AI infrastructure becomes a strategic foundational layer
The discussion about AI usually revolves around what organizations can do with it: analyze faster, automate better, predict smarter. But behind all those applications lies a less visible foundation. AI infrastructure and AI not only require software but also computing power, storage, networks, access to models, and a foundational layer that is robust enough to actually run all those processes.
This foundational layer is increasingly taking on the character of an economic utility. Those who have access to reliable AI capacity can experiment faster, scale up quicker, and integrate AI more easily into business processes. Those who do not or can hardly organize that capacity themselves become more dependent on external parties for precisely the technology that increasingly plays a role in productivity, innovation, and competitiveness.
As a result, AI infrastructure is shifting from a technical subject for specialists to a strategic dossier for companies and governments. It is no longer just about servers and GPUs, but about the question of where AI runs, under what conditions, and who ultimately has control over the underlying systems.
What is AI infrastructure?
AI infrastructure is the technical and operational basis on which organizations can develop, train, run, and scale AI. It is therefore not only about powerful chips or servers but about the complete system needed to reliably, securely, and manageably use AI applications.
This includes:
- computing power, such as GPU capacity
- storage and data processing
- network and platform layers
- access to models and development environments
- facilities for scalability, security, and compliance
AI infrastructure consists not only of computing power but also of storage, networks, models, security, and compliance.
It is precisely that underlying layer that determines whether companies and governments can independently deploy AI or remain dependent on external providers for capacity, storage, and execution.
Even at the European level, attention is visibly shifting from isolated AI applications to the infrastructure underneath. EuroHPC is now talking about AI GigaFactories as the next step: infrastructure hubs with significantly greater computing power, integrated data resources, and more automation. This underscores how quickly AI capacity is changing from a technical prerequisite into a strategic policy dossier.
Why the Netherlands is at risk of falling behind
The Netherlands does not yet have a self-evident advantage in this area. Large-scale AI infrastructure is limited here, especially when looking at capacity that is explicitly located on Dutch soil and operates under European legislation and compliance. This increases dependence on foreign providers, especially at a time when AI is becoming a core component of the digital strategy for more and more organizations.
This dependence does not only affect commercial companies. For governments, healthcare institutions, defense organizations, and knowledge-intensive sectors, the question of where their AI workloads run and how sensitive data and model usage are legally and operationally embedded becomes increasingly relevant. As AI penetrates deeper into decision-making, service delivery, and automation, infrastructure becomes a matter of strategic autonomy.
This dependence is now also explicitly part of Dutch policy. The Dutch government states that the Netherlands and Europe have become too dependent on a small number of foreign players for crucial digital infrastructure. Therefore, the cabinet is working on a national approach to digital infrastructure and on stronger Dutch and European tech capabilities. At the same time, Dutch companies are reassessing their digital dependence and prioritizing sovereignty more.
This does not need to be alarmist, but it is a serious issue. If Dutch organizations primarily use AI through infrastructure, platforms, and ecosystems from outside Europe, a structural dependence arises that goes beyond just technology. It also concerns ownership, regulation, continuity, and geopolitical vulnerability.
The fact that AI infrastructure is now seen as a strategic subject is also evident from other initiatives. In 2025, the cabinet and region are already investing 200 million euros for a Dutch AI factory in Groningen. This makes it clear that the discussion about AI capacity in the Netherlands is no longer just about software but also about physical infrastructure and national positioning.
The bottleneck is not only in technology but also in energy
This brings us directly to a typical Dutch problem: building AI capacity is not only a technological challenge but also an energy issue. The demand for computing power is growing, but the space to deploy that capacity in a traditional manner is limited. Network congestion and pressure on the power grid make the expansion of heavy digital infrastructure more complicated than it sometimes appears on paper.
This energy issue is not theoretical. In the Status of Implementation 2025, Netbeheer Nederland shows that although network expansion is taking place at record speed, the demand for transport capacity is growing faster than the grid can keep up. As a result, connection timelines for new and heavier connections are extending, and waiting lists are increasing.
This has implications for how new AI capacity can be built in the Netherlands. The classic model of one large central data center is not automatically the obvious choice, especially not in a market where speed, scalability, and energy use are simultaneously under pressure. For parties that want to offer AI infrastructure, the question is therefore not only how much computing power they can provide but also how they can practically and energetically organize that capacity.
It is precisely here that AI Mills is trying to position itself.
AI Mills is working on modular AI data centers and AI infrastructure spread across the Netherlands
AI Mills presents itself as a builder of AI infrastructure on Dutch soil. The core of the story is that the company does not focus on one large-scale mega data center but on multiple smaller, modular AI factories that can be placed throughout the Netherlands.
Behind AI applications lies a physical foundational layer of servers, storage, and platforms – this is precisely what the discussion about Dutch AI infrastructure revolves around.
According to AI Mills, this approach makes the infrastructure more flexible and faster to scale. Instead of first developing a large central complex, the company wants to build capacity in phases at locations where energy is technically and practically more available. This should make the rollout not only more agile but also better aligned with how the AI market is developing according to the company: fast, changeable, and highly dependent on demand peaks.
This choice for a distributed model is also the core of its own positioning. AI Mills wants to show that AI infrastructure in the Netherlands does not necessarily mean that there will be another massive digital location, but that you can also work with a network of smaller facilities that together fulfill the same function.
