- 6 hours ago
Digital sovereignty has moved from a policy debate into a strategic opportunity for regional growth. While the global tech stack fractures, local ecosystems are finding new room to thrive by building infrastructure that follows their own rules rather than those of a few distant giants. This shift toward "buying local" allows nations to reclaim control over their data, talent, and supply chains, turning what was once a dependency into a competitive advantage. Business leaders now face a map where national priorities and innovation must move in sync to succeed. Can this push for autonomy spark a new era of regional tech powerhouses? How do we balance the need for local control with the benefits of a connected global economy?
Category
🤖
TechTranscript
00:00Anna Rold Reviewer
00:23Well, good afternoon, thank you for being with us.
00:27My name is Anna Rold, and I will be your moderator or orchestrator today.
00:32And I'll begin with letting my panelists, my wonderful speakers today,
00:38to introduce themselves.
00:39So I'll start, ladies first, with Clara.
00:42We introduce ourselves.
00:45Okay, I was not right for that.
00:47Very, very happy to be here with amazing people on stage
00:50to talk about such an important topic, which is sovereignty,
00:54which I'm so happy we hear so much about at Vivitech this year,
00:58which was definitely not the case before.
01:00I'm Clara Chappez.
01:01I'm the French ambassador for AI and tech.
01:04Prior to that, I spent one year in the government on similar topics,
01:08AI and tech as well.
01:10And prior to that, I headed the team,
01:14which is responsible for startup policies within the Ministry of Economy for three years.
01:18But before that, I come from the private sector.
01:21So I built tech for 10 years.
01:23I spent quite a lot of time in Southeast Asia, in the US and the UK.
01:27And at some point in my career, there was this president.
01:32I mean, he was not president back then, he was candidate.
01:34Emmanuel Macron started talking about how France was a country for innovation.
01:39And that resonated super strongly with me because I have met entrepreneurs all around the world.
01:46And French entrepreneurs were always like, oh, we're building stuff here.
01:49We couldn't build home.
01:51And that really annoyed me.
01:52And that's how I decided to move back.
01:54And that's how I decided to try and help shape what you now do,
01:58which is building and from the other side.
02:00I wanted you to introduce yourselves because I thought that so much interesting information comes out from that first introduction.
02:08But thank you for giving us a little bit of a flavor about tech diplomacy in France.
02:15Please.
02:17I'm Antoine Beausleau.
02:18I'm a professor at EPFL and one of the leads of the Aperitus LLM project that designs very large-scale,
02:27open, compliant, multilingual LLMs for the rest of the world as a trustworthy foundation.
02:36I guess that's pretty much it.
02:38I was also once in the United States, as the accent kind of indicates, though I was also a French
02:42transplant there.
02:43But I moved back to Europe around four and a half years ago to Switzerland, not France, sadly.
02:48But it's still a nice place as well.
02:50You can always come to France.
02:51That's true.
02:52Very easy.
02:53Well, I did right now.
02:56And Raphael, do you want to go next?
02:58Hello.
02:59I'm Raphael Lofant.
03:00I'm Chief Operating Officer at Proton, a European company headquartered in Switzerland.
03:05And I'm an engineer by background.
03:07I started in computer science and networks in Southeast Asia for six years, then California for three years, then England
03:12for a few years.
03:13Then moved back to France and then now in Switzerland.
03:16And Proton, as a share later, is a company that was started 12 years ago that is focused on privacy
03:22and confidentiality of communications.
03:24And we went from three founders to 650 people now today.
03:29And I'm Chief Operating Officer based in Geneva.
03:31Congrats.
03:32And they have the most fun Instagram account.
03:35Just so you guys know.
03:36Damien, we're going to go next.
03:38I'm Damien Lucas.
03:40I'm CEO of Scaleway.
03:42I have a background in tech, actually in engineering.
03:47And in my early life, I was a big fan of open source.
03:50I'm still.
03:51But at that time, I was also a developer.
03:53I'm not anymore a developer, unfortunately.
03:56I was part of the original team of VLC.
03:59Put that open source.
04:00I'm sure you know that.
04:01It's one of the most well-known French piece of software.
04:05And as to that, I had an entrepreneur story.
04:09And finally, joined Scaleway to move Scaleway to the next phase as the European cloud, truly sovereign.
04:16And I think this will be part of the topics we will be discussing today.
04:20Well, we'll dive right in now that we know you so well.
04:23Thank you for taking that moment.
04:25I think actually it was quite important.
04:26I wanted to give you, if you might allow me, a bit of a context on our topic today.
04:31Because in my world, in the policy world, when we think about the word sovereign,
04:37we think about flags and nations.
04:39And we think about something very concrete.
04:41But for our conversation today, I wanted to be a bit careful with the word sovereignty.
04:45And not sound like we're talking about flags and borders.
04:50The conversation is more today about risk and dependency.
04:54Where are we actually exposed?
04:57Which dependencies we can live with?
04:59And which ones can become a liability?
05:01Or even a lever, a powerful lever at that.
