AI in Swiss SMEs: who gains, who loses, and what gets dropped

Every conversation about new companies now turns to AI, so I asked a narrower question that the data can actually answer: since ChatGPT, has AI changed which Swiss companies get founded, and could it affect how many survive?

The direct effect of AI on five-year survival cannot be proven yet. Companies founded in 2023, the first full year after ChatGPT, reach year five around 2030. What follows is strong circumstantial evidence, not proof.

Fast money, slow adoption

Money and attention have moved quickly. Startups working on AI went from about 5% to about 32% of Swiss venture capital in two years.¹ Inside ordinary companies, the change is slower. The share of Swiss SMEs consciously using AI rose from 22% to 34% between 2024 and 2025,² but only about 2% of micro-firms use it systematically.³

What the founding data shows

I compared new company foundings by industry in 2023 with 2022, and grouped industries by how exposed their work is to generative AI, using the International Labour Organization’s 2025 exposure index as a guide.⁴ The grouping is my own judgement. Switch the chart to closure rates to see the second half of the story.

In industries highly exposed to AI, new companies fell by 1.3%, while in low-exposure industries they rose by 7.1%; the national total was flat.⁵ The split inside the exposed group is the interesting part. Businesses that do work AI can now produce shrank: office support fell 16%, publishing 15%, IT services 7% and legal and accounting 6%. Businesses that advise and sell grew: advertising and market research rose 16% and management consulting 4%. Cheap AI tools make it easy for one person to open such a business.

The catch is that the growing sectors already had the highest closure rates, 9 to 10 closures per 100 companies a year, against 5.1 in health. AI is pulling founders into the most fragile markets.

A snowball, not a single effect

Looking across the evidence, the effects reinforce each other. Capital concentrates in a few AI companies, so non-AI startups get less money; Swiss fintech rounds fell from 41 in 2024 to 30 in 2025.⁶ Large AI models squeeze specialists: even DeepL, a celebrated AI translation company, cut about 250 jobs,⁷ and if the AI specialist is squeezed, traditional translation agencies are squeezed harder. Cheap tools crowd the fragile advisory markets. And inside existing companies, badly chosen AI projects drain money that small firms do not have to spare.

What gets dropped

The evidence on company AI projects is sobering. One MIT study found that 95% of organisations saw no measurable profit-and-loss return from generative AI pilots within six months,⁸ and S&P Global reported that 42% of companies abandoned most of their AI initiatives in 2025, up from 17% a year earlier.⁹ Only 34% of Swiss SMEs have rules for what staff may enter into AI tools, and just 23% of firms with fewer than ten employees.² The root cause is rarely the technology. It is where the money goes, missing data foundations and no clear owner. Part 8 looks at why the success rates reported for AI vary so wildly.

My advice to a small firm is therefore modest. Start with back-office tasks such as invoicing, scheduling and document handling, where returns tend to show up first. Write data rules before anyone pastes customer data into a tool. And keep the first experiments small and reversible, with a result agreed in advance that means stop. The full analysis is in section 7 of the study.

EY, Startup Barometer Switzerland 2026. ↩
AXA and Sotomo SME study, via the federal SME portal. ↩ ↩²
UBS AI survey of Swiss firms, via Netzwoche. ↩
International Labour Organization, Working Paper 140: generative AI and jobs. ↩
Federal Statistical Office (BFS), STAT-TAB company demography by industry. ↩
startupticker.ch and SECA, Swiss Venture Capital Report 2026. ↩
Tages-Anzeiger, DeepL cuts 250 jobs. ↩
MIT NANDA, The GenAI Divide 2025, summary. ↩
S&P Global 2025 figures, compiled. ↩