- 2 weeks ago
Ecosystems and deep partner integrations are central to securing AI sovereignty at scale, complementing the regulatory foundations Europe has already established. In this session, Talal draws on Core42's experience constructing one of the world's most deeply integrated AI ecosystems, spanning government, private sector and academia, to explore what it actually takes to embed sovereignty into the fabric of an AI-enabled economy. Joined by Core42's partners NVIDIA and Microsoft, the conversation examines how tightly connected ecosystems, where infrastructure, models and institutions move together, create the kind of strategic independence that fragmented, dependency-heavy approaches cannot. The experience points to a wider theme: that sovereignty tends to follow from the partnerships and ecosystem an economy builds, and how effectively those partners work together.
Category
🤖
TechTranscript
00:00So please prepare to welcome Professor Olivier Hullier to the stage.
00:05Thank you very much.
00:35Access to Fable 5 and Mito's has been cut following an order from the White House.
00:41It was at the same time a reality check and a wake-up call.
00:46When it comes to AI systems and the AI industry, most of the people who are part of the entire
00:55AI world are tenants, not owners.
01:02Following this announcement, there were lots of reactions all over the world.
01:07How can we improve sovereignty and not depend on a few stakeholders?
01:12Yesterday at VivaTech, the French Prime Minister announced an investment of 655 million euros for sovereignty.
01:23It looks substantial, but it covers research, compute, industry.
01:30And when you think that one gigawatt that is being used for powering AI costs, on average, $50 billion,
01:41suddenly the 655 million figure no longer looks so substantial.
01:50The problem when it comes to sovereignty is that no single stakeholder, no single country, no single company can do
02:00it all.
02:01And they have to face bottlenecks.
02:03The oldest bottleneck when it comes to AI systems is a bottleneck that is related to knowledge circulation.
02:11Distilling models, is it possible?
02:14Are we facing models that are locked, solutions that are locked?
02:18And we have to deal with how the know-how would move.
02:23The second bottleneck is infrastructure.
02:27There is a massive gap between the U.S., China, and the rest of the world.
02:32Yet, one could argue that this bottleneck could be closed with capital.
02:40The third one, this third bottleneck that is important, is adoption.
02:48You can have the best technology ever.
02:50If no one is using it or no one has access, it's basically useless.
02:55And the fourth one is intelligence itself.
02:59Think about it.
03:00Think about it.
03:01Today there are still humans in the loop, but we're contemplating recursive AI systems that are training themselves.
03:08And when it comes to intelligence and access to intelligence,
03:13some people are currently riding a horse
03:16while others are driving a Formula One.
03:18Map those bottlenecks to time.
03:22And knowledge circulation is the past.
03:25Infrastructure is today, and recursive AI is tomorrow.
03:31The European Commission is well aware of the issue of sovereignty.
03:37And if you look at the most recent packages, it has integrated semiconductors.
03:42It has integrated compute, cloud, open source in one common agenda.
03:49But is that sufficient?
03:51What is important is to build ecosystems.
03:54Multi-layer systems, because the system will outlast the technology.
03:59We need those systems to be efficient, to be resilient, and to be future-proof for the next wave,
04:05the next technological and scientific disruption, be it quantum or otherwise.
04:09Systems will outlast the technology.
04:13So, today we're here to talk about how partnerships can be built.
04:17How sovereignty is not just a matter of trying the impossible,
04:22meaning building, developing, owning everything in your own company,
04:27in your own country, in your own continent, or supranational organization.
04:32And to speak to the systemic layer, the partnership.
04:38Let me welcome Talal Al-Kaisi, CEO of Core 42,
04:43and Group Chief Global Affairs Officer at G40 Tour.
04:49Talal.
04:56Welcome, Talal.
05:00It is my pleasure also to welcome Nat Ives, Director of Enterprise at NVIDIA,
05:07to speak to us about the infrastructure, compute, and chips layer.
05:17And our last finalist is Philippe Limentour, who is CTO and CSO of Microsoft France.
