Meet one of the brains behind Luxembourg’s supercomputers

Thanks to people like Valentin Plugaru, the chief technology officer at Luxprovide, which operates Luxembourg’s supercomputer Meluxina, Europe is pushing computing boundaries – a global race that’s still anybody’s to win

Is it correct to say that you are the mind behind three supercomputers?

Almost correct. In fact, every supercomputer is not the job of a single person. I want to stress this very much. Even for the first Meluxina machine that was implemented in Luxembourg, the first national supercomputer, there was a team that did the planning, the work. I was very lucky to be part of this team.

That system has now been operational for four years. We are very lucky today to set up what comes next. That is AI-optimised supercomputing, a new flagship machine for Luxembourg and in Europe, that is the computing backbone of the AI Factory of Luxembourg. And also, not a supercomputer but a quantum computer, which will form a trio of high-end computing machines.

The original Meluxina is already middle-aged. Why invest in AI and quantum components for an ageing machine?

It’s a very tricky question. Supercomputers typically have a lifetime between five and seven years. There are two elements that are closely related to this: the advancement of technology – there’s new, better features that enable doing new kinds of things – and the counterpart to this, which is efficiency. Over time, if you want to keep being sustainable and to have new capabilities, you need to reinvest.

Meluxina-AI will, from a timing perspective, be complementary to the first machine. It will be much more powerful for AI workloads, and the software environment around it will be completely different to the first machine.

The quantum machine is very interesting because it’s not a standalone system. It is always meant to be coupled with a classical computing platform. One year ago, the idea was that it would be coupled to Meluxina 1, but then we followed the national strategy for a new machine for Meluxina-AI. So, the quantum computer will now be linked not only to Meluxina 1 but also to Meluxina-AI.

We are not behind. We will be leading the charge.

Valentin Plugaru

Chief technology officer at Luxprovide

Quantum computing is hard to harness, and some say it might never work as we want it to. Is that fair to say or has it changed?

It’s changed. There is this very nice promise called quantum advantage and quantum supremacy. It’s the moment in time where you can run some algorithms on a quantum computer which would not be possible to run on classical computing. From this point of view, we are still far. But there is a community of open-source developers, but also companies that are now investing massively into this: IBM, Google, companies have been created that are working on these topics.

There are many quantum technology pathways that Europe is exploring now: superconducting, trapped ion, photonics, spin silicon. We don’t know which one will be the best in the future. We are trying to build the pathway that Luxembourg believes is the most promising in this direction.

On the software side, many companies are developing tools and platforms that users are adopting, which was not the case four or five years ago. There was no quantum computing standard yet, but we are beginning to see the first standards. There are now conferences and workshops.

How does it compare to AI, which seemed to arrive with a bang, almost overnight?

It depends how far back in time you go. Nothing comes from nothing. For AI, there were tremendous advances. Neural networks came out decades ago. Ten years ago, neural networks started to be implemented and accelerated by GPU technology, which showed it to be tremendously useful for accelerating the workload of neural networks. Deep neural node networks started to be created. The transformers, which are the basis for today’s ChatGPT and the like, were proven to work in the mid to late 2010s.

It didn’t come last year, it didn’t come in the last two years. We’ve been seeing the rise of AI since 2015 and beyond. Quantum is still at the level of being more R&D rather than practical, but it’s following the same path.

What is EuroHPC?

The European High Performance Computing Joint Undertaking, or EuroHPC JU, is an initiative by the EU – plus Iceland, Israel, Montenegro, North Macedonia, Norway, Serbia, Turkey and the UK – that pools resources to develop a world class supercomputing ecosystem in Europe. EuroHPC is headquartered in Luxembourg. It has a €7 billion budget for 2021-27 and currently counts nine interconnected supercomputing centres across Europe, including Meluxina.

On the topic of emphasis and investment, is Luxembourg, and Europe, focusing on the right areas at the moment?

I can never speak for government strategy, though I am advising at European level in one of the advisory boards to the EuroHPC, it’s called INFRAG, Infrastructure Advisory Group, where we’re setting out the multi-annual strategic research programme for Europe.

Typically, what has been said is that you always have investment into platforms and facilities, but you really need to match this with the skill development. The growth of skills is something we need to be focused on for sure into the future. We always have open positions to hire people. Every other supercomputing centre that we talk to always has a skills problem.

Luxembourg has had this idea of digital economy, data economy for some time. It’s writing its own AI strategy, very much in line with the European one. Everybody is working on this today. We are not behind. We will be leading the charge.

