Six out of ten Swiss companies use AI. Only 18 percent are considered advanced users. And just 9 percent have actually made AI the core of their strategy.

These three figures provide a fairly accurate picture of where Switzerland will stand in 2026.

We don't have a problem accessing AI. The technology is available, people are using it, and Switzerland has the research, talent, and capital. What's missing is systematic implementation within companies.

At the same time, AI is already transforming the job market, raising new regulatory questions, and encountering a population that uses it extensively but does not truly trust it.

I'll show you what the current data says about this and what we can infer from it for the next ten years. I deliberately describe these forecasts for what they are: scenarios, not certainties.

First things first: What you should do over the next twelve months

If you only have five minutes, read this section and then stop.

Check whether the EU AI Act applies to you and what your role is. If you sell to the EU or if the output of your AI system is used there, the EU AI Act may apply to you. At the same time, determine whether you are an operator or a provider. Anyone who purchases and uses an existing AI system is usually an operator. Anyone who develops a system and offers it under their own name may be a provider. The obligations differ significantly.
Start with two use cases instead of ten, and define the target in advance. Don’t say, “We want to use AI for quotes,” but rather, for example: “A quote currently takes an average of four hours; in six months, it should take 2.5 hours.” Without a measurable goal, a pilot project can quickly turn into an ongoing project.
Get your data and access rules in order before you buy your next tool. In particular, clarify which company data is allowed in which AI systems. If employees copy customer information, contracts, or quotes into private AI accounts, you have a problem that doesn’t appear in any AI strategy but can still happen every day.
Expect three cost categories. License costs are just one of them. On top of that, there are data preparation costs and the time your staff spends on implementation, training, monitoring, and follow-up work. These two categories, in particular, are systematically underestimated.
Invest in people, not just in software. Before implementation, clarify who will be affected, how these individuals will be involved, and whether training will take place during work hours. If you fall under the EU AI Act, keep in mind that the requirement for adequate AI competence is already in effect.
Continue to hire juniors and apprentices. Entry-level positions, in particular, are under pressure. If companies today fully automate traditional apprenticeship tasks while at the same time training fewer young people, they will lack the next generation of experienced skilled workers in a few years.

Everything else is the reasoning behind it.

Trend 1: Switzerland Uses AI Widely, but Not Deeply

The widespread adoption of AI is a foregone conclusion. The question is no longer whether it will be used, but how.

According to AWS and Strand Partners, 57 percent of Swiss companies use AI. A year earlier, the figure was 46 percent. At the same time, only 18 percent are considered advanced users. That percentage has actually declined slightly because a large number of companies have recently started using AI (AWS/Strand Partners, August 2026).

Based on a survey of approximately 2,500 companies, UBS paints a similar picture: About six out of ten companies use AI, though very few of them do so systematically (UBS Outlook Switzerland, May 2026).

Other surveys highlight the same gap from a different perspective. 47 percent of companies view AI as part of their long-term strategy. Only 9 percent have actually made it a core part of their strategy.

Among people, this trend has progressed further in some cases. According to EY, 89 percent of respondents use AI in their daily work. Accenture estimates that 49 percent of Swiss employees use AI on a daily basis.

In some cases, people are more advanced than organizations.

This also explains why so many companies report both high usage and disappointing pilot projects. The tool is available, but processes, responsibilities, data, and goals are lacking.

A Swiss Adesso survey found that about two-thirds of AI projects do not move into production after the pilot phase.

That doesn't mean the technology doesn't work. It shows that a working model is not necessarily a working process.

In 2026, the bottleneck won't be AI itself so much as data, processes, expertise, and clear goals.

Trend 2: The labor market changes first at the entry level

We need to be cautious when it comes to the job market because several trends are unfolding at the same time.

The KOF at ETH Zurich has examined the extent to which various occupations involve tasks that can be taken over by current language models. Since November 2022, the number of registered job seekers in occupations highly exposed to AI has risen by an average of 27 percent more than in occupations with low exposure to AI.

Among the fields affected are programming, proofreading, web and database development, and journalism (KOF ETH Zurich).