05:25For the last layer that we're going to address today,
05:28the enterprise layer that reaches people, the end users, and the companies.
05:33Philippe, welcome.
05:39Thank you very much for being here, Talal.
05:41Let's start with the fact that when we're talking about sovereignty in Europe,
05:45we see a lot of strategies that look like,
05:48let's depend less and less on non-European vendors.
05:53And I have moved about a year ago to the UAE,
05:56and what really struck me is the fact that the UAE has been able to build a sovereign system
06:01when it comes to AI, in terms of infrastructures and all these layers,
06:05but not with 100% UAE-based created solutions,
06:09but thanks to key partnerships, international partnerships with multi-stakeholders.
06:14Now, can you share with us what the journey was and what the strategy should be,
06:20in your opinion, when it comes to Europe's sovereignty or any supranational body for that matter?
06:26Great. Thank you, Olivier.
06:27And that was a great introduction, by the way.
06:30I think you touched on all the high points.
06:31But maybe what I can say is that I think Europe is definitely asking the right question,
06:37but often in the wrong way.
06:39The issue isn't really whether Europe can build AI alone.
06:44It's more or less whether Europe can retain strategic control
06:49while still participating in global AI ecosystems.
06:53So in our experience, sovereignty is really not isolation.
06:57You need partnerships.
06:59You need control and operation.
07:01You need control over jurisdiction, access, governance, compliance,
07:06and the ability to have it become auditable.
07:09And that's really what separates dependency from real strategic confidence.
07:16So what we've learned in our experience that you described in the UAE, Olivier,
07:20is that partnerships don't weaken the sovereignty operating model
07:25if the model is designed correctly from the start.
07:29And a good example is the sovereign public cloud.
07:32I mean, we work with Microsoft and leverage technologies like in processing or in memory encryption,
07:39along with encryption in transit and encryption at rest,
07:43to have full data lifecycle encryption and enable public cloud consumption for the public sector
07:48and regulated industries for the majority of workloads that the government has.
07:52But on that same side of the coin, you do have secret and top secret data
07:56that require a level of autarchy and isolation that only an air gap or a disconnected cloud can have.
08:01So there needs to be a ratio.
08:03There needs to be a hybrid implementation and there needs to be leverage of partnerships like NVIDIA and Microsoft
08:08where you can have the best-in-class technologies and environments that are suitable for the types of workloads that
08:14you deploy in.
08:15I think a good example of that is the 5 gigawatt AI campus in the UAE, right?
08:19Very much so.
08:20And something you mentioned earlier, I think, was extremely important maybe to double-click on.
08:26When you look at the cost of what it takes to build an AI data center,
08:29and we're talking about data centers that are pulling, I mean, NVIDIA right now with the GB300 is pulling, what,
08:35140 or more kilowatts per rack.
08:36This is opposed to what traditionally happened with 25 kilowatt racks for traditional cloud infrastructure.
08:42So when you think of the energy that's required, the shell and core, the power generation, the liquid cooling, the
08:49networking infrastructure,
08:50that capex that you mentioned of $50 billion a gigawatt is something that you can only multiply by hundreds of
08:58gigawatts
08:58that are required for the demand profile that we are working towards.
09:02And with Jevons Paradox, that macroeconomic theory that the more supply you put, the more demand you're creating,
09:07so it's an insatiable demand for AI compute, there's going to need to be hundreds of gigawatts that are built
09:13globally.
09:14So that's massive in terms of investment and very prohibitive for certain countries.
09:18So when a country like the UAE dedicates 5 gigawatts of energy, clean energy, to manufacture intelligence,
09:25we're really starting to go from a country that exported hydrocarbons and barrels of oil to exporting intelligence in tokens.
09:32Because the ability to serve a 3,200 kilometer radius, about half the world's population, sub 60 milliseconds,
09:41then presents an opportunity for countries that might not be able to have the economic wherewithal
09:46or the political wherewithal to be able to build these data centers in those countries.