05:05For example, in my research, I found out that a surprising share of European companies run
05:10their most basic operations of infrastructure governed by foreign law.
05:16And that's a big deal.
05:17And I wanted to know, was sort of the first question that I had for all of you,
05:21where's the opportunity or maybe the challenge or even they could be both.
05:27And so, with AI, the stakes are even higher.
05:31Because, you know, for AI to build a technology like that, you need not just one or two things.
05:37You need multiple things.
05:39You need talent.
05:40You need to build data.
05:42You need infrastructure.
05:43So, the stack is so much bigger.
05:45So, I've got four people that know everything so much better than me on this one.
05:49So, I'm going to dive right in.
05:51I'm going to start with Damien first.
05:53When you say a cloud is trusted rather than sovereign, what is the risk you're really trying to neutralize?
06:02So, what's a sovereign cloud?
06:05And basically, what are the risks that we want to cover in a sovereign and a trusted cloud?
06:15The first one is a risk around the data, okay?
06:19How is the data secured?
06:21And here, one of the risks is the risk of extraterritorial laws.
06:26We know about the Cloud Act.
06:27We know about the Patriot Act.
06:29And each different nation has this kind of point.
06:32So, at Skyway, we're building a cloud where we're totally immune to those extraterritorial laws.
06:39This is very important.
06:40And how we're doing that, we have no shareholders outside of the EU.
06:44We have no subsidiaries outside of the EU.
06:46And we have no staff outside of the EU.
06:48So, this way, we can make sure the data is safe.
06:51And then, second risk, and we've seen that last week, last weekend actually, is the risk of a kill switch.
07:01We can discuss that, but what we've seen last week is that there is a kill switch that can actually
07:07prevent any non-American to access the latest technologies.
07:13That exists, and the bottom exists.
07:16So, in terms of cloud, what we're doing at Skyway is making sure that we have no dependency to foreign
07:24technologies.
07:25And, obviously, when I say foreign, I mean extra European.
07:28This would be my definition of foreign for tonight.
07:34So, how do we do that?
07:37The easiest part is on the software.
07:39We develop everything.
07:41A lot of open source, obviously.
07:44I'm a big fan of open source.
07:46A lot of open source.
07:47And a lot of work.
07:48And we have our own software Slack.
07:51And then comes the hardware.
07:53And on the hardware front, we have a problem, especially in the world of AI.
07:57We rely on AMD a little bit, Nvidia quite a lot, both Americans.
08:08We need GPUs in Europe.
08:12We need European GPUs.
08:15So, what's our responsibility as a cloud provider?
08:18We need to purchase them.
08:20They're not yet fully ready.
08:22No problem.
08:23So, what we announced yesterday, we placed an order to a company called Visora, which is actually building European GPUs.
08:30They are not fully ready yet, but as soon as they will be ready.
08:34And given the fact that we are ordering them, they will be ready at some point of time.
08:38And that is going to be one dependency less.
08:42And then, data centers.
08:44We deploy our cloud into European data centers so that we can guarantee that we have no technology dependency.
08:52Interesting.
08:53And, you know, this notion of full sovereignty.
08:58We talk about tech sovereignty as if it's a different thing.
09:00For you, Raphael, you've said, full stop.
09:03Tech sovereignty is sovereignty.
09:05So, my question to you is, what's that one dependency that should worry European leaders most right now when it
09:12comes to this subject?
09:14Yes.
09:15I mean, sovereignty, as I said, is, yes.
09:17Tech sovereignty is sovereignty because you need to look at the full value chain.
09:21Starting from the processors, the GPUs, then the software stack, then the SaaS on top of it.
09:26And what we need to do as European continents, we need to do step by step.
09:32It's not going to be overnight.
09:34We need to invest in it, like Damien said.
09:36And when we looked at cloud dependency, the cloud is highly dependent on the U.S. hyperscalers.
09:44We did also on the email and communication services.
09:47We did a study, like we looked at all the domain name servers across the European public listed companies.
09:53And it came out that 74% of public listed companies in Europe rely on U.S. email services.
10:01So, when you have this level of dependency on the cloud, close to 90% on the U.S. cloud
10:07providers,
10:09this level of dependency on email providers, 74% across all the public listed companies,
10:13is not a partnership anymore with the U.S. providers.
10:17It's a full dependency.
10:18And with this dependency, basically, we can see that Europe is a digital colony of the U.S. right now.
10:25So, we will not change that overnight.
10:28And this dependency, as Damien said, is both on the cloud act part, access to the data,
10:32and the kill switch part, which we saw recently, but which already existed in the past.
10:37So, what we need to do is to have European public institutions
10:43and also European private companies invest, like Scaleway is doing, investing in the new chips,
10:51but at the software layer also having European companies saying,
10:55okay, I don't want to leave you this risk of this kill switch anymore.
10:59I need to invest in a software infrastructure where I will not depend on a kill switch anymore.
11:05I mean, I felt like you were ready to say something, but both of you have sort of similar message.