Are we moving forwards compared to Asia and the US, or perhaps slipping back a bit?

It’s competitive. There was a very nice quote I remember: “To out-compute is to out-compete”. It’s always been true. If you have access to the latest technology, you can do things that nobody else can do. And it’s the speed of adoption that really matters.

There are things that Europe is doing in a very controlled way: not doing just any AI, but having secure, trusted AI. It requires policy, regulation, strategic view of platforms, technical implementation, transferring skills, working with much bigger consortiums than ever before. Though Europe today doesn’t have the biggest computing infrastructure, Europe will catch up on this.

© Photo credit: Gerry Huberty

Are we moving forwards compared to Asia and the US, or perhaps slipping back a bit?

It’s competitive. There was a very nice quote I remember: “To out-compute is to out-compete”. It’s always been true. If you have access to the latest technology, you can do things that nobody else can do. And it’s the speed of adoption that really matters.

There are things that Europe is doing in a very controlled way: not doing just any AI, but having secure, trusted AI. It requires policy, regulation, strategic view of platforms, technical implementation, transferring skills, working with much bigger consortiums than ever before. Though Europe today doesn’t have the biggest computing infrastructure, Europe will catch up on this.

You can brute force your approach to AI or you can be smart about it. And this has been proven again and again, that you don’t need this level of resources. When you have constrained resources, in fact, you can do tremendous work at algorithmic level enabling you to do fantastic things.

The first ChatGPT explosion came from the US, and Europe at that moment had AI networks and labs working on a lot of things, but the promise of AI was so exciting that Europe had to find new ways. They shifted tremendous financial resources to bring a European concept of AI to life.

Europe is doing things with financial programmes over seven years. There is the Horizon 2020, then Horizon Europe, which is going up to 2027 and the AI boom came in the middle of this period. So, Europe had to pivot.

You can brute force your approach to AI or you can be smart about it.

Valentin Plugaru

Chief technology officer at Luxprovide

While we’re talking about resources, these machines use a lot of energy and materials. Can they pay it back? Are they going to be central to the green transition and mitigating climate change?

Yes, for sure. I will give you an example: a flagship project of the European Commission called Destination Earth. They’re doing digital twins of the Earth, one digital twin for weather simulation, and one digital twin for climate simulation. The climate is a super-grand challenge because there are so many elements to take into account. What happens if we reduce emissions by this much? What happens if we do this? What happens if we do that?

What you may not be aware of is that when the European supercomputers were built – the big ones like Lumi in Finland, MareNostrum 5 in Spain and Leonardo in Italy – they were always planned from the ground up to support these grand challenges, for example, climate research and forecasting, scenario making for decision-making.

 Meluxina was one of the first machines to be operational in the EuroHPC network, and the team working on Destination Earth requested that Meluxina be added to the list of supercomputers running this initiative. There are only four machines in the European network today that are running this workload. It consumes power, of course, but there are projects working on it that are aiming to improve the lives of everyone, not in the short term, maybe, but midterm and long term.

You say Luxprovide started off going here and there, talking to people and bringing knowledge back. Have you now become the experts that others come to learn from?

I’m happy to say that we are part of these EuroHPC coordination meetings with the others. We are happy to see those others asking us questions, learning from what we do. They have decades more experience. Our computing centre in Finland, Lumi, one of the largest ones in the world, they have 30-40 years of experience. Luxprovide is five years old, almost. So, being able to be on the playing field is already a tremendous achievement.

We have a very innovative concept ahead. Others are telling us, “You’re really at the forefront of what you want to achieve.” I see that we will be at this leading edge for some time. Of course, there’s the concept of what we want to be, and then there’s execution, but we have the ambition for the execution to match our concept.

The government seems to match that ambition. They talk about it all the time and that must be positive for you to hear.

It’s indeed something unique in Luxembourg. This superb close collaboration at all levels, being able to be close to decisionmakers in the country and for them to have such fantastic long-term vision. The data economy I mentioned before, the supercomputing vision, vision on 5G, vision on national data service, all of this came because there was very close collaboration at the national level and the strategic view of where we should go.

It used to be called the Smart Specialisation Strategy and it may not be much in the news today, but we are seeing the effects of this concept in Luxembourg very deeply. Other countries are doing it too, but that it’s happening in Luxembourg to this degree is nothing short of amazing and very few can match this.

This interview has been edited for length and clarity. It was first published in the spring 2025 edition of the Luxembourg Times magazine.