Switzerland’s uphill climb to AI sovereignty  

Switzerland alone will not be able to free itself from its dependence on Big Tech, according to critics. Keystone

Switzerland has invested millions to reduce its dependence on American tech giants. But for many, true sovereignty in artificial intelligence (AI) remains out of reach. 

Artificial intelligence powers global services ranging from translation tools to cloud computing. Yet most of this technology remains in the hands of a few foreign companies, mainly in the United States and China. This has raised fears that political or commercial interests could one day restrict access to essential digital tools.  
 
“Governments have realised that they have to invest in sovereign technologies if they don’t want to be hostage to a small circle of foreign interests,” says David Shrier, a professor of AI and innovation at Imperial College London and an expert on sovereignty in AI. 

For its part, Switzerland has invested millions in IT infrastructure and large language models (LLMs). This has led to the construction of the supercomputer “Alps”, the eighth-most powerful in the worldExternal link, and the development of the LLM Apertus, launched in September. 

Incidents such as Amazon’s cloud services outage in OctoberExternal link, which knocked out several websites, have increased calls to limit dependence on foreign technologies. And recently, the chief of the Swiss Armed Forces, Thomas Süssli, criticised the use of Microsoft within the army over security concerns, in a letter to the government made public by the online newspaper RepublikExternal link. 

There are signs the Swiss population is also concerned about handing too much data to Big Tech. The close result in a national vote on whether to introduce an electronic ID in Switzerland reflected widespread concerns over digital privacy.

Some argue that technological independence is the answer. But with computing power, chips, and data dominated by a few global powers and doubts over the reliability of locally developed technologies, many see true AI sovereignty as a distant prospect. 

US and China control AI 

The US has taken concrete steps to control who can access AI technology. In January, an executive order by then-President Joe Biden narrowed the list of countries that can import cutting-edge AI chips, mainly produced by the US company Nvidia. This dealt a blow to Swiss AI research and development. Nvidia’s chips are used in 90% of the world’s AI models, including Switzerland’s Apertus. The current US administration under  Donald Trump has expanded government control over chips by acquiring 10% of IntelExternal link, one of the largest US semiconductor manufacturers.  

China, meanwhile, is imposing its technological control through political and ideological censorship of algorithms. For instance, the Chinese LLM DeepSeek refuses to answer questions about politically sensitive events, such as the 1989 Tiananmen Square protests. 

More ‘national AI’ as a response to ‘digital colonialism’

As the two superpowers consolidate control, other countries are trying to build alternatives. EU leaders discussed digital sovereigntyExternal link at their summit at the end of October, while politicians in the United Kingdom and Canada have signaled an interest in more home-controlled tech. 

Singapore has taken an early lead in such efforts. The country has earmarked $1.6 billion (CHF1.3 billion) in public funds for AI projects, including Sea-LionExternal link, which aims to train LLMs in 11 regional languages under-represented in US and Chinese models. Switzerland’s Apertus shares a similar goal of inclusivity. Trained on more than 1,000 languages and dialects, it is Europe’s first fully public and open LLM. This means that anyone can download it and analyse and adapt the code and architecture.  

Shrier believes this can counteract the “cultural monopoly” of chatbots such as ChatGPT. A Harvard studyExternal link found that OpenAI’s LLM tends to reflect the mentality of rich, educated and democratic people from industrialised Western nations. The result is a chatbot that thinks like a 25-year-old Silicon Valley software engineer. Shrier likens its dominance around the world to “digital colonialism.”  

Insufficient Swiss investment in sovereign AI

Compared to Singapore, the Swiss government’s spending on sovereign AI is low: around CHF100 million for the Alps supercomputer and another CHF20 million through 2028 for the Swiss AI initiative, which includes Apertus. These projects need about CHF10 million more per year to pay for management and energy consumption.  

“Much bigger investments would be needed” to reach sovereignty, says Marcel Salathé, co-director of the Centre for AI at the Swiss Federal Institute of Technology Lausanne (EPFL). 

Alexander Ilic, head of the federal technology institute ETH Zurich’s AI Centre, agrees that Switzerland will have to increase spending on AI to remain competitive. “In the next ten years we will probably need 20 to 100 times more computing power than we have today,” he says. The European Union and the US have already announced multi-billion dollar plans to build more powerful computing centres. Meanwhile, US partnerships with Nvidia and OpenAI are valued in the trillionsExternal link.  

Salathé says more public funding would also attract more private funding. This is what happened in the 1950s and 1960s in Silicon Valley. “Instead, Switzerland is cutting research and innovation budgets,” he points out, referring to the cost-saving measures contained in the government’s savings package, called EP27. 

Sovereignty in AI creates other dependencies 

Another problem is that building the machines to power AI models requires chips, power and data, which aren’t always available locally. True sovereignty would mean having control at every level of the supply chain. “This is completely impossible for a small country like Switzerland,” Salathé says.                      

The Swiss government is not currently attempting to reach this degree of sovereignty and takes a rather cautious stance. A spokesperson for the Federal Chancellery writes in an email to Swissinfo that building sovereign technologies requires ongoing spending not only on infrastructure, but also on the staff needed to maintain and update it. This, the Chancellery notes, raises concerns about the long-term sustainability of sovereign AI. Meanwhile, open-source models such as Apertus don’t offer additional stability, the spokesperson writes, since they are maintained and updated on a voluntary basis by a global community of developers and IT experts, rather than by a single company or institution.

Unity and strength 

The international scientific community is aware of the limitations to tech sovereignty and is working on ways to join forces.  

Shrier and his colleague Aldo Faisal of Imperial College London have proposed a “federation of sovereign AIs”. It is a network of national AI systems that collaborate by sharing data, infrastructure, and research to reduce costs and improve performance. “Currently only the US and China are able to build AI systems from scratch. All other countries have to make compromises, including Switzerland,” says Faisal, a professor of AI and neuroscience. 

A group of researchers has launched a similar idea in the UK. It takes inspiration from the success of the company Airbus, which was created by a consortium of European states in the 1960s to counter the dominance of the US aerospace industry.  

The “Airbus for AI” initiative borrows this collaborative model, encouraging countries to achieve AI sovereignty by sharing efforts and forming alliances. “No country can compete with Big Tech alone,” says computer scientist Joshua Tan, one of the initiators. “The only way to do that is to unite.” 

Edited by Gabe Bullard/gw