11:11Hardware and software, we need to decouple from and be completely sovereign.
11:17And just to clarify, we're talking about pan-European sovereignty, not just France or just Switzerland
11:23or just Germany, for example, correct?
11:25Actually, there's a risk of sovereignty being seen as a national level because we will not be able to make
11:34large European-scale companies.
11:37I mean, I want Scaleway, I want OVH Cloud to be successful across Europe.
11:42I want Earthner and Cloud to be successful across Europe.
11:45I want Proton and other similar providers to be successful across Europe.
11:48If we're not successful, if we don't become multi-billion dollar companies in a few years,
11:54we will still be in a situation of digital colony in Europe.
11:59I understand.
12:00Clara, from your vantage point, I mean, you speak for France, of course, as a tech diplomacy politician.
12:09Tell us a little bit from France's perspective.
12:12What does that sovereignty really mean?
12:14Yeah, I think what's interesting, just to add up to the point that was made,
12:17is that it's not even just a Europe question.
12:20Part of my job is to go around the world and advocate for what we're trying to build,
12:26which was exactly rightly framed.
12:29Not to say we want to do French tech for French companies and on the whole value chain.
12:34That's not the point.
12:35But being able to understand, where are your dependencies?
12:39And decide for yourself, because I think that's something we haven't done for the last 20 years.
12:44We just went with a solution that was on the shelf and just didn't really realize that then,
12:50suddenly, 20 years later, 90-something percent of the cloud market is concentrated in the hands of three U.S.
12:55companies,
12:55and the risks that's associated with it.
12:58But this conversation of how do we decide, where do we accept some risks, some vulnerabilities,
13:07where do we decide for ourselves?
13:09This has to be a part of the value chain that we need to own as Europeans.
13:13Where do we partner?
13:14It's a conversation I hear everywhere in the world.
13:17In Japan, in Korea, I was in Kenya just a few weeks ago for Africa Forward,
13:22which is a big business summit we hosted.
13:26And part of the summit, a segment of the summit was on the eye.
13:29And we had a conversation with 15 head of states from different African countries,
13:34companies, scientists, civil society.
13:37And if I would have shut off my eyes for a minute,
13:40the conversation was exactly the one we're having on stage.
13:43All the African leaders were like,
13:44we don't want to be stuck between having to pick between Chinese tech or U.S. tech.
13:50We want to be free and independent and have autonomy.
13:54And that means building with Europe.
13:56So the conversation goes even beyond making you succeed in Europe.
14:00It's how do we build some kind of a coalition?
14:02How do we work together?
14:03How do we make sure that when we go to this conference,
14:05and that's what we did,
14:06we have French companies or European companies like Schneider,
14:10Etelsat, et cetera, that are finding markets
14:12because it's aligned with what we're all trying to collectively build.
14:17And I want to add one thing to the conversation.
14:18We talked a lot about the risks.
14:21And, I mean, that's definitely like a very important aspect of this conversation.
14:26You talked about the recent events and export controls and the likes.
14:30I think it has become real to everyone and probably every CEO in the room
14:34that technology is not just technical.
14:37It is super strategic.
14:39And that can basically kill off your whole business and put the whole country at risk
14:43if you're thinking AI applied to healthcare or AI applied to education.
14:47But I also want to add on the opportunity.
14:50Because I think as European, we're very proud of the way we've built Europe,
14:55meaning this is a place where we care for each other.
14:58This is a place where you have access to healthcare, education.
15:01But all of that has a huge cost to our citizens.
15:06And the reality is that because we didn't think strategically about tech in the last tech revolution,
15:12because we didn't create those really big companies that are creating a lot of economical prosperity,
15:18we also missed out on growth.
15:21And we cannot afford our model if we're not going to grow.
15:24So we don't have a choice.
15:25I mean, if we want to continue being Europe,
15:27obviously we need to be like taking steps to reduce the risk,
15:30but we also need to grasp fully the economic potential associated with AI,
15:35because this is a tremendous opportunity for us as Europeans
15:39to just build for ourselves and build prosperity that we need for our lifestyle and our values.
15:45So I think it's an extremely exciting moment.
15:48And also like there is no fatality.
15:51Everything needs like is still to be built.
15:53And when I see people on that stage,
15:55I'm like extremely hopeful and optimistic that we're going in the right direction.
15:59This is very interesting because obviously when I set the stage,
16:02I wanted to make sure we didn't talk about nations and borders.
16:06And obviously that we're talking about the big union here, the European Union.
16:09And so for you, Antoine, you see it in the intersection of research.
16:17You kind of see things a little bit differently.
16:19So is there room for cooperation here?
16:23Does it all sound like to me as the American in the room, in the middle,
16:29that maybe we're losing you as partners?
16:34I think there's definitely room for cooperation.
16:38In fact, I'd say that it's absolutely necessary.
16:41I bring a perspective from the Apertus project on this,
16:45which particularly strives to have a collaborative and cooperative enterprise for building LLMs.
