Five billion against the clock & why the Netherlands must choose now

"AI is a systemic technology that is here to stay, with an impact on all aspects of society. Although this is recognised in the Netherlands, there is too little concrete action and too little investment to actually get to work with AI.", Willem Jonker, chairman AIC4NL [2]

The diagnosis

The position paper opens with a sobering comparison. The Netherlands is investing €250 per capita in AI for the period 2026–2030. Scandinavia is targeting €551, Germany €858, the United Kingdom €858, and France €1,385 [1]. The gap is not new, but its consequences are now becoming visible. The Netherlands ranks 20th on the global AI Index, scoring just 10 to 11 out of 100 for research, development and commercialisation [1]. Only on infrastructure does the country perform reasonably well: 40 out of 100.

The causes are structural. Europe has too long focused on regulation without investing in its own capacity. The AI Act and Data Act are in place, but the companies that must comply with them come from Silicon Valley and Shenzhen. The result is what AIC4NL calls 'technological captivity': dependence on foreign platforms for a technology that is penetrating every aspect of the economy and society.

The solution across three pillars

AIC4NL presents an investment agenda along three lines. The first pillar concerns infrastructure: €1 billion for a Dutch AI factory and sovereign cloud capacity. The AI factory in Groningen, which launched in October 2025 with €200 million, is only the beginning. The coalition points to the 20–80 ratio in successful AI services: for every euro spent on training, four euros are needed for actual use [1]. Following the development investment, a further €800 million must therefore follow for deployment infrastructure.

The second pillar focuses on building a Dutch 'AI Fair Tech' industry: €1.5 billion. This includes an investment fund of €1 billion for scaling AI startups, €250 million for data collections and €250 million for talent. The term 'Fair Tech' refers to AI that meets European values: transparent algorithms, fair data processing, and respect for privacy and copyright.

The third and largest pillar concerns 'AI Grand Challenges' in strategic sectors: €2.5 billion. AIC4NL identifies four domains where AI can have the greatest societal impact: healthcare, energy and climate, defence and security, and agrifood. The coalition allocates €500 million per domain, plus €500 million for retraining and upskilling workers.

Of the total €5 billion, 80% must come from private parties, 20% from the government. This is a deliberate choice: the government as catalyst, the private sector as the primary driver.

The foundation: the AI Deltaplan

The AIC4NL position paper builds on the National AI Deltaplan that Jelle Prins and Michiel Bakker presented earlier in November [3][4]. Where AIC4NL outlines the financial framework, the Deltaplan laid the strategic foundation. The plan contains more than fifty concrete recommendations, drawn up in collaboration with over sixty experts from academia, business and government.

The comparison with the Deltaworks is not coincidental. Just as the storm surge barriers protected the Netherlands from water, an AI delta plan must protect the country from digital vulnerability. The initiators warn of an existential threat. Five years ago, the Netherlands was still in the AI vanguard. Google Brain opened an office in Amsterdam, ASML grew into a key player in chip technology. Today, Dutch companies mainly use American AI models while top talent leaves for San Francisco.

The Deltaplan identifies five strategic opportunities where the Netherlands can excel: vertical AI applications in sectors such as healthcare and agriculture, human-machine interfaces, data platforms for secure data sharing, scientific AI for research breakthroughs, and energy-efficient AI hardware [4]. The most ambitious recommendation is NADI: a Dutch research centre for 'impossible breakthroughs', inspired by the American DARPA, which produced GPS, stealth aircraft and the internet.

Prins, co-founder of AI company Cradle, summarises the urgency: "If we do nothing, we will lose our competitive position. Then all our companies will be using American AI models, until Trump suddenly says we can no longer use them." This is not a hypothetical scenario. Geopolitical tensions make technological sovereignty a matter of strategic importance.

The funding gaps

The AI Deep Dive report by Invest-NL and the regional development agencies, written by AI expert Stefan Leijnen, exposes the funding gaps that stand in the way of the ambitions of both AIC4NL and the Deltaplan [5][6]. The Netherlands has 96 AI-native startups in the ROM portfolio. The average investment per round has risen to nearly €400,000. But that figure is a fraction of what is needed.