It is important to note the following caveat: This is a correlation, not proof that AI caused this increase.

This same period saw a correction in the technology sector following the boom years, as well as a significantly changed interest rate environment. Both of these developments have affected, in part, exactly the same occupational groups.

This change is particularly evident in entry-level positions. According to jobs.ch, from 2023 to 2025, the share of entry-level positions in AI-exposed occupations was 32 percent below the average for the years 2019 through 2022 (jobs.ch AI Report 2026).

At the same time, demand for AI expertise is on the rise. About 25,000 Swiss job postings require these skills. The demand that is growing particularly strongly is not for AI developers, but for people who can apply AI in their existing jobs (PwC AI Jobs Barometer 2026).

Nevertheless, there is no sign of a large-scale wave of layoffs.

According to EY, 7 percent of companies have cut jobs due to AI, 11 percent have not filled vacant positions, and 18 percent have created new AI-related jobs.

This pattern is therefore more complicated than “AI destroys jobs.”

Certain tasks are becoming less important. Other skills are gaining importance. And, above all, the path to starting a career is changing.

For young professionals, this isn't just an abstract statistic. They may be missing out on the very kind of work through which previous generations learned their trade.

For companies, this raises a second question: How will we gain experience in the future if AI takes over some of the entry-level tasks that were previously performed by new hires?

Trend 3: Agents are the big topic, but not yet part of everyday life

So far, we've mostly been using AI as a tool. We feed it a task and get an answer.

Agent-based systems go one step further. They are given a goal and independently carry out multiple steps, interact with other systems, and make decisions within defined limits.

There is still a big gap between hype and reality.

According to AWS and Strand Partners, 24 percent of Swiss companies have never heard of agent-based AI. Only 4 percent are already fully implementing it.

At the same time, technical skills are advancing rapidly.

METR examines how long tasks can take for an AI system to be able to solve them independently with a 50 percent probability. When viewed over longer periods of time, this so-called "task horizon" increases significantly. METR itself points out, however, that benchmarks are only applicable to real-world companies to a limited extent.

That is exactly what matters.

A model can operate autonomously for increasingly longer periods of time. But if no one in the company knows which process is actually supposed to be followed, it simply automates the confusion.

2026 is therefore less the year of autonomous companies than the year in which companies must make their processes agent-enabled.

If you have clean data, defined access rights, and documented processes, you can start using new capabilities relatively quickly.

If you don't have that, you're asking for trouble—and it'll be on turbo.

Trend 4: Sovereignty Becomes a Location Issue

Switzerland has something that is often underestimated in the current debate on AI: its own research, its own computing infrastructure, and its own models.

ETH Zurich, EPFL, and the CSCS have developed Apertus, a fully open and multilingual language model. It was trained on 15 trillion tokens in more than 1,000 languages, including Swiss German and Romansh (ETH Zurich).

In 2026, Apertus 1.5 was released, featuring additional capabilities for image and audio (CSCS).

Apertus doesn't have to outperform GPT, Gemini, or Claude in every benchmark to be relevant to Switzerland.

The strategic importance lies elsewhere.

Open models foster expertise within the country. They enable research. They reduce dependencies. And they can be used in situations where companies do not want to transfer their data to an external model provider.

Added to this is an exceptionally high concentration of talent. According to the Stanford AI Index, Switzerland ranks first worldwide in the number of AI researchers and developers per 100,000 residents.

Capital is on the move, too. According to the EY Startup Barometer, investment in Swiss AI startups rose from about 345 million to about 1.1 billion Swiss francs in 2025.

Switzerland's problem, therefore, is not a lack of talent.

The key question is how quickly we can translate research and capital into productive applications.

Trend 5: Regulation Won't Wait for a Swiss AI Law

Many companies are still waiting for “the Swiss AI law.”

That's exactly what they shouldn't wait for.

The Federal Council has decided not to regulate AI in Switzerland through a horizontal law modeled after the EU AI Act. Instead, existing laws will be specifically amended, and the Council of Europe’s AI Convention will be implemented.

However, that does not mean that AI is unregulated today.

