Sign the statement of support: the Netherlands needs a national ELLIS Institute
The Netherlands excels at training AI talent, but loses its best researchers and entrepreneurs to other countries. A national institute for AI research within the European ELLIS network means that Dutch talent no longer has to leave in order to belong to the scientific frontier, and that the companies which turn that research into value are founded here. That way the Dutch economy shares in the largest technological transition of our time.
Signed by
The largest technological turn of our time is happening right now in Silicon Valley.
In and around San Francisco, researchers are working on computers that can reason. At high speed, AI is growing into the foundation under our economy, our society and our national security. The potential benefits are, just like the potential disruption, enormous. And the power to shape this transition lies with a country that is positioning itself ever more hostilely towards Europe.
In Europe the reflex is to put policy up against that: a model of our own, a supercomputer of our own, a government programme. However, a programme does not automatically generate the next billion euro company.
High tech companies emerge where the people are who do the work. Those people attract capital, while success and capital attract new people in turn.
Talent circulates between companies and research groups. That exchange spreads knowledge and skill and creates the opportunities for new research, applications and startups.
This agglomeration flywheel works the same way everywhere – whether you are talking about high tech in Silicon Valley and Zürich or about Solingen in Germany, world leader in blades and scissors.
Source: opening years accounted for per lab in Sources.
Around ETH in Zürich, one of the strongest technical universities in Europe, Google opened a site in 2004 that has grown into its largest engineering location outside the United States, with around five thousand employees.
Other labs followed. They did not come for an innovation decree from the Swiss ministry of economic affairs, but grew by themselves around a core of research, because concentrated talent is the most important raw material for innovation.
Source: opening year accounted for per lab in Sources.
The same happened in Montreal, where the catalyst was not a university but an institute. Mila was founded in 1993 and has been independent since 2018. Google came in 2016 and Microsoft, DeepMind and Meta all followed in 2017.
Dutch AI talent often ends up abroad
Whoever holds the foundations of AI decides where the power will lie. The Netherlands is exceptionally good at training world class AI talent, among them a number the founders of the current technological breakthroughs. And still we see our best people consistently leave.
Durk Kingma did his PhD in Amsterdam. He came up with the variational autoencoder and the Adam optimizer, two methods that lie under a large part of today’s generative AI. He left for San Francisco to co-found OpenAI and now works at Anthropic.
Laurens van der Maaten, educated in Maastricht and Tilburg, developed t-SNE together with Nobel prize winner Geoffrey Hinton, a technique that is used in virtually every machine learning lab in the world. He worked at TU Delft and left for Meta. There he led the technical development of Llama 3, one of the most influential open AI models of that moment. He now works at Anthropic in New York.
Hado van Hasselt did his PhD in Utrecht and worked on double Q-learning at the CWI in Amsterdam. He is now a senior researcher at Google DeepMind in London. This shows exactly how the breakthrough itself was made in the Netherlands and then left together with the talent behind it.
There is no brain drain going on in the Netherlands: the total number of scientists remains relatively stable. But the researchers who leave the Netherlands are cited 17% more often on average than the researchers who arrive here. That is partly down to the multimillion salaries on offer in the US, but it is also down to something the Netherlands can do something about: research depth.
According to our own calculation based on CSRankings, the Netherlands ranks thirteenth in the world as a country, while the highest placed Dutch university only comes in at place 42. That is exactly the picture of a country with many good researchers but not a single location with enough people.
Own calculation on the source data of CSRankings, 2016–2025. Place in the ranking of institutions.
If all Dutch AI research were to take place in one location, that institute would stand tenth in the world.
Three pillars of the Dutch AI ecosystem
In the Netherlands our AI talent is spread across many universities and cities. There is no shared place where researchers can work together over a long period and at a larger scale. Investments in computing power and in software engineers are fragmented and small scale as it stands. Concentration offers a scale that is within reach of no individual university.
A healthy AI ecosystem consists of three pillars that strengthen each other: breakthrough research, research into applications and AI business activity.
The Innovation Center for Artificial Intelligence (ICAI) organises use-inspired research in partnerships between universities and companies. NADI, the National Agency for Disruptive Innovation, sets out programmes around large challenges and drives the application and the scaling up of breakthroughs. Start-ups and scale-ups then commercialise that technology.
For the groundbreaking work that applications and commercialisation run on, such a home is missing, while that work in particular asks for more than a single chair can carry: large scale computing power, support from software engineers and a concentration of top researchers who work on new breakthroughs together for years.
That is a serious gap, because the most innovative companies can come out of fundamental research. Diffusion models are a good example. That technique was largely developed at universities, among them Stanford, Berkeley, Munich and Heidelberg, and now enables virtually all image and video generation. Out of the group in Munich came Stable Diffusion, and out of that in turn the German company Black Forest Labs. Contract research delivers applications at a company that already exists. New, groundbreaking companies emerge one layer deeper: out of work where nobody knows at the outset where it will lead to. That layer has no place in the Netherlands yet.