09:50So that 5 gigawatt cluster, apart from the domestic consumption,
09:53is going to be very much something for exportable tokens and intelligence from the UAE.
10:01Nat, one thing that people found out last year is a big announcement of a partnership between Mistral and NVIDIA
10:09for a sovereign solution.
10:11And uninformed people of the importance of partnerships and the fact that not everything needs to be located on locally,
10:21were a bit surprised, saying, wait, Mistral, French, European, NVIDIA, American, how is that sovereign?
10:27Can you explain to us the process and building on this, how the partnership between NVIDIA and Core 42 has
10:35helped many organizations
10:37and contributed to the sovereignty of AI in Europe?
10:42Yeah, so great to be back at VivaTech.
10:46That announcement last year was kind of a milestone for France and the ecosystem.
10:50I think when we talk about AI, as your introduction, Olivier alluded to, it's not just a simple software thing.
10:59Actually, the complexity behind making modern AI and frontier AI possible is huge
11:05and requires a big, heavy lift by players all across the ecosystem.
11:10Jensen likes to talk about the five-layer cake.
11:12It starts with availability of electricity.
11:14Then we move into the chips, which is obviously where NVIDIA is well-known.
11:18The infrastructure, so the data centers, the servers, the electrical distribution, the cooling.
11:23Then into the models.
11:24And finally, into the software applications that business and government are going to use to deliver value.
11:30And this layered cake involves many players in many industries all around the world working together
11:38in what can be described as kind of an ecosystem marvel that we're actually delivering this complexity at the speed
11:45that we are.
11:47So I think if you look at previous industrial revolutions, they all drove in considerable amount of infrastructure build-out.
11:57So, yes, the AI revolution is a software revolution, but behind it, we're building out this infrastructure at incredible pace
12:04at a scale and speed and level of investment that no one has ever seen before.
12:08And that's exciting.
12:10But no one can achieve that on their own within one company, within one geography.
12:17And if you look at even at very successful French companies that are historic, you know,
12:21if we look at Airbus or Michelin or L'Oreal or even French nuclear power,
12:27these supply chains behind these very big and successful industries are incredibly global and incredibly complex.
12:37And Europe, France, and the neighbors have been navigating these global supply chains for centuries.
12:45So we're very good at doing it.
12:47So when we think about building sovereign AI, it's not about does the supply chain have to be in one
12:54geography or one postcode.
12:56It's about do I have control?
12:58Do I have choice?
12:59And in AI, you know, there's not one flavor of AI.
13:02One size will not fit all depending on where you want to deploy it, what you want to use it
13:07for, what the data is.
13:09It's a very complex situation to mix the right building bricks together to make the right solution.
13:16So you need to have control over that.
13:19Part of that is definitely open source.
13:21NVIDIA is a big contributor to open source AI, but as is Mistral.
13:24So in Mistral, not only did we see a European champion that we wanted to support and enable,
13:29we saw a kindred spirit in terms of open source and choice, and hence we made that partnership and that
13:36announcement.
13:37Well, it's very interesting to hear the key player in infrastructure, NVIDIA,
13:43remind us how important it is to learn from legacy industry in the AI era,
13:49that supply chain is rarely 100% national or local, and we need to learn from this model as well.
13:58Speaking of which, Philippe, so Microsoft has answered some of the concerns of the European Commission
14:05by developing solutions that are European-based.
14:08So the servers are in Europe, data is stored in Europe, computers in Europe,
14:12yet you remain an American company that has to answer to American laws.
14:18Can you share with us how Microsoft is ensuring, contributing to sovereign European AI this way,
14:26despite one foot in Europe and one foot in the U.S. when it comes to regulation?
14:34Sure, nice to be here.
14:36First, let's go back to the root of what we are talking about.
14:40It's about demand.
14:42So first, we have to acknowledge that there is a huge demand for AI, for new systems, for cyber security.
14:50Companies of all sizes, all industries everywhere in the world want to transform themselves to win on the markets,
14:58win other markets also.