The Swiss Data Protection Act applies regardless of whether personal data is processed by a human, traditional software, or an AI system. In regulated industries, additional requirements apply, such as those set by FINMA.

The EU AI Act could have an even more immediate impact.

Its impact does not end at the Swiss border. Swiss companies may be subject to it if they offer AI systems in the EU or if the output of those systems is used there.

As of August 2, 2026, transparency requirements will apply to certain AI systems and AI-generated content, among other things. For affected companies, the requirement to have sufficient AI expertise has been in effect since February 2025.

The most important distinction here is often overlooked: Are you a provider or an operator?

Anyone who uses an existing system is usually an operator. Anyone who develops a system or brings it to market under their own name may be a provider. The resulting obligations differ significantly.

For a Swiss SME, this overview is usually more relevant than an 80-page summary of the AI Act.

Trend 6: We use AI, but at the same time we don't trust it

That may be the most interesting contradiction in the Swiss data.

Usage is growing rapidly. Trust isn't keeping pace.

According to the 2026 Digital Barometer, 48 percent of the German-speaking Swiss population expects AI to present more opportunities than risks. In French-speaking Switzerland, the figure is 42 percent, and in Ticino, 37 percent.

Other studies paint a similar picture.

In a survey conducted by gfs.bern involving approximately 2,200 people, one of 40 future scenarios scored particularly well: the conscious withdrawal from constant digital presence and AI. Scenarios in which AI has greater control over human decisions were rated significantly more negatively.

This isn't just a matter of communication.

When people are making extensive use of AI while at the same time fearing a loss of control, job loss, or surveillance, this cannot be solved with a better marketing campaign.

The same thing happens in a company.

Anyone who implements AI without discussing its implications for work, oversight, and accountability shouldn’t be surprised if employees use the system but still reject its implementation.

Use does not imply consent.

What this could lead to by 2036

The further into the future we look, the less certain our predictions become.

Nevertheless, four trends are emerging that are particularly relevant for Switzerland.

The Economy: Up to 85 billion francs is a scenario, not a promise

A study by Implement Consulting Group, commissioned by Google and digitalswitzerland, concludes that generative AI could increase Switzerland’s GDP by up to 11 percent within about ten years. This would correspond to additional annual economic output in the range of 80 to 85 billion francs.

That's a huge number.

And it needs to be categorized.

The study was commissioned by a company that stands to benefit significantly from the rapid adoption of AI. Furthermore, any ten-year forecast depends on assumptions regarding technological developments, investments, regulation, and actual usage.

Other economists are much more cautious.

Nobel Prize in Economics laureate Daron Acemoglu, for example, expects a much smaller impact on productivity because only a portion of today's human activities can be automated in a way that makes economic sense.

The difference between the forecasts isn't just a few percentage points. In some cases, they paint completely different pictures of the future.

What is interesting, therefore, is not so much the exact number as the mechanism behind it.

In the Implement study, delaying the rollout by five years drastically reduces the calculated potential.

The reasonable conclusion to draw from this is not that Switzerland will necessarily earn 85 billion francs.

It goes like this: Waiting also comes at a price.

Companies should therefore not try to predict what the situation will be like in 2036. They should invest in 12-month increments and measure what works.

Work and Demographics: AI Is Entering a Labor Market Facing a Labor Shortage

The situation in Switzerland differs from many international debates in one important respect.

We are discussing automation and, at the same time, a structural labor shortage.

Demographic change means that large cohorts will be leaving the workforce in the coming years. This is particularly evident in the healthcare sector, but other industries are already struggling to find skilled workers.

This creates a seemingly contradictory situation.

In some areas, AI reduces the need for certain tasks. At the same time, companies need automation because they can no longer fill open positions.

The question for the next ten years is therefore not a simple one: Which jobs will AI replace?

The better question is: What work should people still do themselves in the future, and where do we need AI so that we can even manage to get our work done?

The transition into the workforce remains particularly challenging.

When routine tasks are automated, learning opportunities disappear as well. Companies must therefore organize skills development more deliberately than they have in the past.

