- 16 hours ago
Our need for reliable, clean power is greater than ever. From electric cars to the massive data centers training the next generation of AI, the demand for electricity is reaching a breaking point. Our aging power grid wasn't built for this digital surge or for the shift to clean energy. To prevent an energy deadlock, we must transform our old wires into a smart network that can manage power spikes in real time. But how can we upgrade our physical infrastructure to meet the needs of our software? And how will we balance the needs of energy-hungry datacenters, for AI and more, with the need for a stable, green future?
Category
🤖
TechTranscript
00:09Test, test. All right, this is working. Good afternoon. My name is Shane
00:14Szykowski. I'm the editor-in-chief of Diplomatic Courier, and I am here with
00:19Philippe Piron of GE Vernova, the CEO of their electrification unit.
00:25Good afternoon. So, these days, energy security is very much at a premium. At the
00:33same time, the global appetite for power is surging. The rapid growth of data
00:38centers has made this all the more acute. Our systems are strained, our grids are
00:43aging. So, we're sort of unready for this energy transition and the overall
00:48growing demand. So, it kind of creates this really urgent question of how do we
00:54actually meet demand reliably and sustainably? So, I really want to start
00:59this off with a framing question to set the stakes. The grid, for most people,
01:04feels extremely invisible up until the point where it falters. When that happens,
01:10it becomes everybody's problem. So, what is the true cost, both economically and
01:17societally, when the grid falls behind?
01:21Interesting questions. I think first, the grid looks invisible, but the grid, as many authors
01:30mentioned, is the largest and most complex machine ever built by a human being. And these machines,
01:38is only when it's failing, it's not only about cost, it's about lives. It's about tremendous
01:46perturbation of the society. Looks at what happened in Spain with the blackout. Looks about what happens
01:55when the grid is attacked in Ukraine. So, I would say, when the grid is falling behind, we are speaking
02:05about
02:06massive impact because the grid is the system that transforms energy into economic activity.
02:14If the grid is falling behind, then the economy is slowing down and, obviously, the digital economy is getting
02:22started. To measure the impact, difficult. I would say, even in the case of Spain, it's really difficult to
02:31understand how much GDP points you are losing because the grid is not at the right level in case of
02:39an
02:39incident, or if the grid is a bit too slow to react to the pace of the economy. So, we
02:45need to go deeper
02:46in the nature of this demand and the impact to really measure that, but it's massive. Once again, for me,
02:53the grid, the poor grid, is the spine of everything. If the grid is not there, no more telecommunication
03:01systems. No more healthcare. No more control systems, even for the defense. It's a mission-critical,
03:09vital infrastructure, which is really absolutely vital for digital economy, but globally for the world
03:18economy.
03:21So, the stakes are incredibly high. So, that also brings me to really thinking about the infrastructure gap
03:28as it exists today. So, the IEA is predicting that data centers alone will consume about as much power
03:38in 2030 as the entire country of Japan did in 2025. That is mind-boggling. So, what does it look
03:47like,
03:48from the perspective of GE Vernova, to bridge the gap between what infrastructure can deliver today and
03:56what will actually be needed in 2030?
04:00It's a question of speed, velocity in order to scale up at the right speed the different needs.
04:07It's not only about just adding some capacity in front of new demand. The demand is really changing. If you
04:15think about what's happening today with the AI factories, the nature of the poor profile is very
04:23different, very different from a traditional cloud data centers. The level of flexibility you need and
04:33this power stability is really incredible. If I take the example of the load factor
04:39of fluctuation between the typical cloud data centers that will have probably maximum fluctuation of 1.5
04:48megawatt into several seconds. In the case of an AI factories, especially on the training phase, we are
04:57speaking up to 15 megawatt in milliseconds. So, you understand the question is obviously to ramp up the
05:07capacity to deliver some more demand, but to build a different need and electrification system to allow
05:15in a completely safe and secure situation to manage this high flexibility or the high fluctuation of load
05:22factors, because the system to generate the poor cannot resist to that. Whatever it is,
05:27a gas turbine or I would say a renewable system like a wind farm or even a nuclear power station.
