustworthy and adaptable, giving a true reflection of enduring Swiss values of neutrality, precision, and reliability. In doing so, Apertus transforms Switzerland’s technological investment into a global public good, making it an AI engine designed for shared sovereignty instead of dominance.
Finally, Apertus’s open-source Apache 2.0 license ensures that it can be freely used, adapted, and fine-tuned by others, including public institutions and smaller states seeking to build AI capacity independently. It is available in two sizes and can be downloaded from public platforms. The model’s guiding philosophy is not to be the fastest or most powerful, but the most trustworthy and adaptable, giving a true reflection of enduring Swiss values of neutrality, precision, and reliability. In doing so, Apertus transforms Switzerland’s technological investment into a global public good, making it an AI engine designed for shared sovereignty instead of dominance.
Opportunities for International Geneva
International Geneva offers an ideal environment to demonstrate the public value of open and transparent artificial intelligence models like Apertus. In a city defined by diplomacy, humanitarian response, and multilateral cooperation, Apertus provides both a technical tool and a governance model. Its openness, multilingual range, and capacity for secure and local use make it particularly suited to organisations that handle sensitive information and operate across borders.
Recent experiments illustrate this potential. During a recent hackathon with UNHCR, the Swiss Federal Department of Foreign Affairs, and the ICT4Peace Foundation in Geneva, participants applied Apertus to refugee survey data, reducing analysis time by nearly 80 percent and freeing up staff for higher-value tasks. Because the model can be deployed locally, even on a protected institutional network, such exercises maintain full data sovereignty while handling confidential information. Another prototype tested Apertus to streamline policy and negotiation analysis, using it to compare hundreds of pages of draft texts across national positions in UNFCCC global climate negotiations. The resulting model could identify areas of consensus and divergence and streamline discussions on unresolved points.
Apertus also offers a path towards greater linguistic and cultural inclusivity. Its training on over a thousand languages allows organisations to process, translate, and generate content in underrepresented languages, making it a significant tool for agencies working with communities whose voices are often excluded from digital systems. For International Geneva, this also opens the door to more inclusive and representative forms of digital participation.
If adopted widely, Apertus could serve as the digital backbone for the international system and lay the foundation for a shared, neutral AI resource that all stakeholders can rely on for a new ecosystem of open-source AI tools tailored to multilateral needs, ranging from humanitarian coordination and peacebuilding to environmental governance.
Beyond these applications, the model’s greatest promise lies in what it symbolises: a shared digital infrastructure built on neutrality, transparency, and trust. If adopted widely, Apertus could serve as the digital backbone for the international system and lay the foundation for a shared, neutral AI resource that all stakeholders can rely on for a new ecosystem of open-source AI tools tailored to multilateral needs, ranging from humanitarian coordination and peacebuilding to environmental governance. In doing so, it could also help define a distinct model of digital governance, where technology is treated as a public good rather than a proprietary asset. Switzerland’s approach demonstrates that sovereignty in the digital age need not rely on scale or secrecy. By prioritising openness and accountability, Switzerland has positioned itself to lead by example, demonstrating how cooperation, instead of competition, can shape the ethical foundations of global AI.
About the Authors
Daniel Dobos is Research Director at Swisscom, Co-Chair of the ITU AI For Good Impact Initiative, and chair of the Swiss AI Standardisation Commission.
Prathit Singh is the Project Coordinator of the Geneva Policy Outlook.
All publications of the Geneva Policy Outlook 2026 are personal contributions from the authors and do not necessarily reflect the views of the institutions they represent, nor the views of the Republic and State of Geneva, the City of Geneva, the Fondation pour Genève, and Geneva Graduate Institute.
