AI Sovereignty: Charting Switzerland’s Path Amid US and Chinese AI Dominance

November 10th, 2026, 09h30-18h30,
Starling Hotel,
1025 Saint-Sulpice

Introduction

The global AI landscape is consolidating around two dominant poles. Global corporate AI investment reached $581.7 billion in 2025 — up 130% year-over-year — while generative AI investment alone surged nearly fivefold to $170.9 billion, with the overwhelming share of this capital concentrated in the United States and China. The capability gap between the two powers is narrowing with equal speed: DeepSeek-R1 matched the leading US model in early 2025, and as of early 2026, the top American model holds only a 2.7% performance advantage over its Chinese counterpart.

For countries outside this duopoly, the consequences are no longer abstract. The first half of 2026 brought two concrete shocks — export controls restricting access to US frontier models for foreign nationals, and the discovery that dramatic falls in per-token inference costs have not translated into predictable or manageable AI spend, with enterprises burning through entire annual AI budgets in months — that together crystallise a fundamental question: how can nations retain meaningful control over a technology increasingly expected to shape economic competitiveness, national security, and societal resilience?

AI sovereignty — the capacity of nations or organisations to govern their own AI technology stack, including data, infrastructure, and models — has consequently moved from the margins of policy debate to its centre. Middle powers that fail to secure influence over the development, deployment, and governance of AI will likely forfeit control over their economies. Switzerland brings distinctive assets to this question: it leads the world in AI researchers and developers per capita, with 110.5 per 100,000 inhabitants, hosts world-leading research institutions, and is home to Apertus, Switzerland’s own sovereign foundation model initiative, Switzerland’s own sovereign foundation model initiative. Geneva further serves as the permanent home of key international organisations actively engaged in AI governance — including the ITU, WHO, and WIPO — and will host the inaugural session of the UN Global Dialogue on AI Governance in July 2026. Whether and how these advantages translate into a credible national AI sovereignty strategy — and what roles industry, civil society, legal experts, and policymakers must play in building one — is precisely what this conference sets out to examine.

This event is organized by the Center for Digital Trust (C4DT), EPFL.

Objectives

  • To examine what AI sovereignty means in practice across multiple dimensions — technical, legal, cultural, economic, and geopolitical — and assess which levers are realistically available to middle powers today.
  • To learn from the strategies of other middle powers — including Singapore, Canada, France, and others — and identify the approaches most transferable to Switzerland’s specific assets and constraints.
  • To diagnose Switzerland’s current position — mapping its distinctive assets, exposing its dependencies, and assessing whether existing policy instruments amount to a strategy or merely a plan.
  • To produce actionable recommendations toward a Swiss national AI sovereignty strategy, identifying the institutional actors, investment commitments, and partnerships needed to give that strategy force.

Discussion Points

The discussion points below are proposed. The confirmed agenda will be published in advance of the event.

1. Setting the Stage: Why AI Sovereignty Matters Now

AI sovereignty has to be approached with five overlapping dimensions: who controls the infrastructure and models (technical), whose law governs the systems a nation depends on (legal), whose values and social norms are encoded in them (cultural), who captures the value AI creates and bears the financial risk (economic), and who has the power to cut off access (geopolitical). The 2026 stress tests — access restrictions and cost volatility — made these dimensions concrete and urgent. This opening session defines the framework and makes the case for why middle powers need to act.

  • What does AI sovereignty mean across its five dimensions, and how do they interact in practice?
  • What is driving the global turn toward sovereign AI strategies, and what are the principal obstacles?
  • What do the 2026 stress tests reveal about the nature and depth of current dependencies?
  • What strategic options are realistically available to middle powers?
  • Where does legitimate sovereignty end and counterproductive insularity begin?

2. Sovereignty Levers Along the AI Technology Stack

Full-stack technical control is not a realistic goal for any middle power. What is achievable is operational leverage — the ability to run inference on domestic infrastructure, audit model behaviour, choose which model sits under your harness, and avoid being cut off by a foreign government or pricing decision made elsewhere. This session examines where that leverage actually lies, across the technical, legal, and cultural dimensions of the stack.