Infrastructure and Energy: AI Is Getting Cheaper, but It's Not Free

The cost of comparable model performance has been falling dramatically for years.

This suggests that AI is being integrated into more and more software, and that many applications currently marketed as standalone AI projects will become standard product features in just a few years.

For companies, this means that long-term dependence on individual models or providers is risky.

At the same time, the absolute demand for computing power is increasing.

According to a study by the Swiss Federal Office of Energy, Swiss data centers consumed approximately 2.1 terawatt-hours of electricity in 2024. By 2030, that figure could rise to approximately 3.5 terawatt-hours under the most extreme scenario analyzed.

The same principle applies here: efficiency and fuel consumption evolve in tandem.

Individual calculations become more efficient. However, if we perform a hundred times as many of them, overall energy consumption may still increase.

Energy, computing capacity, and digital sovereignty are therefore also becoming relevant for companies that do not operate their own data centers.

Society and Education: The Skills Gap Determines Distribution

Technological progress is not automatically distributed equitably.

People with strong digital skills can adopt AI sooner and increase their productivity. Companies with good data and established processes see benefits faster. Large companies can invest more than small ones.

This creates the risk that a surge in productivity will simultaneously give rise to a new skills gap.

Continuing education is therefore particularly important.

PwC shows that skill requirements are changing much more rapidly in jobs that are highly exposed to AI. At the same time, employers are indicating a growing willingness to pay for AI skills.

However, these figures say little about the people who are already with the company.

Someone who has been working in a clerical, production, or administrative role for twenty years does not automatically benefit from a newly posted job opening that requires AI skills.

The main task, therefore, is not recruitment.

It is the professional development of the existing workforce.

What this means for the people in the company

Implementing AI is not just an IT project.

People need to know why a system is being implemented, how it will affect their work, and what happens to the data it generates.

Stakeholder involvement should therefore begin before the pilot project, rather than only after the tool has already been purchased.

Continuing education must take place during working hours. Anyone who tells their employees they need to get up to speed on AI but shifts the time required for this to their personal time has not implemented a continuing education program.

And not everyone needs the same thing.

An IT professional who uses AI on a daily basis needs a different training program than someone who has been working in logistics or administrative support for twenty years. A single half-day course for the entire company does not solve this problem.

On top of that, there's a line we should discuss early on: surveillance.

AI systems that support work processes often simultaneously generate data about the people who work with them. Companies should therefore clearly define which types of analysis are permitted and which are not.

Trust isn't built by telling employees that they have to trust AI.

It results from clear rules.

My thesis: Switzerland doesn't have an AI problem; rather, it faces a challenge in building expertise.

When I put the numbers together, I get a pretty clear picture.

Switzerland has the technology.

It has outstanding universities, its own infrastructure, capital, businesses, and a population that has embraced AI at an astonishing rate.

What's still missing is the translation into competencies and processes.

That is exactly why I consider AI literacy to be one of the key priorities for our region in the coming years.

Through swissAI, we are calling for our education system to adapt more quickly to this reality as well. Not because everyone has to learn to code or become an AI expert.

Rather, it is because almost everyone needs to learn how to work with AI, evaluate its results, and understand when it should and should not be used.

The same applies to companies.

You don't need a hundred new AI tools.

You need people who understand how to make good use of what's already available.

100 hours of genuine engagement with AI over the course of a year amounts to about 16 minutes per workday.

That's a modest investment for a technology that will likely transform a significant portion of our work.

Conclusion

Switzerland will have no problem accessing AI in 2026.

About six out of ten companies use it. Only 18 percent are at an advanced stage. And only 9 percent have truly integrated AI into their core strategy.

The actual task lies exactly between these three numbers.

Anyone who wants to shape the coming years should therefore spend less time thinking about which model will be released next week.

What matters more are clean data, measurable use cases, clear rules, and people who understand how to use the technology.

No one knows today whether this will actually result in 85 billion Swiss francs in additional economic output by 2036.

It is much easier to answer the question of whether we are failing to realize a significant portion of our potential due to a lack of expertise.

Where does your company stand today: at the stage of utilization, systematic application, or has it already been strategically embedded?