Why Is Research Infrastructure More Than a Laboratory?

I used to think of research infrastructure as a place. Now I think of it as the ability to see the same thing in different ways.

Where could this be tested?

I hear this question often.

Sometimes the answer is a laboratory. Sometimes we need a real-time simulator. Sometimes a real power network. And sometimes the answer should not first be sought in the physical world at all, but in data, a model or a scenario.

And often we need all of these.

At that point, the question is no longer really where a particular device can be tested.

A much more interesting question is: how can we gain enough understanding of what happens when this solution becomes part of a real system?

In my previous post, I wrote that the purpose of validation is not to eliminate all uncertainty. Its purpose is to produce sufficiently reliable evidence to make the next decision.

But where does that evidence come from?

Not all research infrastructure fits inside a laboratory

When we talk about research infrastructure, my own first mental image has long been very physical.

A laboratory. Equipment. Instruments. Cables. Perhaps an entire test network.

But some of the most important research infrastructure does not fit inside any physical room.

It can be data. It can be a power system model. It can be a computing environment, a simulation platform or a virtual model of a system that does not even exist yet.

And at the other end there may be real devices, physical laboratories, hardware-in-the-loop (HIL) environments, pilots and, ultimately, the operating energy system itself.

Physical, digital and virtual are therefore not alternatives to research infrastructure.

They are different layers of it. And then the most interesting question becomes what happens between them.

We need both a microscope and a map

Take data centres, for example. They are an exceptionally timely case.

In Finland, the planned combined capacity of data centre projects that already have connection agreements is approaching five gigawatts. Together with other electricity consumption projects that have connection agreements, they could, if fully realised, increase Finland’s electricity consumption by almost 40 per cent compared with the 2025 level. Not all projects will necessarily be realised in full, but their scale alone shows how significantly the power system may change.

If we want to understand the electrical behaviour of a single data centre, we need to zoom in very closely.

How does its load behave? What happens during a disturbance? How do protection systems, power electronics and control systems respond? Can its electricity consumption provide flexibility when the power system needs it?

For this, we need detailed models, simulation, real-time environments and sometimes real hardware.

We need, in a sense, a microscope.

But at the same time, we need to be able to take several steps back.

What happens if there are many data centres?

Where will they be located? Do we need new networks, more generation, storage or balancing capacity?

And from the perspective of an ordinary electricity user, the question quickly becomes very concrete: what does all this mean for my electricity bill or network charges?

This is already the subject of public debate. One view is that the growth of data centres will attract new electricity generation and therefore will not necessarily raise prices permanently. At the same time, questions have been raised about whether responsibility for ensuring sufficient generation and flexibility should also lie with large consumers.

So the question is not only how much electricity a data centre uses. What matters is what its arrival does to the system as a whole — to generation, networks, flexibility, investments and, ultimately, how costs are distributed.

Research infrastructure does not tell us who should pay those costs.

But perhaps one of its roles is to help us see consequences while they are still alternatives — before they become costs. It should help us produce the knowledge needed to make better decisions.

What does a data centre really do to the grid? Where would its location cause the least need for additional investment? What happens if several large consumers behave in the same way at the same time?

And what happens if industry, heating and transport are electrified at the same time while the system also gains more weather-dependent generation?

At that point, the microscope is no longer enough.

We need a map.

The microscope and the map must not tell different stories

We cannot build the entire power system in a laboratory just to see what happens. Part of the future system does not even exist yet.

So we need scenarios, data, optimisation, models and simulation.

But the microscope and the map must not tell different stories. A physical experiment should refine the model, and the model should in turn tell us what is worth testing next. If something unexpected happens in a pilot, it should be taken back into simulation: why did we not see this earlier? At the same time, simulation should reveal situations that we would not have known to test in the laboratory.

This is not a one-way journey from model to reality.

It is continuous movement between the two.

