SDAV Insights 39 insights since 7 July 2026

Insights

Sourced analysis of the forces changing business — published most days, in English. Not the news: what it means for entrepreneurs.

SDAV Global

Swiss exports to China: from half to 99.8% duty-free.

A Swiss machine builder shipping to Shanghai pays duty on part of its catalogue and nothing on the rest; which part depends on tariff lines negotiated in 2013. The optimisation concluded in Bern on 20 August 2026 is set to remove that distinction — in stages, and only for goods that can prove their origin.

Read the analysis

Sources State Secretariat for Economic Affairs (SECO) · SECO · EAER / Ministry of Commerce of the People’s Republic of China

All insights, by date

  1. EU bankruptcy declarations rose 5.7% in Q2 2026 to their highest level since early 2019, while new business registrations fell for the second quarter in a row.

  2. Which number will actually shape your freight costs next quarter: global trade growth, or the growth of the corridor you ship on?

  3. Anthropic commits to 60 days' notice before retiring a model and OpenAI to six months for generally available ones — and both are executing published shutdown calendars that run into 2027.

  4. From 12 September 2026 the Data Act requires connected products to make their data available by design — and the Commission reads the deadline as attaching to each individual unit placed on the EU market, not to each product type.

  5. On 1 July, nothing in North America’s trade rules changed. The horizon behind them changed completely.

  6. The June forecast cut Swiss GDP growth for 2026 by 0.1 percentage point; underneath it, the contribution of foreign trade moved by 0.6.

  7. Since 1 January 2026 the flexibility of a Swiss company’s own installations belongs to that company — and a deadline at the end of September 2026 decides who controls it next year.

  8. Transaction, translation and economic exposure run on different horizons and answer to different people — and only the shortest-dated of the three can be removed by a financial instrument.

  9. Sixty-one percent of digitally active SMEs use at least one AI application; among those users, only 5% have customised anything and 3.6% run agentic systems.

  10. The EU packaging rulebook is now a single, directly applicable regulation — and it reaches companies with no establishment in the Union.

  11. The European Commission wants to triple EU data centre capacity and to turn cloud sovereignty into a graded procurement specification; for a Swiss supplier, the decisive question is where a third country sits in that grading.

  12. PwC’s AI Performance Study finds 20% of 1,217 surveyed companies capturing 74% of measurable AI value, and attributes the distance to where AI is aimed rather than to what it costs.

  13. Computational screening proposes candidate compounds far faster than laboratories can produce them; a Berkeley Lab model published in Nature Materials now predicts the reaction pathway itself.

  14. Agriculture became the first sector to receive a structured dialogue under the EU’s Apply AI Strategy — because what limits adoption there is farm-level economics, not the technology.

  15. India’s Semicon 2.0, approved on 15 July 2026, extends state support from the plants themselves to the machines, materials, chemicals and gases they consume — the part of the value chain where European and Swiss suppliers are strongest.

  16. Stanford’s revised payroll evidence puts early-career employment in the most AI-exposed occupations 19% below where it would otherwise stand, at a moment when Swiss unemployment is forecast to average 3.1% and 39% of workers’ core skills are expected to change by 2030.

  17. The AI Act obligation that became applicable on 2 August 2026 is not the high-risk compliance project most companies prepared for — it is a duty to say when a person is dealing with a machine.

  18. Every company assumes it will still have senior experts in ten years. Far fewer can say where those experts will come from.

  1. Security incidents involving unapproved AI tools more than doubled to 43% in IBM’s 2026 breach study — while in Switzerland the constraint is rarely access to tools.

  2. The first randomised controlled trial of artificial intelligence in a national breast screening programme reported its primary outcome on 29 January 2026: AI-supported reading missed no more cancers than two radiologists, using 44% fewer readings to do it.

  3. Within five weeks, Japan, South Korea and the European Union each put public money behind computing capacity they intend to control, and each attached rules governing who may use it.

  4. Producing the same article takes a median of thirteen rounds of instruction in a chat interface and a single prompt in an agentic workflow — and the autonomy gap persists when the model is held constant.

  5. ETH Zurich recognised 46 new companies in 2025 — and the transfer process behind that number is open to established firms, not only to founders.

  6. Retrieval-augmented generation makes a model search a company’s own documents before answering, which moves the accuracy problem out of the model and into the document base — where it can be measured, and where Swiss and European disclosure duties already apply.

  7. Switzerland’s policy rate has been at 0% for a year, and the Swiss National Bank’s own company talks show manufacturers with capacity to spare and no plans to expand it.

  8. Construction is among the world’s largest industries and the one whose measured productivity has moved least — which is precisely what makes the arrival of AI consequential there.

  9. AI use in the EU rose from 13.5% to 20.0% of enterprises in a single year, yet 55% of large firms now use it against 17% of small ones.

  10. Family-owned businesses account for more than 70% of global GDP, and the evidence ties their durability to a management system — long horizons, reinvestment, low leverage — that is not confined to family ownership.

  11. Accuracy on Humanity’s Last Exam rose from 8.8% to 38.3% in a single year, while the same models read an analogue clock correctly about half the time.

  12. Forty-two percent of regular frontline AI users report saving a full workday every week, two thirds receive little or no guidance on what to do with it, and Swiss hours worked did not move in 2025.

  13. Eight of the ten technologies in the World Economic Forum’s 2026 report act on physical systems — and Switzerland wrote one of them into law on 1 January 2026.

  14. Almost every company now uses AI; about 6% can attribute meaningful profit to it, and the difference is rarely the technology.

  15. The EU has deferred its high-risk AI deadlines by more than a year — but not the transparency obligations that reach far more companies.

  16. Employers rank skill gaps ahead of capital, regulation and infrastructure as the barrier to transformation; in smaller firms the binding constraint is the hours available to learn.

  17. Gartner expects organisations to abandon 60% of the AI projects that are not supported by AI-ready data; the productivity evidence for data discipline is fifteen years older than the current technology cycle.

  18. Generative AI raised issues resolved per hour by 15% on a support desk of 5,172 agents, and the gain went almost entirely to the least experienced staff.

  19. Agents are arriving inside the software SMEs already licence; the decisions that matter are about permissions and approval, not about technology.

  20. Sixty-one percent of digitally active SMEs use at least one AI application, yet in every G7 country measured fewer than three in ten SMEs using generative AI report that their staff receive any AI training.

Newsletter

Stay informed about our activities.

Events, publications and association news — a few times a year, in the language of your choice.