16:51And for me, I think that stems from the fact that I think one of the most fundamental issues
16:57that comes into play when we talk about AI sovereignty or sovereign AI is just the capabilities of the systems
17:04themselves.
17:06You know, we can speak, I guess, about where the data is housed and where the data centers are,
17:11which is an insanely important problem today.
17:14But as we look towards the future as well,
17:16we have to start thinking about the ways in which AIs are actually going to be deployed and used in
17:22the future.
17:22They're going to be a base for pretty much everything we do in enterprises, in government, in schools, in hospitals.
17:30And ultimately, you're going to want the most powerful AI imaginable.
17:35And if you actually want to have sovereign AI capabilities,
17:38you need to be able to construct those yourself so that you can make sure they're aligned with the goals
17:43that you actually have.
17:44I think there's this kind of narrative out there today that, you know, AI is really a function of compute.
17:50And if you throw enough compute at the problem with AI, it's going to solve it.
17:55But the truth is, that's not quite true.
17:57You know, AI doesn't just spend compute, it kind of directs it towards the goals that it is given
18:01and the goals that it initially natively has from how it was trained, from how it was designed.
18:07And if you kind of miss that second part, you can end up deploying AIs that are completely unaligned with
18:12the goals that you actually have a priori.
18:14And so a lot of our focus is, okay, well, how do we actually make sure that we can design
18:19AIs that are quite powerful enough,
18:21but still kind of locally controlled by the actors that want to use them?
18:24And I guess to, you know, jump on the bandwagon of talking about the news from last week,
18:29I think the most important thing that I saw in the news cycle about, I guess, the blocking of the
18:34Fable release,
18:35or the Fable release initially, was that they had managed to find a way by which the AI would lie
18:41to you
18:42if you asked it how to design other frontier AI systems.
18:45As somebody who works in this space, that was, you know, quite alarming.
18:48A, that we could now no longer use these tools to help us self-improve,
18:52but also that they essentially announced that they could make AIs lie to us in an undetected way
18:57if we started asking questions about things that the company itself did not want to answer.
19:01Like, that's an incredible technical leap, and it's also a particularly dangerous one
19:06when you think about being able to use models that come from other actors
19:09without necessarily having a full view into what is possible.
19:13So, at least for me, it's incredibly important that, you know, we kind of combat that impulse
19:17because, at the end of the day, right now, the only people who have these types of sovereign AI capabilities
19:21are in the US and China.
19:23And if Europe and the rest of the world, you know, actually wants to play in that same arena,
19:27then they need to work together and they need to find transparent bases by which to design AI.
19:31Very interesting.
19:33And, indeed, I saw those news, the same news as well.
19:36And all I could think of as a main question to all of you is, as I was writing, you
19:42know, the notes,
19:44is AI sovereignty even achievable at this point?
19:49Like, do you all believe, like, is it a yes or no?
19:51Is it achievable?
19:52What do you all think?
19:54Maybe it's a quick rapid fire.
19:56Yes?
19:56No?
19:57Clara?
19:58I mean, you don't do my job if you answer no to that question, and I think you don't do
20:03the entrepreneur's job either.
20:04You need to be very realistic to where we stand.
20:07Otherwise, you cannot solve the difficult questions.
20:09But, obviously, you need to also be very optimistic of where we can go.
20:13And sometimes it's easy to think, like, oh, things are going so fast that it's too late.
20:21But it's not because of the speed of the technology.
20:24It is unprecedented.
20:25And it's normal we're asking ourselves those questions.
20:28I always give this example.
20:29But in France, it took 20 years for half the citizens to have a personal computer at home.
20:33It took two years for everyone to have AI in their pocket and use it regularly.
20:38Like, the speed of the transformation is such that obviously creates, you know, so much tension in the system,
20:44because humanity hasn't been, you know, trained to go through this speed.
20:48But the speed also means there's so much still to build.
20:53And it's only the beginning.
20:55I know it probably sometimes doesn't feel like it.
20:57But when I'm seeing teams like Yann Lequin's team building world models out of the global world,
21:03but out of Paris as a headquarter, we know there are a lot this technology still can bring.
21:08And there's still a lot to figure out.
21:10And I'm sure you would agree to that.
21:12So yes, sovereignty is achievable.
21:15But it requires a hand lot of commitment, financial commitment, purchasing commitment,
21:22because it's great to talk about sovereignty all we want.
21:25But if we continue buying non-sovereign solutions, things are not going to change.
21:28So I really want to highlight the strategy that Skyway has taken with Visora,
21:32because this is almost a political move.
21:36You commit to say, hey, I want my technology to be European at some point,
21:40so I'm going to go and purchase those technologies even though they're not even ready.
21:44But if we all do that, think about the mountain we can move around.
21:47I mean, we have the money.
21:49It's just not spent here.
21:51And it's also not invested properly.
21:53But that's a whole other topic we can discuss at some point.
21:56At another panel, where do you stand?
22:00I mean, sovereignty of AI is at two levels.
22:02One is on the training itself.
22:04The other side is on the inference, the running, the production.
