🔴 Hearing with Octave Klaba, CEO of OVHcloud, on his group and current digital challenges

Hello. Thank you, Mr. Chairman. Ladies and gentlemen members of parliament, I thank you for this invitation to speak before your committee, where industry, energy, innovation, and digital technology intersect. These are all subjects at the heart of cloud activities. Mr. Chairman, you recalled our successes but also the size gap that still separates our activity from that of the hyperscalers. This gap is real. We are going to ask this question. They all essentially boil down to one. How can we scale up? Critical mass—I will return to this after a few introductory words about OVH. OVHcloud is 27 years old. I created OVHcloud in '99 in Roubaix, without any capital. For over 16 or 17 years, I grew the company without any capital at all. The first round of funding I did was in 2016, for 250 million. The second round of funding, so to speak, was the IPO in 2021. We also did a capital reduction where I returned 350 million to investors, so the company grew with very few external resources. Yet today, we generate over 1 billion in revenue. We have 3,500 employees. We are deployed in Europe, the United States, and also in Asia. But it's a story of freedom and independence, probably because my personal history led me to be a little bit paranoid about how to build companies and see them over the very long term. When I started OVH, I was already thinking 20 years ahead, and today that is still the case.

Today, the cloud meets every cloud need, and soon it will meet more and more of what we now call artificial intelligence needs. I mentioned our revenue, but you also have to see that we serve different types of clients. We have, of course, the public sectors in France, but not only there. In Germany, in England, and what we call the Rest of Europe, meaning England, the Netherlands, and Poland. We are also in Canada and the United States. So we have these business sectors. But if we had focused solely on this type of activity, we wouldn't have had the critical mass needed to build a company of this size. And that's why, coming from digital, we'll say, the digital sphere. Where customers can simply come to our websites and order services with a simple bank card, we have that business too. It is very important because it currently accounts for about 60 to 70% of our revenue. Corporate, private, and public companies represent about 20% of our revenue. And so, it's really important to understand that in our business, critical mass and critical size are essential to build this type of company. Because the investments behind it are so significant that if you only target one type of client, you won't get a return on investment and you won't be able to keep investing every month, every year, and keep growing the company.

Our ambition is to reach 2 billion in revenue in the coming years. I won't give exact dates today because we are a listed company, and as you know, it's highly regulated, and investors scrutinize these types of investments. I will answer the questions you asked me, and for the first question, I will answer it right at the end. What changes to public procurement rules? As I said, it's about critical mass; for instance, take OVHcloud in France, the government's public cloud is 84 million euros annually in 2025, and OVH is number one with 33 million. 33 million, compare that to 1.2 billion in revenue. So you see, it's great that there is revenue, but it's more in the tens of millions of revenue. And compared to many of our competitors who are more modest in size, unfortunately, this type of revenue in the tens of millions. It's fine for a growing player, reaching a billion in revenue, which is European, global, has a presence in many countries, and has clients in 140 countries, but this public procurement, around hundreds of millions or fifty million, is actually anecdotal compared to the overall growth needed for a cloud and AI company. But it's good, it's good.

It has changed in recent years, and we have clearly seen this gradual shift in public procurement to start using European infrastructure. We host AIFE, for example, which is the electronic invoicing system implemented in France on September 1st. So for AIFE, we host many intermediary providers, and behind all that, these elements are part of our clients. But these are quite modest details today compared to more global growth. Cloud contract from the European Commission. It is very interesting because following various shocks we have experienced over the last 2 or 3 years, the American administration is part of it, but not the only one. There is another one called VMware, which was bought by Broadcom. All these shocks that happened recently, but in a somewhat unpredictable and unforeseen way, have triggered reactions within private or public companies. There are two types of reactions. The first reaction is, "Well, this will pass, things will eventually return to normal and it will all be fine." So we keep a low profile, and eventually, everything will sort itself out. And then there are others who say no, we really need to change the approach. And the European Commission is part of that group. They created a reference framework. They used these frameworks to be able to select European players. We were chosen as one of the winners alongside Deep and CleverCloud for a 180 million contract over four to five years.