  • The harness — the software layer that orchestrates how models are deployed and used — is increasingly where capability and control reside. Does this mean the race to control frontier models is the wrong race?
  • Can inference sovereignty compensate for the absence of training sovereignty, or does dependence on externally trained models introduce irreducible risks regardless of where inference runs?
  • What security and governance risks arise from running AI workloads on hyperscaler infrastructure, and what can sovereign cloud architectures realistically offer in response?
  • What do different degrees of model openness — from closed proprietary to open-weight to fully open — mean for sovereignty?
  • A Swiss organisation using a US-hosted model is subject to the CLOUD Act. What does this mean for public institutions and regulated industries, and what architectural choices can reduce this exposure?
  • Should Switzerland align with the EU AI Act, develop its own framework, or accept a degree of legal dependency?
  • Alignment is a sovereignty problem, not just a safety problem. If Switzerland relies on foreign-aligned models for public services, healthcare, and legal interpretation, who controls the normative layer of Swiss public life?
  • How can privacy-preserving techniques and data space architectures enable sovereign data sharing without sacrificing scientific openness?

3. Governing Sovereign AI: Instruments, Actors, and Accountability

Operational leverage over the AI stack creates the conditions for sovereignty, but governance determines whether that leverage is actually exercised. This session examines what mechanisms, standards, and institutional structures are needed to make sovereign AI deployment accountable, trustworthy, and capable of keeping pace with technological change — from procurement rules for critical infrastructure to the coordination of actors across federal and cantonal levels.

  • How can Switzerland evaluate the trustworthiness and safety of foundation models before deploying them in critical infrastructure?
  • What technical standards, audit requirements, and procurement rules are needed to ensure accountability over the full AI lifecycle?
  • What governance mechanisms — legal, organisational, technical, market-based, and normative — are available to Switzerland, and how can they be effectively combined?
  • No single actor currently has both the legitimacy and capacity to own Switzerland’s sovereign AI agenda. Which institutional structure is best placed to bring the Federal Council, SERI, SECO, and the cantons together?
  • What treaty frameworks or multilateral agreements could constrain the unilateral use of AI access as a geopolitical lever, and what role can Switzerland play in advancing them?
  • If Swiss firms use US or Chinese AI tools to generate productivity gains, does the economic surplus flow disproportionately abroad — and what policy instruments can address this?
  • Is public trust — not technical capability — the binding constraint on sovereign AI adoption, and what does rebuilding it require?

4. What Middle Powers Are Actually Doing: Lessons for Switzerland

Middle powers are not passive observers of US and Chinese AI dominance. Singapore has earmarked $1.6 billion in public funds for AI projects including multilingual foundation models. Canada has adopted an “AI for All” strategy that treats AI as critical infrastructure on par with energy and defence. France has backed Mistral as a European frontier model challenger. European initiatives including Apertus, Open Euro LLM, GPT-SW3, and Alia are pursuing sovereignty through openness, compliance, and cultural fit rather than raw capability. This session examines what these strategies have in common, where they differ, and what Switzerland can learn from them.

  • What strategic choices have other middle powers made — on compute investment, model development, governance, and international positioning — and how transferable are those choices to Switzerland?
  • What can European and Swiss sovereign LLM initiatives realistically deliver as instruments of sovereign AI strategy, and what are their principal obstacles?
  • Does building EU AI Act compliance into a model’s architecture from the outset create a market position that foreign providers have little commercial incentive to match?
  • show How can European LLM initiatives build genuine normative sovereignty — ensuring models reflect the values of the societies they serve — rather than simply replicating foreign alignment choices in a local language?
  • What role should Switzerland play in the broader European open foundation model ecosystem, and which bilateral and multilateral partnerships best serve its sovereignty interests?

5. Switzerland’s AI Sovereignty Strategy: From Assets to Action

The closing session turns from analysis to action. Switzerland’s distinctive assets — world-leading research institutions, Geneva’s governance ecosystem, a rule-of-law tradition, political neutrality, and Apertus — are real but currently underexploited. This session asks what a credible Swiss sovereign AI strategy would actually look like across all five dimensions, who has the legitimacy to define and own it, and what it would take to move from the current implementation plan to a genuine national strategy.

  • Does Switzerland currently have an AI sovereignty strategy, or merely an implementation plan — and what would it take to close that gap?
  • What are Switzerland’s key differentiators across the five dimensions of AI sovereignty, and how do they translate into strategic options?
  • What should the core pillars of a Swiss national AI strategy be, and why should the Federal Council treat this as a priority?
  • How should Switzerland position itself between the competing demands of European alignment, bilateral partnerships, and its own distinctive neutrality?
  • What investment commitments, institutional decisions, and international partnerships are needed to give a Swiss sovereign AI strategy real force?

Closing

The day will conclude with a synthesis of the key findings across all sessions and a structured discussion of actionable recommendations toward a Swiss national AI sovereignty strategy. Outputs will be shared publicly following the event.