22:08So at Proton, we bet on the open source models.
22:12So we run open source model on our infrastructure.
22:14And we make sure that this infrastructure is only in Europe.
22:17So we leverage open source models, so there's no lock-in to any provider.
22:23One and two, we run in Europe, so we have no dependency on US infrastructure for that.
22:29What are you thinking?
22:30Of course.
22:31Of course.
22:32Yes.
22:33That's a big yes, as you can imagine.
22:36Yes.
22:37And we need that.
22:38Yeah.
22:39And we need that.
22:40Commenting on the opportunity of being a little bit more sovereign.
22:43When it comes to cloud, the European market, this is 80 billions.
22:48Most of it goes like 85% of it.
22:50Let's say 100% of it goes to the US hyperscalers.
22:54You know what's the difference between a US hyperscaler and an hyperscaler like Scaleway?
23:0050 cents out of a euro.
23:03When you spend one euro with Scaleway, this is 68 cents that stays into the European economy.
23:11When you spend one euro to a US hyperscalers, it's less than 20.
23:1750 cents out of a euro, 80 billions.
23:21Well, we could actually get 40 billions into European economy every single year.
23:31Easily achievable.
23:32What can we do with that?
23:34Can we get prepared for the AI revolution?
23:37Can we get prepared for quantum revolution?
23:39Of course.
23:40Of course, yes.
23:42And the good news is it is happening at the moment.
23:47I mean, we've seen the European Commission, European Commission selecting a cloud.
23:50They selected a European cloud.
23:53It may sound trivial, but it was not the case two years ago.
23:57That's a big difference.
23:59We've seen the health data hub that used to be on Microsoft moving to Scaleway.
24:04Finally, it took some time, but finally, it is happening.
24:09And we're seeing more and more private companies.
24:11Yesterday, we announced LVMH.
24:13We announced La Maif, which is an insurance company in France.
24:17We announced RATP, moving all of them to Scaleway.
24:21And CHAP's vision, because cybersecurity is probably quite important as well.
24:25They are also moving to Scaleway.
24:27Can you imagine that?
24:29This is really happening.
24:30So, of course, yes.
24:32I have a follow-up question for you to challenge you a little bit.
24:36I do find it fascinating.
24:39So, I wanted to hear more from the perspective of, okay, these are the ingredients to become
24:44fully autonomous, fully sovereign, because this is the word of the day.
24:50Which of these boxes have you checked?
24:53For example, if you keep operations, shareholders, staff inside the European jurisdictions, you've
25:00got the governance box checked, the finance box checked.
25:05What about chips?
25:06What about the hardware?
25:07I know you mentioned some of those things.
25:10Do you feel like you can check all those boxes fast enough?
25:14I'm absolutely convinced we can.
25:17And this is what, again, this is what we're doing.
25:19We're committed.
25:20We're committed to Bureau, for example, to build the servers.
25:24We are committed to Visora, to build the GPUs.
25:27We are committed to Cyperl, to build the CPUs.
25:31We have a lot of great companies in Europe.
25:35We just need to think a little bit more European.
25:40We can do that.
25:41We can do that.
25:42We have great data centers in Europe, and not only in Europe.
25:46We have great European data centers in Europe.
25:49That's great.
25:50And you know what?
25:51We are uniquely positioned to be very successful in AI.
25:56And the reason for that is very simple.
26:00Electricity.
26:01We probably have the best electricity in the world.
26:05We have electricity in quantity.
26:07We have electricity in quality.
26:09So there is no best place to deploy AI than Europe,
26:14and maybe especially France, maybe Sweden,
26:17and a few of those European countries.
26:21That's a great opportunity, and we should care about that.
26:25Consider that AI running in Texas or AI running in Arizona
26:32is not as green as AI running in Paris or Stockholm.
26:37Think about it.
26:39So it's also part of our responsibility to actually make sure this happens,
26:43and the way we run AI, the way we use AI,
26:46is actually way more European and way more green because it's European.
26:54If possible, I would like to build on what Damien just said.
26:57Because, I mean, what do you need to try and achieve what we're trained to achieve?
27:02First, you need talents.
27:04There is no better place in Europe to find these talents.
27:08Actually, you find French and European people in every single lab in the entire world
27:13because they've been trained here, and we have some of those best quality training
27:18when it comes to AI and math in particular.
27:21Two, again, you mentioned it, but I want to repeat that one more time.
27:25Everyone is looking for energy.
27:27Like, there's not a single of the hyperscaler who is not looking to build more data centers,
27:32and to build data centers, they need energy.
27:34In the US now, you have states that are, like, just banning the construction of new data centers
27:39because they're struggling with the energy consumption.
27:42We exported in France the equivalent of the entire Belgium consumption last year,
27:4792 terawater.
27:48The energy is here.
27:49It's abundant.
27:50It's predictable.
27:51It's green.
27:52So, I mean, this is an amazing edge, and an edge that is quite difficult to compete with
27:57because we invested in nuclear for the last 25 years, or more than 50-something years.