An ELLIS institute adds to these initiatives. It develops new methods and trains the people that applied research, such as in the ICAI labs, can build on.
ELLIS, the European network for AI research
The European network this revolves around already exists. The European Laboratory for Learning and Intelligent Systems (ELLIS) has been building shared infrastructure since 2018 and by now counts 47 locations in nineteen countries. Within it the Netherlands has three units, in Amsterdam, Delft and Nijmegen, with more than a hundred and sixty affiliated researchers between them.
Source: ELLIS, ellis.eu/research/sites, 7 September 2026.
An ELLIS unit is a research group within a university. The network is spread across Europe, attached to existing knowledge institutions.
ELLIS recently took a new step by founding independent institutes, which stand apart from existing universities and are fully focused on fundamental research, talent development and valorisation. Two exist now, in Tübingen and in Finland.
By founding an ELLIS Institute of its own, the Netherlands can set the innovation flywheel in motion and reap its fruits.
An institute with a different way of working
What makes such an institute fundamentally different from a large university research group are four design principles.
- Autonomy
- The scientific leadership decides for itself on the research agenda, the recruitment and the spending, without depending on annual funding rounds.
- Talent
- Positions are filled on international top quality. The pull of the institute depends entirely on who already works there.
- Mandate
- A clear mission and multi-year, predictable funding that holds up across cabinet terms.
- Appetite for risk
- The room to take uncertain directions, agreed that way from the start instead of trying to organise it afterwards.
Research within the institute is led by Principal Investigators with far reaching autonomy. Researchers work at the institute for part of their time and keep their appointment at their own university. The time at the institute is entirely for research. Built in flexibility means scientists can put their appointment on hold temporarily to set up a startup. That way the lines between the ELLIS Institute and the commercial ecosystem stay short.
An institute of this kind only works at sufficient scale and with sufficient stability. Breakthrough research asks for critical mass: several groups at once, shared computing infrastructure and young talent that benefits from senior colleagues. And it asks for predictable, multi-year funding. What it comes down to above all is a stable investment. Without that stability the autonomy and the appetite for risk disappear, and with them the working of the entire model.
Why now
For government, AI policy is no longer a side issue. The Wennink report, the AI Deltaplan and the way those two are worked out in the coalition agreement show that the political momentum is there. The building blocks are ready: the Dutch ELLIS units, an exceptionally strong talent base and existing international collaborations mean the institute does not have to start from zero.
Source: ELLIS, ellis.eu/research/sites, 7 September 2026.
Set against that are the costs of waiting. Top researchers, venture capital and European research programmes concentrate in places where research and industry are both strong: London, Paris, Zürich, Tübingen and Helsinki. Those are cities that have already made their choice. But in the Netherlands too all the ingredients are there to bundle strongly. The talent is there, the research is there and the companies are there. The only thing missing is one place with enough concentration.
The longer the Netherlands waits, the stronger the pull of those places becomes and the more of our own talent finds a home there.
A Dutch ELLIS Institute sets itself apart not through more money, but through how it is organised: independent, concentrated and in close collaboration with the places where AI is already being used. That makes it the pillar that is structurally missing right now.
An institute of its own puts the Netherlands in a position to contribute to what AI is and not only to what is done with it.
Sign the statement of support
Companies and organisations from the Dutch AI ecosystem state in a statement of support to the Taskforce Toekomstige Welvaart en Vestigingsklimaat and to the cabinet that the Netherlands needs an ELLIS Institute. Signatories commit in the way that suits their organisation, through talent, collaboration or financial support, once the government commits its own contribution.
Signed by
Questions
The short answers to what is asked most often.
What is an ELLIS Institute?
ELLIS is the European Laboratory for Learning and Intelligent Systems, a European network for AI research that has existed since 2018 and by now counts 47 locations in nineteen countries. Most locations are units: research groups within a university. An institute is something else: independent, separate from existing universities and fully focused on fundamental research, talent development and valorisation. There are two at the moment, in Tübingen and in Finland.
Does the Netherlands not already have ELLIS units?
Yes, three of them: in Amsterdam, Delft and Nijmegen, with more than a hundred and sixty affiliated researchers between them. A unit does remain a group within a university. What is missing is one place with enough concentration: several groups together, shared computing power, support from software engineers and top researchers who work on new breakthroughs together for years. The units are the building blocks that mean an institute does not have to start from zero.
Do only the universities with an existing ELLIS unit take part?
No. Every Dutch knowledge institution can join. ELLIS affiliation is not a condition, neither for the institution nor for the researcher. Positions are recruited openly and internationally, and every candidate goes through the same assessment.
Is an ELLIS Institute part of a university?
No. The institute stands apart from existing universities and has its own leadership: Principal Investigators set the research agenda, not a university board. Researchers often also keep an appointment at their own university, but the institute does not belong to any university in particular, just as with the existing ELLIS institutes in Tübingen and Finland.
Can companies take part as well?