It's good. It's good because it's a first step. I think it's the method that we should focus on more than the amounts currently at stake. It's true that 180 million over five years for four players isn't exactly what needs to be done yet, but the method is what interested me most. That is to say, the European Commission says: "We are going to put up 180 million to test and develop with you, to give you the chance to go after subsequent tenders from different agencies and prove that you are capable of handling all, or at least some, of the activities currently hosted by the hyperscalers." I think it's super interesting because you have to keep these 180 million in perspective, which will roughly represent about 40 million per year across 4 winners. You're facing about 1.2 billion in revenue that AWS generates with the European Commission per year. Per year, okay? And you have another 500 million with Azure. So you just have to put that into perspective. Meaning you have 1.72 billion in revenue generated between the European Commission and the hyperscalers, facing this first call for tenders. So you have to look at the method; it's a big business, but it also involves a lot of responsibility. It's a lot of products, a lot of things for which you need proof. What we call proof points in English, which means being able to prove that we are a trusted player, and that's why the method interested me more.

If you look at Europe more globally, there is about 10 billion in revenue generated between the various governments, commissions, and ministries with hyperscalers in Europe. So the critical mass isn't really there; it's not about creating new budgets. It's actually about redirecting them toward European players and being able to trust them to help them grow. And so, there is the money, the money is already there, it just takes a little bit of courage to ensure that this money goes to European players. Another question you asked me about the SecNumCloud and the other platform we just obtained: what are the next steps and difficulties? I like what was done by ANSSI because they provided a technical framework, unlike all the other standards which are just standards saying "you figure out how you want to do it." Thus, they went much further, much further by stating how it must be done, bringing a certain expertise that they have on securing infrastructures. And that is why it is extremely complicated and long to obtain, because it's not just about saying "I have papers, I follow processes," but about how you build the infrastructures. And so we fully embrace these frameworks. We are already used to it; that's three certifications we've passed with them, and we will continue to enrich our products. Today, on the cloud platform, we have a certain number of products that we are going to enrich. The goal is for this portfolio to be as rich as possible and to benefit all clients, including the state and private actors.

We have 46 data centers in France. What guarantee of service continuity? It comes from the technical design we do around the data centers, but also around all the activities that take place in those data centers. The operating principles—having one data center—that's in the past. That's from the 2000s; since 2020, and in the years to come, it's more about distributed data center services. Meaning you have what we call a "region," a region of three data centers spaced 30 km apart, powered by different energy sources with different fiber optics, so that everything clients run can withstand a failure without impacting overall operation. And so, this typology of distributed designs is what we are currently deploying, changing the entire configuration of our data centers. You have the same type of principles in AI inference. Meaning that today we can clearly see that inference, so everything that allows you to type in the chat and receive tokens, we see the criticality of these types of activities and so it cannot go down. Right? It is just as important. It is becoming as important as, for example, telephony or all of communication because there are many companies that rely on tokens, so you cannot put all of your activities in a single data center in a single location. You must put them in many different places and be able to connect them on the network through different schemes. In any case, that is our approach to being able to provide resilience.

Electricity, land, water, the barriers to investment in France. Well, there is a long, long list, actually a long list of regulations. I started 27 years ago. Right. So I have a little bit of, yes, I have a track record, in fact, on managing different issues and in different countries. It has become more complex in recent years, to the point where it takes several years to be able to deploy a data center or even be able to upgrade one. You just have to look at the ICPE level; when you have a data center and you want to expand it, you must of course submit the file, it's normal. Right? But the act of submitting the file blocks all investments on the sites for 2 years. Meaning you are prohibited for 2 years from doing anything at all on the sites. So I challenge you to manage an activity where you have growth of this type where you need to grow quite quickly and you are told don't touch anything for 2 years at that location because we are studying your file. We will tell you if you can grow in 2 years. It's that you cannot. Right? You can, and the list of these types of constraints is long.