28:02So, and three, you need dynamic ecosystem, agile ecosystem,
28:07who's going to be able to play catch-up to build those amazing things we need.
28:10And I think Vitec is probably the place where you can doubt that this ecosystem is here.
28:15I mean, everywhere you go on those three floors, you see tremendous entrepreneurs who are so eager to build.
28:22But there is maybe one thing that I think we all, like, suffer a little bit from.
28:27It's a confidence in ourselves.
28:30Do you think great builders go around in the U.S. or in China wondering if they are great enough?
28:36I don't think anyone does that.
28:38Like, if we want to be sovereign Europeans, the very first thing we need to do is believe in ourselves.
28:44And that's probably where we need a bit more of a mindset shift.
28:48Like, we have all it takes.
28:50Just, you know, let's believe we can do it.
28:53And let's do it.
28:54I was, when we organized the AI Action Summit a year and something ago,
28:59everyone was talking about AI.
29:00That was great.
29:01That was one of my KPIs.
29:02Let's get people to talk about AI and measure how much of a positive sentiment we can bring around the
29:07technology.
29:07Because this technology, I mean, creates fear and it's normal.
29:11And so it's also the role of the state to show the positive out of it.
29:14But little did I imagine that by trying to achieve this goal, one of the side effects we created was
29:21everyone on TV, companies, everywhere, press, radio, talked about ChatGPT, to talk about AI.
29:29That was crazy.
29:30Like, there was a French call.
29:32I'm not going to name them because since then they've changed who they work with.
29:35But they did a press release to announce that we're working with ChatGPT.
29:42Would you imagine Harvard Business School doing a press release saying,
29:45oh, we're so proud we're working with Mistral?
29:47I don't think they would do that.
29:49Would you imagine, like, the daily newspaper in the US saying,
29:52oh, like, to talk about AI, oh, Mistral, Mistral, Mistral, Mistral?
29:56They would never do that.
29:57So we need to be a bit more conscious of who we are, the great talents we have,
30:02and have conviction that we can do it.
30:04And that goes with being a little bit broader.
30:07Let's get a bit of this US marketing that we sometimes miss a little bit.
30:11Because I think that's also how we can be sovereign, you know.
30:14Very interesting.
30:16You know, both Antoine and you, Rafael, you've talked about sort of this open source model.
30:23So I wanted to double click a little bit on that.
30:24So maybe I'll start with you, Rafael.
30:28You've argued before that Europe has to stop treating its tech stack as a coast and start treating it as
30:35an investment.
30:36And you mentioned in a conversation the China automotive bet.
30:41I would love for you to talk a little bit about that because it is a path to what Clara
30:46and Damien has been so passionate about talking about today.
30:50Yeah.
30:51And before answering the question, I would agree with Clara on the fact that, yes, there needs to be a
30:55mindset shift in Europe.
30:56I mean, we need to, we as Europeans, believe in Europe technologies.
31:01I mean, China did China first.
31:03US did China, US did US first.
31:06We need Europe to do Europe first.
31:08And we need to believe in ourselves for that.
31:11To look at China 20 years ago, the vehicle industry was not great in terms of electric vehicles.
31:19They realized that they were depending too much on oil.
31:23And so they decided to look at it as an investment.
31:27So we are going to invest in our local industry of electric vehicle.
31:30And now if you look 20 years later, close to 20 years later, they have a very robust electric vehicle
31:36industry.
31:37And they have reduced their reliability on their, their reliance on, on, on oil and on foreign producers.
31:44So we need to do the same in Europe.
31:46That means like Scaleway is doing, like other providers are doing, investing into our own players.
31:51So that we, we invest and we will reap the benefits both short term and long term.
31:57And, you know, you, you talked about, I don't know who mentioned it earlier that we don't want to be,
32:03we no longer want to be sort of colonized.
32:05It feels like being colonized.
32:07And I was thinking, okay, what's the other word?
32:09Well, Europeans were pioneers actually.
32:11And so I was thinking, can this model, this open source model you were talking about, are there limits to
32:18it?
32:19What are, what are some of the limitations?
32:20What could be some of the unintended consequences?
32:25Yeah, that's a, that's a really good question.
32:27I think, you know, I think that classical open source is a really good foundation for how we can create
32:34the cooperative environments that are needed for AI.
32:36But AI is also a very different model, no pun intended, than what we normally did with classical open source.
32:44When we think about classical open source, we might think about something like Linux, for example.
32:49You know, collaboratively developed, you know, everybody contributes to it.
32:53You can clearly see what changes are made.
32:55And, you know, people can agree on essentially what goes into the kernel that everybody gets downstream.
33:01And then on top of that, great application, and a great application layer has formed on top of it to
33:05actually build products and to commoditize the underlying open source technology.
33:10With AI, it's kind of a bit different in that setting up that same collaborative environment where everybody can contribute
33:18to an AI development is much tougher to do.
33:21Just because of the centralization of resources that has to take place.
33:26You know, the scale at which you have to end up training.
