On 6 November 2026 the Dutch cabinet approved its AI vision, Kiezen voor AI voor iedereen (Choosing AI for Everyone), on a proposal from State Secretary Willemijn Aerdts, who handles Digital Economy and Sovereignty. According to the Rijksoverheid announcement, she coordinates the approach while individual ministries apply it in their own fields. The vision aims to let the Netherlands make its own choices on AI and reduce reliance on foreign tech giants. This article assesses it on one question: does it give Dutch firms and public bodies more room to build and adopt AI, or just more paperwork?
The strongest case for the strategy
The case for the vision is serious. Most of the compute, models and cloud capacity that Dutch hospitals, schools and ministries rely on belongs to a handful of non-European firms. If a supplier changes its terms or a geopolitical relationship sours, a country with no alternatives has little leverage. DutchNews reported that the 28-page letter to parliament seeks partners beyond the US and China. It names Canada, Japan, Singapore and South Korea as "middle powers" to work with, alongside other EU member states. It also calls for "stable and predictable relations" with Washington and Beijing, and notes that the Netherlands has joined the US-led Pax Silica partnership on chips, computing power and AI infrastructure. Diversifying partners is a sensible hedge and does not amount to protectionism.
What is concrete and what is not
The concrete content is thin so far. DutchNews says the only funding figure in the vision is €120 million for Dutch participation in an EU-wide AI project, already set aside in September. IO+ adds that the money is available from 2027 and is meant to develop new technologies and infrastructure and strengthen the Dutch and European AI value chain. The letter gives no amounts for other investment. It calls for "targeted and structural investment" without saying how much. The Rijksoverheid release says infrastructure and computing power needs will be set out in early 2027, and DutchNews puts the compute and datacentre plans in the first quarter of 2027. IO+ also says a plan for the Dutch AI factory in Groningen is due then. The Groningen detail appears only in IO+, not in the official release.
This matters because sovereignty is mostly a matter of capacity. A strategy that names the goal but defers the compute plan leaves open how a Dutch start-up would get GPUs, power and grid connections in 2027. Volt leader Laurens Dassen, quoted by DutchNews, called the plan "a toothless tiger." That is a political judgement, not an analytical one, but it points at a real gap: ambition is stated, and money and megawatts are not yet.
The pro-innovation parts deserve credit
Several elements point the right way. The official release says the cabinet wants to "remove barriers that hold back SMEs" from using AI. It also wants to create room to experiment under supervision with new AI applications. IO+ reports a digital marketplace for validated AI applications and more joint government procurement. Small firms benefit from these measures. Shared procurement and a vetted marketplace cut the cost of due diligence that only large firms can absorb alone. Supervised experimentation, in the form of regulatory sandboxes, lets rules be tested against real products before they harden.
The regulatory frame is already set by the EU. The release notes that the AI Act "sets requirements for safe and responsible use and prohibits AI that carries unacceptable risks." The Dutch contribution is to decide how the Act is supervised. In April the cabinet put its implementation bill out for consultation. Under it, existing regulators supervise AI in their own domains, the Autoriteit Persoonsgegevens (AP) takes the role where no clear supervisor exists, and the AP and the Rijksinspectie Digitale Infrastructuur coordinate. That distributed model beats creating a new central AI agency. A firm in medical devices or finance faces regulators who already know its sector, not a generalist body learning from scratch.
The safety institute question
The cabinet is also studying whether to set up a Dutch AI safety institute. The Rijksoverheid release says it could "help identify and address new risks in time." DutchNews says it would be modelled on the UK's AI Security Institute, with an advisory report due in December. The model is a good one to copy if the Dutch institute follows the UK's approach: technical evaluation of frontier systems, published findings, no licensing power. It would become a problem if it grew into an extra approval layer on top of the AI Act. The December advisory report is the first place to check. Does the institute test models, or does it gate products? Does it publish what it finds? Does it work with the existing supervisors or compete with them?
What to watch
The vision is a reasonable statement of intent, and its tone on SMEs and experimentation is better than the usual "regulate first" reflex. Three tests will show whether it is more than that:
- Compute plan, early 2027: whether it commits real capacity and grid access, or only another study.
- Spending beyond €120 million: the letter gives no figures, so any later budget line will show how serious the autonomy goal is.
- Safety institute design: an evaluator that publishes its work supports open innovation, while a gatekeeper does not.
Sovereignty achieved through capacity, open procurement and proportionate supervision would be good for Dutch industry. Sovereignty achieved through duplicated rules would just add costs to the firms the strategy says it wants to help. The Netherlands has chosen the first path on paper, and 2027 will show whether it follows through.