The gaps are structural. ROMs invest an average of €400,000; Invest-NL only starts at €10 million. In between lies a 'valley of death' where promising companies stall. At the early stage, capital is lacking for teams that want to train large models. At the scaling stage, there is no breakthrough capital of €100 million or more. The Leijnen report identifies three critical funding gaps: the early capital-intensive phase, scale-up and breakthrough capital, and the rapid seed and Series A financing that is crucial in AI due to the speed of the market.

A notable finding: only 2% of Dutch AI startups focus on hardware and energy, yet this category attracts 18% of invested capital [6]. The market sees opportunities in the Dutch position around semiconductors. But the reality is challenging. ASML invested €1.3 billion in France's Mistral, not in a Dutch AI champion [7]. The capital is there; the companies to invest it in still need to grow.

The broader context: prosperity at stake

Peter Wennink, former CEO of ASML, places the AI investment agenda in a broader perspective [8][9]. In his report on the route to future prosperity, he calculates that the Netherlands must mobilise €151 to €187 billion in productivity-enhancing investments by 2035 in order to sustain the welfare state. In that context, AI is not a side issue but a prerequisite.

The bulk must come from private parties. But they prefer to invest elsewhere: Dutch pension funds invest three times as much in American as in European companies [10]. Wennink uses an apt metaphor: the Netherlands is a car with four flat tyres. The engine is running, but the conditions are missing. Talent shortages, slow permitting, grid congestion, nitrogen problems. A company in the Brainport region that has to wait until 2033 for a power connection is not being hindered by the market, but by policy.

Energy prices compound the problem: the Netherlands pays €95 per MWh, France €32 [6]. For AI companies that depend on computing power, this is a structural competitive disadvantage that no investment agenda can offset.

The way forward

Together, the four reports paint a coherent picture. The AI Deltaplan sets out the strategic direction and the urgency. The Leijnen report from Invest-NL and the ROMs exposes the funding gaps that block the growth of startups into scale-ups. The Wennink report places everything in the context of national prosperity and identifies the preconditions that must first be put in order. And the AIC4NL position paper translates all of this into a concrete investment agenda with price tags and an implementation structure.

The consensus is clear: do not compete with the US and China on the largest language models, but invest in vertical applications and sectors where the Netherlands has traditionally been strong. Healthcare, agrifood, logistics, high-tech, defence. AIC4NL advocates a 'Dutch AI Pact' in which government, business, knowledge institutions and civil society organisations coordinate their investments. A national AI summit should provide the impetus.

The coalition warns against fragmentation: the Netherlands has a tendency to generate many small initiatives with little impact. Only a consolidated effort creates the critical mass that is needed.

What does this mean for startups?

For the 96 AI-native startups in the ROM portfolio and the dozens of others beyond it, the €5 billion agenda is more than a policy document. It is a potential lifeline. The valley of death between €1 million and €10 million where promising companies currently stall could be bridged by the proposed €1 billion investment fund. The €250 million for data collections provides access to training data that individual startups could never assemble on their own. And the €500 million for talent per sector makes it easier to compete with Big Tech salaries.

More concretely: a Dutch AI startup in the healthcare sector could benefit from both the sectoral data collection and the €500 million Grand Challenge for healthcare. A hardware startup focused on energy-efficient chips connects with the infrastructure pillar and the Dutch position in the semiconductor industry. The agenda creates not only capital, but also a market: the government as a launching customer for AI Fair Tech solutions.

But there is a downside. The 80–20 split means that €4 billion must come from private parties, the same private sector that currently invests three times as much in American as in European companies. The AIC4NL agenda is ambitious, but its execution depends on a willingness that does not yet exist. Dutch pension funds and corporates must be convinced that investing in local AI startups is not only socially desirable, but also commercially viable.

The next five years are critical, AIC4NL concludes. For startups, this means a dual challenge: growing within an ecosystem that is still under construction, while at the same time helping to shape that ecosystem. The 96 companies in the ROM portfolio are the foundation. The challenge is to turn hidden champions into visible winners who demonstrate that Dutch AI Fair Tech can compete.

Investing now means becoming a player. Waiting means remaining a prisoner. For the Netherlands as a country, but also for every startup trying to scale without the capital and infrastructure that competitors in the US and China do have.