33:29The decisions that you make around data sets that go into it and that end up aligning the system towards
33:34particular behaviors or another.
33:36And I think what we need to do is find the way by which we can create these same types
33:40of historical open source ecosystems, but specifically for AI development.
33:44And that's one of the things that we're working on quite a bit with Apertus, is how can we bring
33:48in these voices from across the world, across different parts of Europe, across different institutions,
33:52and actually give them a say as to how the AI is developed, such that it can serve as a
33:57powerful foundation for what they want to do after.
33:59And then, they themselves can build the same application layers on top of it that they need, either for different
34:04products, for different applications, for their own national or local sovereignty if they need it.
34:09But I think we really need to find a way to democratize the costs that actually go into that base
34:15system that is there.
34:16Because it is supremely expensive.
34:18These days, a frontier model can cost billions.
34:21In a few years, it will be tens of billions to develop from scratch, even not taking into account the
34:25infrastructure that's necessary to put it there.
34:28That's not, you know, what a classical open source project normally looks like.
34:32And being able to put in those resources requires a very large cooperation to do in the first place.
34:37And when you say cooperation, you mean cooperation across all of the European nations?
34:42We're still being sort of talking about Europe only.
34:46So, you know, I think it's a good starting point in that there's already a structure in place for European
34:51nations to work together in the EU and the Council of Europe.
34:54But I view it more broadly than that.
34:56You know, I'm a personal believer, and I think most people here are probably as well, that there's a lot
35:00more that unites us than divides us as people.
35:02And I don't think there's necessarily a big reason why we couldn't have entities in, let's say, Japan or South
35:07Korea or India or Indonesia or Thailand.
35:10You know, I work with people from all of these countries, and we discuss the Apertus project, and we discuss
35:15the efforts that they're putting into open source AI development in their countries as well.
35:19There's no reason why we can't all have kind of a central place by which we try to work on
35:23this incredibly expensive proposition together.
35:26But I think it will have a much higher chance to succeed if we're actually able to bring in those
35:31voices into the room together.
35:32I was looking to you to be sort of like my United Nations of AI right now because I felt
35:37that we needed to have a bit.
35:38Well, Rafael is the one that lives in Geneva.
35:40I live a couple of kilometers away, so.
35:43And I guess the question that I have for all of you is when I think about sort of the
35:49hyperscalers and how much, even in this European conference right here, how much of it, how much of them have
35:55a presence here?
35:56A presence, you know, so a lot of the discussion is aspirational, is inspiring and interesting.
36:02But in reality, is this conversation right now about cutting out the US hyperscalers entirely?
36:11Or is it more of a balanced approach, balanced purchasing approach, so that, you know, there's maybe a bit of
36:19a line there, like a little bit of both, at least in the beginning, because I'm not sure that you
36:25can go back on Monday to work and say, we're done.
36:28We're not only doing this.
36:30Or maybe it is.
36:31I don't know, Damien, you're laughing, so I don't know what's going to come from you.
36:34I think choice is very important.
36:38Choice is very important.
36:41So it's not a question of having only European.
36:45It's a question of having a European profit 55% on US hyperscalers.
36:52The problem is not it relies partially on US hyperscalers.
36:57The problem is 85.
37:00This is too much.
37:02If it was like relying 40% on US hyperscalers would be great.
37:07I mean, competition is key to make sure that we stay at the edge, technologically speaking.
37:16This is key.
37:17So we need to have competition, including with US hyperscalers.
37:21We need three or four US hyperscalers in Europe.
37:25But we also need to compete against Chinese hyperscalers, US hyperscalers.
37:30But according to me, the problem is a problem of balance.
37:33It should be 40% Europe, 40% maybe US, 20% Chinese.
37:39That would be much more balanced.
37:41And it would mean a much better impact on our economy.
37:49So, no, I don't think, I don't think it, I mean, those, the way, maybe you remember the way China
37:58reacted to the question of GPUs.
38:01What did they do?
38:02What did they do?
38:02China.
38:03When they realized they were too much dependent on the US, and the US was starting to cut, or at
38:10least to limit the exportation of GPUs.
38:16They blocked importation of the GPUs.
38:19So the US said, you're limited in the number of GPUs that you will receive.
38:24The answer to China was, great.
38:26I will stop importing any GPU.
38:29The consequence is, within three years, all of the Chinese companies have been able to develop GPUs.
38:38This is a bit extreme.
38:40Of course, if Europe would rule that no European company is allowed to actually use a US hyperscaler, it would
38:47end up being, we're going to use the European hyperscalers.
38:52I think it's a bit too brutal for Europe.
38:58But introducing some kind of European preference in order to balance the market repartition, this is what we're looking for.
39:09That's fully aligned.
39:10I mean, we need to get to a position of, like, ultra dominance, 85% on the cloud, 75%
39:16on email, to a situation in a few years where it's 40-40, 50-50.
39:21I mean, it's like, it's balanced, and European companies, CIOs, IT departments have a real choice, alternative, where they can
39:29feel safe to say, okay, now I can choose a US or Europe, and both are the best solutions, and
39:36okay.