05:37Our
05:37system has not been built to manage directly so much fluctuation so fast. So, that's why at GivaNova,
05:43we are designing solutions in order to bring the right level of power stability and flexibility in order
05:49to make this possible. So, it's a question of ramping up the capacity to follow and that's what we're doing.
05:56our company has been doubling all the manufacturing capacities in terms of equipment and system over the
06:02last two years. But it's as well the nature of what we have to adapt to this new economy and
06:08this new
06:08profile of load, which will make the grid much more smarter and capable to manage this high
06:17versatile profile of demand and load, which is really unprecedented in the history of the electrical industry.
06:28it's a really unique time and it's going to be very expensive to do this. And that's how the energy
06:34and electrification infrastructure always is. But it also has this really long lifetime where it needs
06:41to be used in this way for the whole of its life to be sort of financially viable. And so,
06:47it makes the stakes
06:48of getting it right really high. And so, that makes me wonder, for some time this has been something that
06:55producers and grid makers have been thinking about, sort of the question of how to prioritize
07:02sort of between clean and traditional infrastructure in terms of cost and innovation and so on like this.
07:09And what I'm really wondering in this context is how these sort of innovations in tech, whether we're
07:15talking about AI or something else, how that is changing this conversation, this equation.
07:22Yeah, I would say that globally, the current
07:26trilemnia we are facing in all countries in the world is how to make the right optimization or the right
07:34optimized compromise
07:36between energy security, including reliability, energy affordability, because cost competitiveness is
07:45the key, and obviously energy sustainability. I think that in the case of Europe, the choice has
07:52been very clear to go to sustainability, but it has to be as well, compound by the necessity of
07:59industrial competitiveness. So today, we are trying to find the solutions that make really both possible
08:06and at the paramount, the energy security is clearly what is the most important. I think that today,
08:14when I look to the demand for AI in terms of energy, what is looking to me, ultimately very important,
08:23is going to be beyond the prerequisite of being the right level of power, quality, security, reliability,
08:30which is a must-have when you have 99.99% of systemability as a requirement, is going to be
08:37the competitiveness first. Why? Because ultimately, I would say the cost per token is going to arbitrate
08:47who is going to lead the race between the different
08:51continent of the world in this AI race. And the cost per token, like for like,
08:57is going to be much more influenced by the cost of energy or the energy efficiency in the different system,
09:05than definitely with the other kind of cost that obviously you can face on the data centers. Obviously,
09:13the servers, the GPUs, but today, I would say GPUs are really delivered by a very limited number of
09:19companies. So typically, we can expect that will be the same level of cost. Maintenance and miscellaneous
09:25freed is not really the big topic, it's the energy. So what we need to consider is how to make
09:32sure
09:32that we can run an electrical infrastructure that could deliver the really best cost per token
09:42driven by this cost of energy. And today, I see two points. Because between North America,
09:49Europe, or China, you've got very different level of average cost of energy. Typically, what we call
09:55level of cost of electricity, it could be very different. In the US, based on cheap gas,
10:01you can reach 40 ish, I would say dollar per megawatt. In China, a lot of energy and including
10:12renewable energy quite subsidized will be as well in this range. In Europe, it depends, we are more in
10:18the 50 ish. And that could fluctuate between what we call in France as the most, I would say,
10:25interesting opportunity on nuclear, which is just below that. But as well on renewables, when it comes to
10:31offshore wind, that would be much higher, solar will be in the same range. So the question for us is,
10:37in Europe, at least in terms of overall competitiveness, and I would say sovereignty,
10:44how we built data centers on AI factory that will have the best energy efficient infrastructure.
10:52It's about cooling, it's about the power distribution architecture and topology we are selecting,
10:57and it's about this ability to have really very resilient and available systems for the world
11:05functioning of the data centers. I think that today, Europe has really some interesting advantage.