I will give you the example of our site in Roubaix. So we are from Roubaix, we gradually bought different buildings in a neighborhood. Right. So each time it is a different address. The buildings are spaced 200m, 500m apart, sometimes it's 700m. And then, well, gradually, we did ICPEs for the different locations. But in fact, since they were 2 MW, 5 MW sites, it went, it went relatively well. But at one point, they came to us saying you must consider all of this as one single location. Right? And so from now on, they want everything to be considered as if it were a single address, the entire district. Right? So we were blocked for a certain amount of time to be able to bring all of that into compliance, so to speak. They imposed rules on us, saying well, different buildings, different locations, it doesn't really matter, you must consider them all together. So you have a certain number of investments that you must make because of a certain number of constraints. We had to put in, I think, 5 or 6 million extra to be able to bring, in quotes, a neighborhood into compliance. Right, with the complexities of distances, well, all of that, so that it's just an example. Okay, I mean, we live with that, and it's a bit longer than not having all those constraints, which are important in absolute terms, they are important. Sometimes, it's because it protects us, it ensures that we operate in a safe place. But still, it's the reality for entrepreneurs, it's the reality of this activity.

And today, when we talk about data centers, I don't think we will have gigafactories in France, or even in Europe. Meaning that places where there will be 1 GW of electric power deployed in one data center at one address, for me, will never exist in Europe or in France. The Americans are building for, well, with all the feedback we're starting to get, actually 2 GW or 3 GW, with all the issues of CO2, energy, off-grid, water consumption, and so on. Well, in France, that will never happen because it will simply be blocked beforehand. If you want to build a 100 MW or 200 MW data center today, it already takes you between 4 and 5 years to get there, even with everything that has been done previously by the president to push this entire activity.

And what are your recommendations for the legislator? I like the freedom of action, actually, and that's what I like, is that, somewhere, personal initiative, individual initiative can change things. That someone gets up in the morning, has an idea, takes the risks, creates businesses, and so I like to let companies act and, behind that, give as much freedom as possible. But also, I realize that we live in a slightly undisciplined country, at least one that isn't focused on the long term, which is satisfied with a certain rhythm of operation or activity from one year to the next where change, well, we don't really like that part. So, I just have one idea, which is that somewhere—and I don't like it being done by law, but in any case, it's really a reflection I'll leave with you—it's 15% of activities. Meaning there's no need to redirect, uh, 30, 40, 50% or earmark 50% of digital activities toward European players. If you put in 5, 10, or 15%, you will create giants in Europe, in France, in the digital sector. That is enough because it's a certain micro-critical mass, which is already not so micro, that allows you to develop, to invest, to rotate capital, and to ensure we have more of these types of companies in Europe and in France.

And the second element I would mention, though I don't know how you can intervene here, is financial culture. Financial culture in Europe, in France, is very much oriented toward: I invest, I lose money. Right? So, somewhere, are you going to invest money? Are you going to be capable of losing money? Profitable investment, making investments that then create value or generate growth. We are in this "startup nation" approach. Somewhere, we've lost sight of that. I created the company without outside money, and that is my DNA, really. I don't know how to do it any other way.

And that is what we are missing today. On one hand, we have investors who want cash, free cash flow, dividends, and on the other hand, we have a whole population saying, "Well, I want to raise money to...for this m...to do development, to do things that amuse me, but I'm not really looking for revenue, or activity, or anything sustainable." Fortunately, between the two, there is still a slew of companies that are on the right track, but insufficient in scale in my opinion.

As I was saying, actually, thank you. When I said that, it's actually 15% of revenue redirection, or at least of orders, that allows for the creation of European players of a decent size. We also have to take into account the issue of critical mass, meaning that France alone does not have it today and cannot solve this problem. The answer must be European. If you don't have the European critical mass to provide revenue to French or European companies, you will have champions like those currently on the ground, around 200, 300, or 500 million in revenue, because that is the revenue cloud companies will reach if they only focus on France. Beyond that, you can create multi-billion dollar revenue champions because you will have access to several markets. So why is it that today OVH reaches 1.2 billion and is on the path to 2 billion in revenue, while all our competitors are stuck around 100 or 200 million? Because in 2004, after 5 years of the company's evolution, I decided to deploy to every country in Europe.