39:36Today, there needs to be pushed, like, European preference, but it's not, we need to, as providers, we cannot shield
39:44ourselves behind this European preference.
39:46We need to be the best, and Proton, without VC funding, we went from zero customers to 100 million users
39:53around the world, personal users, and 100,000 companies using our products without VC funding.
39:59So it's possible, Scaleway grew, we grew, now we need to go in the next few years to the next
40:03stage where we are, like, European hyperscalers.
40:06Yeah, and we're in our sort of last few minutes, so I wanted to just maybe check or measure the
40:13pulse one last time with all four of you.
40:17It's VivaTech's 10-year anniversary, now we see months, you know, decades happening in months with the development acceleration of
40:24AI.
40:25Let's fast forward 10 years, I'm not looking for you to forecast or foretell the future, we're building it today,
40:31that's what VivaTech's motto is.
40:33The question I have for all four of you, do you believe this notion of digital sovereignty, tech sovereignty that
40:40we talked about today will have changed anyway, in any shape or form?
40:46Our notion today is a bit nationalistic, and that makes sense, this is what we know and understand.
40:51We're driven by policies and political issues.
40:55What will it look like in 10 years?
40:57Do you think something will change?
40:59Will we look at a new ism, a new type of competition?
41:02We can go with you, Claire, and we'll just go down.
41:04Yeah, I think to answer this question, I'll go back to something Damien has said, which is, it's good for
41:09innovation.
41:11Because when you have monopolistic companies, it just kills innovation, monopolistic companies are not good to create the best technologies
41:19and innovation.
41:20And I think quite a lot of economists agreed with that, and we've seen some of the rules related to
41:25that in the US actually, pretty strong over the course of the industrial revolution.
41:31So I guess if we get going, the conversation is not going to be about sovereignty a bit more like
41:38it feels today as a defensive concept, but it's sovereignty for us, for Europeans, for innovation, for the best of
41:44the technology.
41:45I think that's probably the switch we are going to see happening, luckily.
41:49And I just want to add one thing, because it feels like in this conversation is a bit Europe against
41:54the rest of the world.
41:55I think one thing we should not forget, and there was an AI launch during the G7 presidency meeting in
42:01Evian just two days ago with head of states from G7 and plus India, Kenya, et cetera, and CEOs of
42:09the companies.
42:10And the idea that was discussed on that table also holds very strongly, which is, this is too important to
42:16think even by region.
42:18There needs to be, I go back to the UN idea, but actually the UN is hosting the very first
42:22AI dialogue in Geneva in just a few weeks from now.
42:25There needs to be global coordination.
42:27It's not because we're working on building up sovereignty and capability.
42:32This is not against like collaboration internationally.
42:36It's for, again, the collaboration, because if we have more capacity for ourselves, then we have a stronger voice when
42:41it comes to how we want to operate with partners internationally.
42:44And this cooperation is fundamental.
42:47We need to share models.
42:48We need to share with open source, actually.
42:51We need to share standards.
42:52We need to share capacity building.
42:54And this is not just like countries or regions against regions.
42:58It's all together.
42:59If we want this technology to build the best for humanity.
43:02Yeah.
43:03It's about options.
43:07Yeah.
43:07I think the one thing time shows us is that our definitions today will be completely inadequate 10 years from
43:15now.
43:15We'll probably have an incredibly different idea of what digital sovereignty means about specifically what AI sovereignty and sovereign AI
43:22mean as well.
43:24I think it'll be an interesting story to see how that process develops, but it'll look completely different from today.
43:30Yeah.
43:32Yeah.
43:32I would say, I mean, we have, it's very binary.
43:34I mean, either in 10 years, we'll have managed to like a grow cloud market share and like to the
43:3940% that Damien was saying.
43:40We'll have grown like email providers to like 40, 50% market share.
43:45So we'll have reduced the dependency.
43:47Then we will look at it as a success story or we will have just been too slow and we
43:53will be doing a post-mortem.
43:55I believe in the first one.
43:56Uh, and I, I look forward to, to this time where we can say maybe in 10 years, in five
44:00years where like sovereignty is dead, long live sovereignty.
44:04Very good.
44:06Would you like to have a last word?
44:07Very quickly.
44:09Five years ago, running a cloud was only about running CPUs.
44:15Today, in terms of power, we're running more GPUs than CPUs.
44:20In five years time, 10 years time, I'm convinced we will be running more quantum than anything else.
44:28And you know the good news?
44:30There are more quantum startups in Europe than anywhere else on the world.
44:35We are first on quantum.
44:37So in 10 years, it will be completely different because the tech would be revolutionized by quantum and Europe would
44:45be leading this revolution.
44:49Very interesting.
44:50Thank you so much to my panel and thank you to you for watching.
44:53Thank you very well evening.
44:53Thank you, should God bless you.
44:53Thank you very much.
44:54Why not?
44:54Thank you very much.
Comments