The central selling point: sovereign AND energy-efficient
The main promise of AI Mills is twofold. On the one hand, the company wants to provide AI infrastructure that is entirely located on Dutch soil and operates within European legislation and compliance. On the other hand, it claims that this model can be rolled out without burdening the power grid in the same way as classical central data center concepts.
This claim is attractive because it addresses two current concerns at once: dependence on foreign AI capacity and the practical limitations of the Dutch energy infrastructure. According to AI Mills, the solution lies in modular construction and placement at locations where energy surplus or more sustainable generation is available, for example, near wind or solar energy. In that story, data sovereignty under pressure from smart AI plays an increasingly important role.
This is a clever positioning, but it remains important to distinguish between claim and proven practice. That a modular model potentially better addresses network congestion does not automatically mean that all operational and technical bottlenecks are solved. It is clear that the company is explicitly profiling itself at the intersection of AI capacity, energy-efficient building, and sovereignty. This also fits into a broader movement where digital control, sovereign cloud, and AI threats are coming together more prominently.
More than GPU capacity: AI Marketplace as an extra layer
AI Mills does not limit its story to pure computing power. In addition to GPU-as-a-Service, the company also wants to offer an AI Marketplace where organizations get direct access to applications and base models, such as language models, AI agents, and image and video models.
This makes the proposition broader than just infrastructure rental. In practice, it means that customers do not necessarily have to start completely from scratch with model development but can build on existing models and further train or adapt them to their own use case. The comparison with a library is functional: companies essentially borrow base models and use them as a starting point for their own applications.
This shows that the infrastructure discussion is not only about hardware. Access to AI capacity only becomes truly interesting when organizations can also quickly connect that computing power to usable models and concrete workflows. This shifts the narrative from pure infrastructure to adoption speed, development time, and cost control.
What makes this interesting for customers and the market
If this model works as AI Mills outlines, it can be attractive for customers in multiple ways. First of all, it may lower the threshold to acquire AI capacity closer to home without immediately being fully dependent on foreign hyperscalers. Additionally, a local infrastructure layer can better meet organizations that have stricter requirements for compliance, data sovereignty, and legal control.
For Dutch companies, this could mean that AI can move faster from experimentation to production. For governments and regulated sectors, it is especially relevant that infrastructure is not only technically available but also becomes better administratively and legally applicable. Sectors such as healthcare, government, and defense look at risks, control, and data use differently than an average commercial startup.
This is also interesting for the market as a whole. If more serious AI capacity arises in the Netherlands, it can lower the threshold for new applications, new business activities, and a stronger local ecosystem. This means that infrastructure is not only a technical provision but also a condition for establishment and innovation. You can also see this reflected in broader market movements where AI infrastructure in the Netherlands is increasingly gaining momentum.
What still needs to be critically assessed
At the same time, caution is needed. Ambitious infrastructure plans are not automatically the same as proven market readiness. That AI Mills is choosing an interesting direction does not mean that all major questions have already been answered.
The first question is how much capacity will actually become available and in what timeframe. Additionally, it is relevant how competitive such a model becomes against established hyperscalers, who still have a strong advantage in scale, price, and ecosystem. The energy issue also deserves more than just a conceptual solution: how is energy use practically arranged, at which locations, and under what technical conditions?
There is also an administrative layer to consider. "Sovereign" sounds attractive, but it ultimately needs to be concrete for customers. Does it mean that data remains fully in the Netherlands? That operational control is organized locally? That the infrastructure is legally and technically independent of non-European dependencies? It is precisely on such points that the market will scrutinize more closely as the plans become more concrete.
This scrutiny will also become more relevant as AI regulations become clearer. In April 2026, the cabinet brought the Dutch implementation law for the European AI regulation into consultation. This clarifies that data quality, risk management, human oversight, and transparency are not just policy words but hard conditions for certain AI applications. For organizations that want to assess what that practically means, the question is also relevant what companies need to do now with the EU AI Act.
This is exactly where an editorial article should make a difference. Not by dismissing the initiative as wishful thinking but by making it clear that infrastructure only becomes relevant when promise, scale, energy, compliance, and operational feasibility come together.
A signal of a broader Dutch shift
That is why AI Mills is particularly interesting as a case. The company shows where the AI discussion in the Netherlands is heading. Not just talking more about tools, copilots, and models, but about the question of where that AI runs, how it is powered, under what legislation it operates, and who ultimately has control over it.
This shift is not only national. The European Commission explicitly focused on more own AI capacity with the AI Continent Action Plan, and the network of European AI Factories grew to 19 locations in 16 member states, including the Netherlands. This shows that AI infrastructure is increasingly seen as a strategic foundational layer for European competitiveness and autonomy.
This makes AI infrastructure more than just a technical niche topic. It touches on energy policy, digital autonomy, innovation power, and economic positioning. In that sense, AI Mills is less just a business story than an example of a broader shift: AI only becomes truly strategic when the foundational layer is also strategically organized.
For organizations that want to seriously deploy AI, that is the real lesson. Those who only look at applications see only part of the playing field. Ultimately, it will also be decisive where that AI runs, under what compliance conditions that happens, and how robust the underlying infrastructure is. It is precisely there that it becomes clear why AI in the Netherlands is not just a software issue but increasingly an infrastructure issue.