15:00So it's not only about being local, but being an international company,
15:06except, you know, public sector companies specific.
15:11So this is a complex world.
15:14If you want to follow the pace of NVIDIA's roadmap and lifecycle of products,
15:20if you want to follow the pace of updating, patching against cyber security threats,
15:27you have to have an infrastructure which is highly, not only scalable, but an adaptive one.
15:34And the hyperscaler world provides that possibility.
15:41Then you add the choice because people don't want to be dependent on a single component in the supply chain
15:49lifecycle.
15:51So we are a platform.
15:54Microsoft is, from the past 50 years, have been building a platform with an ecosystem of partners.
16:00That's what we did in the UAE with CoreCut 42.
16:03That's what we are doing every day with NVIDIA everywhere in the world, with Mistral in France and Europe.
16:09So we work with partners.
16:12But the first thing is, yes, we are, Microsoft France is a French company with a U.S. headquarter.
16:22First to remind that almost all the companies are subject to data requests.
16:28There's a new e-evidence in Europe, which is the equivalent of the cloud out, which holds also data access
16:35in Europe.
16:36So if you do some business in another country, and you bet you should, because if you want to be
16:44a big player on the market,
16:46you can't just play locally.
16:48So that's the first thing.
16:49The other side of the complexity is regulatory environment.
16:54And every three to five days, every three to five days, there is either a new regulation or an update
17:02on the regulation somewhere in the world.
17:04So how do you keep pace of keeping your business compliant?
17:10Because if you're not, you don't have the license to operate in a country.
17:15And each country has its own set of legal constraints.
17:20Even in Europe, you know, each country, there is a European level and then the French level.
17:25So we need partners to adapt to ensure that the customers are delivered with a platform of choice.
17:33We have 11,000 models available in Microsoft Foundry, for example.
17:38But at the same time, we scan them automatically.
17:41We ensure that they are secure.
17:43We ensure that no one has broken the weight inside it.
17:47That's the same for the world's securization of the assets.
17:52That's the same of ensuring that everything is up to date, that there is no supply chain attack.
17:57Excellent.
17:58So that's really where the partnership is built.
18:01So we have our own data centers.
18:04We upgraded the infrastructure in Europe by 85% in the last two years.
18:12We have a continuum of platform.
18:14So we are working locally, like in UAE, the same in Blue in France or Delos Cloud in Germany.
18:20Let me build on this, Philippe, because that really is a great segue to you, Talala.
18:28And the fact that Core 42 has a partnership with Microsoft amongst other companies, big partnership.
18:34But also, you have deployed, you have headquarters in Europe, you have deployed in France, in Italy, and beyond.
18:41And you, too, have to face those constraints, meaning that you're a UAE company that has deployed infrastructure in Europe.
18:52How is Core 42 dealing with this complexity?
18:56By maximizing optionality.
18:58I think a key component of what we've done in the UAE successfully in enabling a digitally sovereign nation by
19:09leveraging partnerships and figuring out the balancing of what needs to go in an air gap versus what needs to
19:14leverage a public cloud with the right policy set, control set, technical solutions, is one component.
19:20But there is also operational sovereignty.
19:22There are multiple layers of sovereignty.
19:24It's not just data sovereignty that's important.
19:27Different types of workloads require different levels of isolation, different levels of operational sovereignty.
19:32So in the case of Europe, we know the security imperatives that the European Commission has are such that there
19:40needs to be a level of localization involved as well.
19:43So it's not only about us bringing GPUs and putting up an environment in the continent and then saying, we'll
19:51do everything for you.
19:52No, we look at local partners that can help with the different elements of sovereignty for every particular use case.
19:58So these are the types of things that I think, in maximizing optionality, we're able to bring to the table
20:03to emulate the success story we've had in the UAE in enabling these types of environments.
20:09And to expand it to other territories and beyond UAE, we saw recently some big signing between President Macron and
20:20Khaldun.
20:21Yes.
20:23Things are progressing, and the fact that there are these collaborations, and we're looking at what is happening at government.
