GPT-NL: Why the Netherlands Is Building Its Own AI Model
GPT-NL Netherlands is the country’s attempt to build a national artificial intelligence language model that is not controlled by American tech companies. Developed by Dutch public research organisations TNO, SURF, and the Netherlands Forensic Institute (NFI), it moved from laboratory to live pilots in early 2026 and is now being tested by municipalities, government agencies, and public institutions across the country. The broader commercial roll-out is planned for the second half of 2026.
The project sits at the intersection of two of the most pressing questions in technology today: who controls the AI tools that shape public life, and whether Europe can build credible alternatives to Silicon Valley.
What GPT-NL Is
GPT-NL is an open large language model built specifically for the Dutch language and context. Unlike ChatGPT or other commercial models, it is developed entirely within the Netherlands and Europe, with full transparency about its training data, decision-making processes, and compliance with Dutch and EU privacy law.
The model is funded by €13.5 million from the Dutch Ministry of Economic Affairs, allocated through the Netherlands Enterprise Agency. It is not being built to compete with the most powerful American AI systems on general benchmarks. The goal is something more specific: a trustworthy, sovereign model that Dutch public institutions can use without routing sensitive data through foreign servers.
As TNO puts it: “Our Dutch digital sovereignty on a critical technology like AI is at stake here.”
From Lab to Live
GPT-NL spent 2024 and most of 2025 in development. Since late February 2026, five organisations have been running feasibility studies with the beta version, according to Computer Weekly. That number is set to grow to ten by spring 2026, with a broader commercial roll-out via professional licensing planned for the second half of the year.
One of the most concrete applications is being tested with Gem, the virtual municipal assistant already used by around 2.8 million residents in cities including Utrecht, Rotterdam, and Eindhoven. ICTU, which manages the Gem platform, has installed GPT-NL on its own AI infrastructure to test whether the model can enhance Gem’s capabilities, expand what it can answer, and reduce maintenance costs. Results are expected after summer 2026.
Government agencies including TNO and the NFI are also piloting the model internally. The NFI, which carries out forensic investigations for the Dutch justice system, sees particular value in a model that keeps sensitive case data within Dutch legal jurisdiction rather than processed on American cloud infrastructure.
Why Sovereignty Matters
The argument for GPT-NL is not primarily about building a better chatbot. It is about control.
When Dutch government agencies, hospitals, courts, and municipalities use ChatGPT or other commercial models, they are sending potentially sensitive data to servers operated by US companies under US law. The EU AI Act, which came fully into force in August 2025, imposes strict requirements on transparency, accountability, and data governance for AI systems used in high-risk contexts. Commercial models do not always meet these requirements or make it easy to verify that they do.
GPT-NL is designed from the ground up to be transparent, auditable, and compliant with Dutch and European values. The training data, the model weights, and the decisions made during development are all documented and open to inspection. This makes it suitable for applications where accountability matters, such as public sector communications, legal proceedings, and healthcare.
The model also serves a linguistic purpose. Dutch is a relatively small language compared to English, and commercial AI models trained predominantly on English data perform less well on Dutch-language tasks, particularly for regional dialects and technical or legal Dutch terminology. GPT-NL is trained specifically on Dutch-language data, including a large archive of Dutch news articles provided by media organisations including ANP.
The Dutch AI Landscape
GPT-NL is part of a broader push to make the Netherlands a serious player in European AI. The National Growth Fund has allocated €1.1 billion to AI-related projects, including the AiNed programme worth €276 million and the ROBUST AI initiative at €84 million.
The Netherlands is also getting an AI factory, a European supercomputer for training and developing AI models, to be built in Groningen. Funded in part by EuroHPC, the EU’s high-performance computing initiative, the centre of expertise starts operations in 2026 with the supercomputer running at full capacity in 2027.
The 2026 State of Dutch Tech report, produced by Techleap, found that the Netherlands has 11,301 active tech companies and attracted €2.64 billion in venture capital in 2025, placing it among Europe’s frontrunners. However, the report also flagged a persistent bottleneck: the Dutch scaleup rate, the percentage of startups that raise more than €10 million, stands at 21.6 percent, compared to a European average of 24.1 percent and a US figure of 52.2 percent.
AI attracts 27 percent of Dutch venture capital but the conversion from startup to meaningful scale still lags behind European peers. The AI factory, GPT-NL, and the restored €565 million in university research funding are all aimed at closing that gap.
The EU AI Act Dimension
The EU AI Act has added urgency to the GPT-NL project. The Act came fully into force in August 2025, and around 3,200 Dutch businesses are directly impacted, particularly those using AI in high-risk categories such as HR systems, credit scoring, and medical applications.
As of Q1 2026, 67 percent of Dutch businesses are not yet fully compliant, according to Searchlab. The Dutch Data Protection Authority has been designated as the AI supervisory body, with its Algorithm Supervisor unit already conducting its first audits in 2026.
For public institutions, GPT-NL offers a path to compliance that commercial models cannot easily provide: a model where every aspect of its development and operation is documented, auditable, and aligned with European law.
What Comes Next
The second half of 2026 will be the real test for GPT-NL. The move from controlled pilots to broader professional licensing will reveal whether Dutch institutions are ready to adopt a sovereign model over the convenience of commercial alternatives.
The team at TNO is also exploring a next-generation model with multilingual capabilities, which would extend GPT-NL’s usefulness beyond the Dutch language context to broader European applications.
If GPT-NL succeeds, it could become a template for other mid-sized European countries that want AI infrastructure aligned with their own values and legal frameworks, rather than dependent on the choices made in Silicon Valley. If it struggles to gain adoption outside the public sector, it will remain a valuable but niche tool.
Either way, the Netherlands is making a serious bet on the idea that Europe needs its own AI, and that public institutions should not have to choose between capability and sovereignty.