11:12Obviously, if you take the image of the AI layer cake of NVIDIA, Europe is not necessarily the world,
11:21the best position when it comes to the GPUs. But historically, on the electrical design,
11:29electrical infrastructure energy, we have been really the historical engineers player here,
11:35whatever is the company in terms of OEMs. So the question is going to propose in the future,
11:41the best possible system infrastructure for the data centers to minimize power losses,
11:47and to make sure that we have really the best cost efficient systems in terms of cost per token.
11:54Sure, that makes perfect sense to me. You really covered really well what the challenges
11:59for meeting this demand. I'm wondering, just in terms of, we also know that AI should be able to have
12:07some empowering impacts, whether we're talking about grid optimization or predicting failures or
12:13something like this. And so I'm wondering if there's an extent to which sort of GE Venova has looked at
12:19this
12:19AI helping to make the grids themselves better.
12:23Well, AI is very important. That's the debate as well. We've got
12:28power for AI or AI for power. So typically in our company, we are injecting AI in each and any
12:35core
12:36processes of the company we have to become more cost efficient and more productive and reducing our lead times.
12:42So that's really an enterprise journey we are entering and we are going at full speed.
12:48After, I would say, our destiny, because we are serving the hyper-scalors, is really to inject as well
12:54in our product, in our systems, some AI level and algorithm. The best example today is
13:02typically our solutions when it comes in terms of pure software, how to supervise the overall grid or
13:09how to supervise global data centers from the pro-generation to the rack, is really to make sure that the
13:18grid orchestration software we have is allowing the best level of simulation, prediction, and capability
13:25to optimize the whole chain. That's where we are really working in when we are serving this
13:32hyper-scalors. We are offering power to rack from on-site power generation, power evacuation, but as well
13:39grid connectivity, power stability, and enhance power distribution. All of this is already
13:47optimized by AI. It will continue to be in order to have the most automated solution that offers
13:53the best cost per token.
13:55That makes sense to me. You also spoke a little bit about the regionality. When we spoke in the past,
14:05you talked about a sort of regional race involving China, Europe, and the US. I'm curious,
14:14what are the measures of this race and what's at stake?
14:19Well, I would say the race is everywhere. It's a global race because ultimately all these different
14:28continents are entering into a race which is for overall economical competitiveness, national security
14:35too. So everybody is going to look first at the level of system availability of their different
14:43infrastructure. So when I was speaking about 99.99, that's a reality. And after we come back to the
14:49question from my viewpoint of competitiveness, obviously Europe, which has a particular goal
14:56and purpose about sustainability, is going to have really a different metrics because the level of low
15:02carbon energy is going to play a role. It has to be connected to the competitiveness, but mixing
15:11the quite cost-efficient nuclear energy on existing power plants with some of the renewables or whatever
15:19it is, hydro, onshore wind or solar, then you can have as well an interesting equation to be brought
15:27in front of this competition. So for me, yes, the question is going to be around these three dimensions.
15:33First, the reliability security, and just after that, the cost per token. And finally, I would say to what
15:42extent the low carbon footprint is going to play a role, not only because of the sustainability, which is
15:50critical for the future generation, and that's what we are committed to at Giovanova, but as well in energy
15:56security, because the renewable is still homegrown energy resource in Europe, so that's the best
16:04resource to protect a certain level of sovereignty regarding these critical questions here on this
16:09continent.
16:12So each of these three major regions, China, Europe, the US, and I know it's global, but I'm really
16:18interested in these because you've said that Giovanova is involved in all three of these, and they're
16:23really driving a lot. They each have different approaches and strengths, but I'm interested now
16:30in digging more specifically into data centers, and grid build out, and the greening and modernization
16:36of grids. So if this is in some way a race, then what does winning look like for Europe?