20:29I would like to go back to you, Nat.
20:32Your perimeter covers France, Benelux, Nordics.
20:35You have a wealth of experience and data points when it comes to adoptions and some of the hurdles.
20:44Which layers are being covered at the moment and which layers are not being invested in or considered in the
20:55governments that you've been working with?
20:56So I think, you know, in general, all layers are being considered and invested in and are being covered.
21:03But obviously, different geographies, different industries as well find themselves in different situations.
21:09France has the fortune and great choices historically to have basically all the layers covered, actually.
21:17If we look at France with power, with the ability to build our infrastructure, you know, great capability locally around
21:25data infrastructure, frontier labs like Mistral, H Company, and others based in Paris, AMI.
21:34And actually, I'm pleased to say, like, the French public, there was a survey that came out about a week
21:40ago.
21:41So 60% of French people are now using AI regularly, and I think that's, like, the fifth country in
21:48the world in terms of AI adoption.
21:50And over half of French people are using AI at least weekly.
21:54So if I look at the other geographies, the Nordics, you know, they have most of those bases covered.
22:02I don't really like comparing between geographies because a lot of the constraints, like for Benelux, in terms of space
22:08and electricity, are historic.
22:10But, you know, they're all working together, and we bring the ecosystem together.
22:15As I already said, these data centers, these projects are extremely complex, and it's only through extreme co-design that
22:24we can actually achieve the level of density and capability that's required for modern AI.
22:31And that's why working with partners like Core 42 and Microsoft is super important.
22:36Wow.
22:38So, Philippe, one of the things I like a lot about your title is your CSO and CTO.
22:45As an academic researcher and tech entrepreneur, I appreciate the fact that science and technology are together because you cannot
22:52consider one without the other.
22:54That was my public service announcement of the day.
22:57With your experience, when it comes to governments in Europe, the layers are one thing, but then adoption, speed, innovation,
23:08where do you think that there is the biggest gap to fill?
23:13Or if you had to push and hold the hand of some of the decision makers in this supra-national
23:20body that is the European Union, where do you think the priorities should go at the moment?
23:27Everything should go in thinking about how to innovate as fast as possible.
23:32That's the target, because we are talking about risk management, and innovating is taking risks, and then you have the
23:40counterpart to control those risks.
23:43I think, as Nat said, the French people are the most advanced in the world in adopting AI.
23:53AI, I think if I would have to focus on one thing, it would be training on AI or learning
24:01on AI at the executive committees and board level.
24:05I think...
24:05AI literacy?
24:07Yeah.
24:07I think that's where, because it has to go at some point within the strategy of the company, because being
24:14a frontier firm really means using AI not to incrementally move, but to reinvent the processes, reinvent the way you
24:21work, reinvent the way the organizational model is set.
24:25That has to come from the top of the company.
24:28Then the next layer would be, there are things to consider, like, okay, data access, as you mentioned, but shadow
24:38AI is a key one.
24:40So are you developing, deploying AI fast enough in your company to prevent your employees to use non-authorized tools,
24:51which threaten your data, your assets, your brand?
24:58And I'm seeing a lot of it, and not a lot of people being accountable for screening and controlling shadow
25:04AI.
25:04So that would be the second one, cybersecurity, top of mind risk, far, far ahead from any other risk.
25:14So innovation, taking risks, training, learning at the EXCO level, understanding, framing the right strategy to be a frontier firm,
25:23to win on your market, ensuring that you deploy and develop AI,
25:28because it's both a top-bottom strategy to be executed, but it's also a bottom-up one.
25:36Absolutely.
25:37Because people who know about the pains to solve are the people who are working on those assets.
25:42But it's so important.
25:43This is such an important point at the moment, because you've got most organizations have not achieved their digital transformation.
25:50They have been hit by the generative AI tsunami, and now there is the agentic AI revolution.
25:56No one is ready.
