The AI Act and its risk tiers
Under the act, AI systems are sorted by risk. Some practices are banned outright. High-risk systems, such as systems for hiring, credit, medical use and critical infrastructure, need conformity assessments, documentation and monitoring. These duties apply from 2 December 2027 for uses such as hiring and credit, and from 2 August 2028 for AI in regulated products such as medical devices. General-purpose models have transparency duties. Above certain compute thresholds, they also have duties for systemic risk. Most other systems only need basic transparency. As a founder, find out your tier, and build the technical file while you build the product.
Compliance and enterprise sales
Large business buyers now check suppliers against the act. A supplier who already has the model documentation, a summary of data governance and the risk classification passes the security review. Also handle your data in the EU (see data residency in the EU). With this file, you close deals faster, while US competitors are still writing theirs.
Compare the Netherlands with Ireland and Germany on tax, substance, running costs and time. See setting up a Dutch subsidiary for a foreign parent company.
What the Netherlands offers an AI company
With a BV, you can sign contracts in 27 markets from inside the EU. With the WBSO, you pay less wage tax on your research (see the WBSO wage credit). Profit from the model and the product can fall under the 9% innovation box (see the innovation box). With the highly skilled migrant permit (kennismigrant), you can hire people from anywhere. See hiring international developers. In Amsterdam, you also find computing power, talent and capital, as described in the financing map.
A provider outside the EU needs an authorised representative in the EU. For this and for the AI Act dates after the 2026 amendment, see the AI Act and the EU representative.