For data centres, this is also becoming very concrete through regulation. Fingrid is preparing new KJV2026 requirements for large consumption facilities above 30 megawatts, including data centres. The requirements cover, among other things, disturbance tolerance and the ability to limit the power drawn from the grid. The target is for the requirements to enter into force in March 2027.

At that point, the research question becomes a practical one.

How do we demonstrate that the requirements are actually met?

Then the question is no longer only how the system behaves. It must also be possible to demonstrate it. That requires models, measurements, simulation and physical testing.

A break can occur even when nothing is packed into a box

When we talk about fragmented research infrastructures, we tend to think first of a physical problem.

A device is tested here. Then it is packed into a box and shipped to the next laboratory. We wait for a slot. The next test is carried out. That is a very real problem.

But a break can occur even when no device moves at all. Data remains in one system. The assumptions behind a model are not known at the next stage. The initial conditions of a simulation do not correspond to what was measured in a physical experiment. A model created in one project cannot be used in the next.

Or a test produces a good result, but the next actor does not know enough about how that result was produced to rely on it in their own decision-making.

Then we lose something that may be even more valuable than time.

We lose continuity of learning.

The valley of death is not always only about money

The innovation valley of death is often discussed as a funding problem. And of course, money can run out.

But sometimes something else is missing: evidence that gives the next actor enough confidence to make a decision.

A company knows that its solution works in its own test. But a distribution system operator needs a different kind of evidence. A customer asks a different question. A certifier needs a documented procedure and results that can be revisited later. An investor, in turn, wants to understand how much uncertainty still remains.

The same technology. Different decisions. Different questions.

That is why it is not enough that a test has been carried out. The evidence also needs to be understandable to someone who was not there.

At that point, jointly agreed procedures — and, where needed, accreditation — start to look like something very different from bureaucracy.

They are one way of building trust across organisational boundaries.

Perhaps shared does not mean a shared building

The microscope, the map and the connection between them are therefore not only technical questions. They determine whether understanding, once created, survives when the stage, environment or actor changes.

This has also changed my own view of what shared research infrastructure means.

It does not have to mean a place where all equipment is gathered together. It can mean shared access to different capabilities. Compatible models and data. Interfaces through which physical, digital and virtual environments located in different places can be brought together around the same research question. And common ways of documenting how evidence was created and what it can — and cannot — demonstrate.

It makes sense for few organisations to build everything themselves.

And they do not need to.

What matters is whether we can build the connections.

This is exactly the question we are currently examining in the Novel Power Infra work. For me, the most interesting question is no longer what can be done in one location, but what we can understand when different capabilities are made to work together. The concept is described in more detail in the Novel Power Infra white paper, published in September 2026.

What can we see only when physical, digital and virtual environments are connected to one another?

Perhaps the most important product is not a test result

Research infrastructure does not, of course, solve everything.

It does not remove uncertainty. It does not turn bad technology into good technology.

And maintaining shared infrastructure is not free. Someone has to keep the data usable, the models up to date and the environments operational even when no project is using them at that particular moment. That raises difficult questions about ownership, funding, openness and who orchestrates the whole.

These are questions I am currently studying myself.

And I do not yet have answers to all of them.

But one thing is clearer to me now than before.

The value of research infrastructure does not ultimately come only from its equipment, data, models or computing power. It comes from whether we can move between the microscope and the map without losing contact with reality.

From an individual device to the entire system.

From the physical to the digital and the virtual — and back again.

From knowledge produced by one actor to the basis for the next decision.

And perhaps we also need a compass. The microscope helps us see in detail, the map helps us understand the bigger picture, and the compass helps us use that understanding to decide where to go next.

Perhaps the most important product of research infrastructure is ultimately trust.

Not certainty that everything works.

But enough confidence that we understand what we know, what we do not yet know — and what we should test next.

And when research infrastructure can also consist of data, models and virtual environments, one more interesting question appears.

Can we test an energy system before it even exists?

Perhaps we can at least make some futures visible before we build one of them.