25:57There is no curriculum at the moment, except maybe at the Mohammed Ben Zahid University of Artificial Intelligence in Abu
26:03Dhabi,
26:04that is preparing authorities and stakeholders to work with agents.
26:10However, one thing that people need to realize is that agentic AI will help achieve the digital transition,
26:17will help achieve the generative AI adoption.
26:20So the tools are actually not contributing to more complexity, but could help solve.
26:26And this, as we enter the last part of our conversation, gentlemen, short answers from all of you.
26:35Nate, speed, infrastructure, et cetera.
26:40Philippe really accurately mentioned cybersecurity, which is a top, top, top priority at the moment,
26:47especially with everything we heard about on Tropic and Metos.
26:52What are the lessons that you've learned that you would put on top of the agenda?
26:57Philippe shared his.
26:58We'd love to hear yours when you're advising organization, public and private.
27:03So obviously when we talk about especially agentic AI that's touching tools and systems inside an enterprise,
27:10inside a government, et cetera, this becomes a very complex IT problem, security, data, access to systems,
27:18or blocking access to systems becomes very important.
27:21It's complex, but actually it's solvable.
27:26IT teams and all the partners involved can work on these problems and solve them.
27:33So it's a constraint, it takes time, it takes money, it's complex, but it's definitely addressable,
27:38and successful projects are addressing that.
27:41But I think actually the other side of the coin, which is more the cultural leadership challenge
27:48around adopting AI into business, into government, is the piece that's actually constraining companies
27:55and governments right now, having the curiosity to play with AI, to understand the possibilities
28:04and to imagine new possibilities also takes time.
28:08So leaders need to find that time.
28:11They're very busy people, but also they need to create the time for their people to play
28:15and explore the new capabilities of AI.
28:18It's moving so fast.
28:20It actually sort of doesn't matter if you haven't started, because if you start today,
28:26you're learning stuff that's only a month or two old anyway, so you're not behind.
28:32But the quicker you start to play with the technology and understand the possibilities,
28:36the quicker you can imagine how you can use it to power new revenues,
28:43make access to public services more efficient and progress.
28:48Talala, one thing that G42 has introduced is digital embassies and green shield.
28:55We have little time, but can you just summarize what these initiatives are about?
29:00I think it's a central piece to the topic of this panel.
29:04When you think of sovereignty in today's world as interconnected countries and companies
29:10and the collaborative nature that things have to happen in, rather than striving for technology independence,
29:18you need to look at strategic interdependence.
29:21And a digital embassy is a product of that.
29:23The conceptual understanding that I try to convey is that sovereignty isn't predicated on only having data residency anymore.
29:33In today's world, you have to be able to find like-minded partners in terms of companies and countries
29:40where you could then create digital resilience in a form that is hard to engineer an architect
29:47into a traditional cloud infrastructure if you're only looking at data residency as the biggest imperative.
29:53So when we came up with the concept of digital embassies,
29:56there is a G2G element in it that gives you the policy and legal air cover that's required
30:03to prevent something like a Cloud Act from getting in the way
30:07and allow a country to host its data in another country under its own data imperatives and security laws.
30:15And vice versa.
30:16This could be a reciprocal ordeal.
30:18So we're very happy that both France, India, and a couple of other countries
30:24that we have G2G agreements with are signed up to explore this further
30:30and begin to implement some of those solutions strategically.
30:34So we learned today that, and we were reminding the importance of sovereignty,
30:39but sovereignty as a multi-layer, multi-stakeholder system that can be informed
30:44that what happened with legacy industries, the fact that it's all about partnerships,
30:50having the different stakeholders complement each other
30:53and working hand-in-hand.
30:55Please join me in thanking Talal Al-Kaisi from Core 42 and G42.
31:06Philippe Limantour from Microsoft.
31:14And Nat Ives from NVIDIA.
31:20I'm Olivier Oulieu.
31:21Thank you very much for your attention.
31:23And thank you, VivaTech, for this opportunity.
31:26Have a great day.
31:28And thank you, Olivier, for that.
31:31Well directed and moderation.
Comments