They can. A company can fund a research position and put forward a candidate for it itself. Being a knowledge institution is not a condition. That candidate is assessed by the same international committee as every other applicant. Co-funding pays for a position, but it does not buy an appointment and it does not buy a say in the research agenda.
What is the difference with ICAI?
The Innovation Center for Artificial Intelligence (ICAI) is a bilateral collaboration between universities and companies, in which the intellectual property usually goes to the funder. An ELLIS Institute is collaboration at national scale, aimed at breakthrough research. The two strengthen each other: breakthrough research flows through to use-inspired research such as in the ICAI labs. An ELLIS institute adds to ICAI.
What is the difference with NADI?
NADI, the National Agency for Disruptive Innovation, formulates programmes around large challenges and challenges external researchers and companies to find solutions. NADI does not employ researchers itself. An ELLIS Institute does the opposite: it appoints its own researchers and carries out the research itself, without distributing money to others. NADI could well fund a programme that builds on the institute's research: ELLIS then delivers the knowledge such a programme can carry forward.
Why not a programme or a larger university lab?
Because the way it works sits in the form of organisation. Four design principles make the difference: autonomy over its own agenda, recruitment and spending; recruitment on international top quality; a clear mandate with multi-year funding; and room for uncertain directions. In practice that means: research is led by Principal Investigators with far reaching autonomy, researchers keep their appointment at their own university and spend their time at the institute entirely on research, and anyone who wants to start a startup can put the appointment at the institute on hold temporarily.
What does it cost?
This page deliberately does not name an amount. It is not about a single large sum, but about a stable investment. Breakthrough research asks for a critical mass and for predictable, multi-year funding that holds up across cabinet terms. Without that stability the autonomy and the appetite for risk disappear, and with them the working of the entire model.
How can my organisation support this?
By signing the statement of support. Companies and organisations from the Dutch AI ecosystem state in it that the Netherlands needs an ELLIS Institute and commit in the way that suits them, such as through talent, collaboration or financial support, once the government commits its own contribution.
Contact
Questions about the proposal or the statement of support? Get in touch.
An initiative of
Sources 8
Every figure on this page is listed below, with its source and the date on which it was established.
- Attention Is All You Need mentioned in the text only
- CSRankings mentioned in the text only
- ELLIS — Research sites
- 44 units, 1 associate unit and 2 institutes in 19 countries, 47 locations in total.
- International mobility of AI scientists
- Field-weighted citation impact on Elsevier data, 1996-2019. Leavers 2.49 against arrivals 2.12, which is 17% higher. In the same publication Rathenau characterises the Netherlands as a brain exchange country.
- Opening years of AI labs around Mila in Montreal
- Mila was founded in 1993 by Yoshua Bengio and has been an independent non-profit since 2018 (mila.quebec). Google Brain Montreal 2016. Microsoft January 2017, through the acquisition of Montreal based Maluuba (blogs.microsoft.com, 13-01-2017). Samsung SAIT AI Lab 17 August 2017, at the Universite de Montreal (news.samsung.com, 30-11-2017). DeepMind October 2017, its second international lab (betakit.com). Meta FAIR end of 2017. Thales opened cortAIx in 2018, announced 10 October 2017 (thalesgroup.com). Borealis AI of RBC opened in 2018. ServiceNow set up its AI Innovation Hub in January 2021 through the acquisition of Element AI (servicenow, 30-11-2020).
- Opening years of research labs around ETH Zürich
- Google Zurich 2004, grown to around 5,000 employees and the largest engineering site outside the United States (swissinfo.ch; blog.google). Disney Research Zurich 2008, with ETH. Meta through the acquisition of Zurich Eye by Oculus, November 2016 (Wyss Zurich; VentureBeat). Anthropic 2024 (Greater Zurich Area; S-GE). OpenAI 2025, its fifth European site (Greater Zurich Area).
- Own calculation of the Dutch position based on CSRankings
- Own calculation on the source data of CSRankings (generated-author-info.csv and institutions.csv), derived from DBLP. Retrieved 8 September 2026. Subfields: the full AI group of CSRankings, that is Artificial intelligence, Computer vision, Machine learning, Natural language processing and The Web & information retrieval. Window 2016-2025, the last ten complete years; 2026 is not finished and would let a partial year weigh in. Scope worldwide, 638 institutions. The score is the same measure CSRankings publishes itself: per subfield the adjusted publication count, in which each publication counts once and the credit is divided equally over the coauthors, and over that the geometric mean of (count + 1) across the five subfields. Country scores are calculated in exactly the same way, with all scientists of that country together as one whole. CSRankings does not publish a country ranking itself; that ranking is therefore ours. Own calculation on the source data of CSRankings, window 2016-2025. Three figures come out of it: the 42nd place of the highest ranked Dutch institution, the 13th place of the Netherlands as a country and the 10th place of a bundled Dutch institute. The faculty lists of CSRankings are maintained by the community; coverage differs per institution.
- The agglomeration mechanism
- Schematic rendering of the mechanism. No measurement data.