AI trends 2026: Between sovereignty, agent economy and regulatory tipping point

Artificial intelligence is developing rapidly - technologically, economically and geopolitically. While companies are increasingly relying on AI, questions about regulation, location attractiveness and technological independence are also becoming more acute. Against this backdrop, EY Switzerland identifies key AI trends that will shape 2026 and which will be discussed in depth on March 24 at the EY National AI Conference 2026 will be presented in Zurich.

The agent revolution: powerful, but not quite ready yet

AI agents are no longer experiments. According to EY's CEO Outlook 2026, 97% of CEOs report that their AI initiatives are meeting or exceeding expectations. The change is most evident in software development: by the end of 2025, almost half of the program codes on the Github platform had already been written with the help of AI tools. AI agents write tests, fix errors and perform complex tasks for hours or days without human intervention.

However, what looks convincing in a demonstration is not the same as reliable operation in a productive environment. The output of AI agents based on large language models (LLM) can vary greatly for the same tasks. If several AI agents are integrated into a process, any errors are multiplied, which makes the development of complex multi-agent systems difficult.

These challenges can be met with a structured orchestration of AI agents. For most companies, the greatest effort is not so much in the AI model itself, but in the accompanying requirements such as security, test environments, incident management and stable deployment processes. The productive use of AI agents therefore requires organizational and technical structures that are familiar from software development - with clearly separated test and production environments as well as development processes. While the technology is constantly evolving, many companies still lack the operational experience to operate such systems reliably and securely.

Sovereign AI: Technological independence is becoming a location factor

«Sovereign AI» describes the efforts of countries and economic regions to develop their own AI models, data infrastructures and computing capacities in order to reduce dependencies on foreign providers.

Adrian Ott, Chief AI Officer at EY Switzerland, says: «Today, a large proportion of Swiss companies that use AI productively rely on models from major providers in the US. The artificial intelligence that supports business decisions in Switzerland has therefore often been trained, regulated and operated outside of Switzerland.»

What happens if access to the most powerful AI models is suddenly withheld for political reasons? Although there are some European and Swiss models, they currently lag behind the leading models from the USA for particularly demanding applications. In addition, Europe currently only has around 5 to 10 percent of the global AI computing capacity for the development of new models, compared to around 60 to 75 percent in the USA.

According to the Global Economic Outlook 2026, trade policy interventions and geopolitical uncertainties remain among the key risks to growth. Digital sovereignty is of crucial importance for Switzerland. For companies, this brings the following key questions to the fore: Where is data processed? Who controls the underlying models? And how robust is their own AI architecture in the face of geopolitical and regulatory changes?

Physical AI: The silent revolution on the factory floor

While the public debate focuses on chatbots and text-based applications, physical AI is becoming increasingly important and is one of the major AI trends of 2026. Physical AI refers to the use of AI to control robots and machines.

A key driver of this development is the training of robots in fully simulated digital environments. With the help of so-called digital twins, millions of hours of work processes are mapped in virtual, simulation-based environments in which systems learn how to deal with irregularities and unexpected situations before this knowledge is transferred to the physical world.

The question for companies is therefore not so much whether physical AI will change manufacturing and logistics, but how quickly this development will progress and who is prepared for it and how well. Companies that are already using robots controlled by physical AI are collecting valuable operating data and information on sources of error, tolerances and borderline cases. As the technology matures, this experience can develop into a significant competitive advantage.

After all, a robot that has been in industrial use for more than a thousand hours is not only a machine worker, but also a supplier of an extensive training data set. Competitors without direct access to real production environments can hardly replicate this.

Adrian Ott says: «Physical AI currently receives less public attention than generative AI, but it will permanently change the largest and most important industries and therefore also the global economy.»

Adaptive AI strategy: strategy, governance and regulation are becoming differentiating factors

Although most companies are already working intensively with artificial intelligence, they often still lack the organizational foundations to be able to use it effectively. Experience from numerous EY projects across various industries shows that the biggest challenges lie less in the technology itself than in the structure: a lack of AI strategies, unclear responsibilities and governance frameworks designed for a slower pace of innovation.

At the same time, the regulatory environment is evolving rapidly. Governments are introducing new frameworks or clarifying existing laws and regulations to address issues of transparency, liability and data protection. Switzerland is pursuing a sector-specific approach with a focus on international connectivity and is attempting to reconcile innovation with legal certainty. For companies operating across borders, this results in an increasingly complex set of rules that varies depending on the market and sector.

In the Swiss financial sector, for example, this dynamic is already clearly visible. According to the EY Banking Barometer 2026, 78% of banks are actively working on the introduction of AI. In the previous year, this figure was still around half of the institutions at 53%. At the same time, data protection, regulatory requirements and operational resilience remain key challenges.

The difference is not so much whether companies use AI, but whether they have the necessary governance structures in place to operate AI in a responsible, scalable and adaptable manner. This includes clearly defined roles and responsibilities for AI decisions, robust control mechanisms that can keep up with the speed at which models and use cases evolve, and transparent decisions on data and model architectures that enable both internal oversight and regulatory scrutiny.

Ott says: «AI governance takes on the crucial task of anticipating regulatory developments at an early stage and continuously adapting internal processes before they become a risk.»

The AI Trends 2026 will be presented and discussed in depth in keynote presentations, business panels and breakout sessions at the EY National AI Conference 2026 on March 24 at Stage One in Oerlikon.

www.ey.com