And so I did that for 4 years. So I covered Poland, Spain, Italy, Portugal, Ireland, England, the Netherlands, the Baltic states, Germany, and Switzerland. So I did a lot of European development from 2004 to 2009. In 2010, we were asking ourselves how we were going to get into the United States. We didn't go right away, we went to Canada, and so in 2011 and 2012, we also expanded through Canada and also through different data centers in France and Europe. In 2016, for me, it was the final moment in our history to say we must become a global player. And that is why I raised 250 million in 2016 to go and build data centers all over the world. That's when I went to the United States. I went to Singapore, I went to Mumbai, I did Australia and other countries in Europe like Poland, Germany, and so on. The critical mass. And so, the stakes, which are the investments, are so heavy that if you only have one single country you can count on, that's fine. And...I think, actually, we need to redirect a portion of the revenue towards Europeans. But if France does not have this activity and this initiative is not European, we will not manage to build multi-billion dollar companies.

Many questions. I will try to answer some, though I won't be able to answer them all. Regarding artificial intelligence, for me, it is essential. For me, it is essential and it is a real challenge. Today, technologies are mainly American or Chinese. Right? And innovations come largely from China. In Europe, we don't have any tech players willing to invest in artificial intelligence, in the actual technology of artificial intelligence. Of course, you have use cases and how to implement them in companies? How to make it part of daily use in processes and increase company productivity. But there is the underlying technology, which is extremely important. Today, many place a lot of trust in open source just because it's good, but it is Chinese. Meaning we don't know what will happen tomorrow with that source, and we don't know what the strategy behind it is that companies will adopt in one, two, or three years. Is this open source temporary to make us dependent on this type of model, or is it permanent? Will there be companies willing to invest billions while continuing to outsource the models? In any case, we have reached the conclusion that it is too risky not to invest in artificial intelligence. That is why, since the beginning of the year, we have started acquiring companies with talent to strengthen ourselves technologically from within, so we can then train models from scratch and enrich them for the fields we need. And we have three of them. The first field is around security, offensive or defensive. Meaning we see very clearly to what extent security issues surrounding artificial intelligence have become colossal. These are real problems because AI allows for discovering security flaws or penetrating systems at a speed never seen before. So, at our level, we believe we have a lot of our own data—not client data, our own—because we manage large data centers, we manage enormous amounts of activity, we deal with many DDoS attacks, and we have a massive amount of data we can use to train our systems. That's where we come from, right? We must be good at security. The second element is coding, developing software. We use LLMs for development, but we also want our own LM models to develop software. And third is agency, so we can have agents for our business areas like finance, legal, HR, and the supply chain. And so we believe that if we develop these LM models for ourselves and our needs and are happy with them, we will also make them available to our clients so they can benefit from them. We could talk a lot about artificial intelligence, but that is a very quick answer.

Thank you, Mr. Deputy, Mr. Chairman, CEO. I will start with the last question on ecology and data centers. That is one of my favorite subjects because since 2004 we have been cooling data centers with water. All right? We don't have AC in our data centers. We have 500,000 physical servers, meaning physical computers that are cooled with water, fluid. So, what does that allow us to do? It allows us to cut our electricity consumption in half. So it's not, it's not anecdotal, it's a lot. And since our, all our data is public, well, I can say today we pay about 70 million a year on energy, energy, in fact, in different data centers per year. If we didn't have those technologies, we would have paid 140. All right? So these are real savings we make that we pass on, in fact, that we are, and it allows us to be cheaper on the market and to offer cheaper services. So we pass this added value on to our customers. These are types of technologies that we developed over 20 years ago and that we have improved over time. The same types of innovation we have done regarding the circularity of parts. Which means that we build all our servers in our factories. We have two factories. We have more than a hundred people, 120 people, who build our servers. We don't buy servers elsewhere. We have the entire global supply chain where all components arrive in Croix, in the north of France, and near Montreal, in Beauharnois, where we have two factories in which we manufacture a little over 50,600 physical servers per year, which allows us, in fact, to customize, to really...

