Proximity

I spend a good part of my life between Switzerland and Iceland.

Both countries are small enough that a new company can be two introductions away from a regulator, a university and somebody who knows where the good coffee is. Both are also small enough that one awkward system can become everybody's problem.

The usual AI map makes places like these look marginal. Frontier models require vast amounts of capital, electricity, chips, data and engineering talent. The United States and China can assemble those resources at a scale that Switzerland and Iceland cannot.

Fine. Training the largest model is one contest. Building trusted businesses on top of increasingly available models is a much larger one.

Small, high-trust places can win there. They can move specialised knowledge into software, test rules with companies rather than at them, and give global customers confidence that somebody remains responsible when the machine gets creative.

Trust helps, but only when it is earned through good systems. A flag, a tidy airport and a respectable banking sector do not qualify.

Small markets force global ambition

A Swiss or Icelandic AI company cannot spend five years selling to its home market and call that scale. The domestic market is too small.

This apparent weakness forces a useful discipline. The product has to travel. The contracts, data controls, language support, pricing and customer service need to work across borders early.

Many founders in large countries can survive for years on local demand. A founder in Reykjavik or Zurich learns quickly that the world does not care where the company was incorporated. The customer in London, Boston or Singapore wants to know whether the product solves a painful problem, works with existing systems and survives a security review.

That pushes small-country companies towards narrow, valuable problems.

Switzerland has deep knowledge in pharmaceuticals, medical devices, finance, insurance, advanced manufacturing and industrial engineering. Iceland has useful operating experience in energy, fisheries, climate, language technology and digital public services. A company does not need to recreate a general model to build a strong product in any of these fields. It needs access to people who understand the field, reliable data, a clear workflow and customers willing to test.

The winning product may look rather unglamorous. A system that checks a medical-device dossier. Software that finds exceptions in an insurance book. A tool that helps an energy operator predict maintenance without sending sensitive data into the open internet.

Good. Boring systems with real customers pay salaries.

Trust is economic infrastructure

AI creates a new cost before it removes an old one. Somebody has to verify the output.

That verification cost becomes serious in law, health, finance, public services and industrial operations. A customer needs evidence about the data, the model, the human review, the failure process and the person who will answer the telephone after something goes wrong.

Small places can make this easier because the distance between institutions is shorter. A university lab, a regulator, an industry specialist and a founder can work on the same test without needing a summit, 14 sponsors and a lanyard.

Switzerland already has many of the ingredients. The IMD World Digital Competitiveness Ranking placed it first in 2025, with particular strength in knowledge and future readiness. Switzerland signed the Council of Europe AI Convention in March 2025. The government is now preparing a consultation draft that keeps cross-sector rules focused on fundamental rights and uses sector-specific amendments elsewhere.

This gives Switzerland room to connect AI oversight to the rules that already govern a hospital, bank or machine manufacturer. Done well, that produces clearer responsibility. Done badly, it produces 19 slightly different interpretations and a consulting festival.

Iceland has a different opportunity. More than 90 per cent of its population uses digital identity, according to the OECD, and its public services score well for user-centred design and proactive delivery. A country of this size can test a service across the whole system and see the consequences quickly.

Yet the OECD's 2026 review also found weak points in data governance, open data and AI oversight. That is useful evidence. Trust cannot repair data that departments cannot connect or a decision nobody can explain.

High trust does not mean automatic trust

The phrase “high-trust country” can make governments and founders a little too pleased with themselves.

The OECD's 2025 survey found that more than 40 per cent of people in Iceland expressed low confidence in how government might use AI. Switzerland, by contrast, had one of the more positive results. The wider finding matters more: people in high-trust countries do not automatically trust public-sector AI.

They look for the same things everybody else looks for. Will my data be protected? Can I understand the decision? Does a person remain in charge? Can I challenge it?

Years ago, I wrote about Iceland's Kennitala system. It was elegant for Icelanders and remarkably awkward for me as a foreigner. I existed in the Alien Registry, yet basic services still treated me like a polite administrative ghost.

That experience is dated, and the systems may have changed. The design lesson has not. A trusted system can work beautifully for the default user while blocking the person at the edge.

AI will find every edge. Foreign founders. Minority languages. Unusual medical cases. Small companies without compliance teams. People whose lives do not resemble the average record in the dataset.

Dear small-country government, a national AI strategy means very little if the international engineer you want to recruit cannot open an account, register a company or move their family without a local guide and three contradictory forms.

Openness has to reach the operating system.

The constraints are real

Small countries have limited talent pools. A few successful companies can absorb much of the available technical workforce. Global firms can pay salaries that local startups struggle to match.

Late-stage capital is another constraint. Switzerland has money, but much of it is trained to avoid exactly the kind of uncertainty that early AI companies create. Iceland's venture market is smaller again. Founders often need foreign investors when the company moves from a promising product to a serious international sales operation.

Compute will remain external for most companies. Cloud platforms, chips and foundation models come largely from abroad. That creates dependency on pricing, access, export controls and product decisions made in another country.

Regulation travels too. Switzerland sits outside the EU, but a Swiss company selling into Europe still has to understand the EU AI Act. Iceland participates in the European Economic Area, and the Act is still under review for incorporation into the EEA Agreement. Geography does not create a legal hiding place.

Then there is the social cost of smallness. Close networks speed up cooperation, but they can also protect incumbents, recycle the same people and make dissent expensive. Everybody knows everybody. Occasionally everybody has invested in everybody too.

Trust without competition becomes comfort.

Build the place as a test environment

The sensible strategy for Switzerland, Iceland and similar countries starts with a few choices.

Pick fields where the country has real operating knowledge. Build evaluation environments around those fields. Give startups access to regulators before launch, with written answers that a future customer can understand. Make public procurement small enough for a young company to enter and serious enough to prove the product.

Create shared infrastructure for model testing, secure data access and incident reporting. Do not spend public money on a national chatbot with a flag in the corner. Help companies prove that their systems work.

Make the country easy for outsiders. Fast visas, functional digital identity, English-language company processes and a banking path for legitimate founders will do more than another innovation conference.

Finally, make regulatory answers portable. When a startup tests a system with a hospital, regulator or public buyer, record the assumptions, results and limits in a form the next customer can use. In a small country, one good pilot should become infrastructure for the next ten companies.

Small countries will not win every layer of AI. They can become very good places to build the layer where powerful systems meet medicine, money, machines and public life.

That requires trust people can inspect, talent that can enter, capital that can take risk and rules that produce evidence rather than paper. Start with those four things.