reduce costs, and to provide servers that are specifically suited to the various businesses we serve for our customers. And so, in this way as well, once the servers go into the data centers, after a certain number of years, well, in fact, they have to be removed, recycled, and... So, this whole recycling chain is something we have had in place for years. So, for our data centers, we have built many of them on brownfield sites to reduce CO2 consumption. And again, I could talk about this for thirty minutes. It is an extremely important subject for us, in order to precisely reduce both the impact and consumption, and subsequently to be able to deliver services at a large scale.

On the supply chain, it is a primordial subject. I will just give you a figure; in fact, right now we are experiencing a crisis in the cost of RAM. Memory is the memory you have in computers. You have it in cars, you have it in smartphones. The cost of RAM has, from memory, increased 6 to 9 times in the space of, let's say, 12 months. Right? So it's a major impact on the overall costs of cloud and artificial intelligence. And it keeps rising for a simple reason: much of global production has turned to AI, and the global output of the three main suppliers has been redirected specifically to GPUs and AI, leaving other activities behind. And so, supply and demand—the demand is still there. Supply has driven up prices because global production hasn't increased. Uh, Europe buys 3% of the world's total RAM volume. Right? 3%. Europe, the entire European territory, buys only 3% of RAM by volume. Meaning that's not where the action is. The production, all the assembly, it's not happening there, actually. The assembly of all these electronic activities. That's happening in the United States and in Asia. So, Europe hasn't been up to date, or rather, it hasn't been current in terms of electronics, for years. Uh, we still have Bull, we are practically the only European computer manufacturer. There aren't many others. There are a few factories in Turkey for HPE, but there are extremely few. It's not manufactured in Europe anymore. It's far, far off; we are dependent on all these activities that are actually outside of Europe. I'm not even talking about chips, and so on. Even the United States is struggling to bring all this activity back onto its own soil. Regarding artificial intelligence, there is a lot, a lot to answer. I think that I...

So regarding AI, in relation to the questions you had, innovation in artificial intelligence and the development of new models requires having GPU clusters in our data centers. Well, it turns out that I invested in quite a few data centers about ten, fifteen years ago. I have 250 megawatts of electrical power. So, about more than half is in use, but we'll now be able to accelerate the use of the other half. Uh, and so I am a bit ahead, actually, compared to all the phases of all the players saying, "Well, I don't have any more electrical power, I don't have any more space," and so I am a bit more comfortable today scaling up artificial intelligence. Still, the technological designs of the data centers that they use—that host artificial intelligence—are totally different from other types of infrastructure. The concentration of energy is enormous per square meter, and that requires, in fact, a certain number of electrical cables or completely different electrical systems. And that is where history proves us right again. Water cooling comes along to save the cooling, since you cannot put that much electrical power into a square meter without being able to extract that much heat without glycol water or water. And so, being experts in that technology for over 20 years, well, we can clearly see that we master this entire GPU supply chain, and we have deployed at an extremely fast pace in our data centers. But it changes a lot in terms of space and the way it's brought in, and behind that, you just have to take into account that the technologies are still evolving. Meaning that if you invest in a data center today and say, "There, I have an artificial intelligence data center," be sure that in 18 months, 12 months, you will have to retrofit your data center to adapt it to new technologies, to the GPUs coming out in two years, because your energy or cooling chain will no longer be up to date. And so, it also completely changes the way we invest in data centers. You cannot build a data center today and say, "That's it, I've built something that will last 20 years." You can build shells in which, over time, you will deploy technologies that are incompatible with the GPUs of the moment.