16:43Well, I think Europe has a definite play. For me, I talked about the fact that in this AI
16:53stack, we do have really some interesting assets and experience and expertise in order to optimize
17:02the data center electrical infrastructure, probably as well on the cooling side. We've got as well some
17:09interesting startups when it comes to Europe with new models. I would say that even
17:20today, we have quite a long battle regarding the LLM models with the US and to some extent Chinese,
17:26but if I take the case of AMI Labs, for example, with the world models and so on, I think
17:33there will
17:33be as well some new startup company providing some disruptive innovation that could play really an
17:40interesting game here. And finally, the energy, the first layer of the AI layer cake is going to be a
17:50real topic for us,
17:51if it's well organized between the models, the infrastructure of the data centers, and the foundation.
17:59overall, the AI is all these different layers on the stack. It's not only the pure model, it is the
18:06nice
18:07intrication interconnection between the software application, the models, obviously, the data center
18:15infrastructure systems, the chipsets, and the energy. And that's what we think, in terms of expertise,
18:22experience, we have really a play to go for in Europe. It's going to be a tight and fierce race
18:30to the excellence.
18:34Thanks for that. I want to think a little bit about the future.
18:40I want to think a little bit about the future. So, GE Vernova, again, is deeply involved in these three
18:46regions. And I'm wondering about the world in 10 years. What does an optimum, realistic state of
18:54affairs, do you think, look like, if you can in all three markets, but especially in Europe, that sort of
19:03is both optimum, but also viable? Could you reformate it, please? I do not get that.
19:11I'm sorry. So, I'm thinking of the state of the world in 10 years, and the grid build-outs, China,
19:20Europe,
19:20the US. What would an optimum state look like, especially in Europe? But if we have time,
19:26you might be able to touch on the other two markets as well, because GE is involved.
19:34Difficult questions, I would say. For the future, I think that the AI is going to be inevitable for
19:43the three continents. The question is, to what extent we'll be able to develop really the right
19:50solutions? I think in Europe, we are going more into vertical models, which could be very much
19:57leading the way with different industries. North America has been selecting much more global approach.
20:05And when it comes to China, more complicated to say to what extent it will impact the rest of the
20:10world. What I know is that every continent is going to play this game. We have different
20:17weaknesses and strengths here, and I will have some difficulty to predict what's going to be a
20:23winner. Probably all of them will have some good win in different fields of their preference.
20:31We'll see that in five or ten years. It's still a long game to play. I think we're going to
20:40see
20:40different stage. As we experienced when it was the revolution of the internet in the 2000 years,
20:48I think we're going to have different stage and the race is not over at all. We'll see over the
20:55next
20:55years how it's going to develop. Beautiful. Thank you.
21:05I guess I would ask... I'm thinking more specifically in terms of...
21:18Sorry, one moment. I actually missed this here. Yeah, so...
21:26We had talked about these kind of specific strengths and weaknesses that each region had.
21:34And that's sort of what I was thinking of when I was talking about the optimum outcomes.
21:40So I guess I didn't quite pull that out of your answer. I'm sorry. So...
21:49The optimum outcome in Europe, I think that you sort of spoke toward. But when we had spoken on the
21:55phone
21:56before, you also spoke about, within the US especially, kind of what they're moving towards
22:02for an optimum outcome with their grid build-out.
22:10Not sure to follow. Honestly, your question here, what I would say that
22:16for me, what is important is the construction of the ecosystems as we are building here at Viva Tech.
22:22Because today, the US ecosystem is very dense, is very active. It is a necessity that the European
22:31ecosystem continue to muscle. You are all an example of these efforts. And we'll see
22:40as well different companies here, trying to continue to develop not only this technical superiority or race,
22:48but as well the necessity to have really some
22:52growing footprint in the different regions. And that's the case of many of the European companies.
22:58Like GivaNova, we are really investing a lot in order to have really this ability to play on the different
23:06continents,
23:07Asia, North America, and Europe. And that's what we're very proud to do in order to support
23:13this different development that will serve a different future for the world.
23:18GivaNova has 25,000 employees here in Europe, 60 plants, and most of our R&D centers have been some
23:26footprint here in Europe. We need to continue to play this global game, because that's what I think the
23:32world is asking for us, and to make sure that we have a better future together.
23:38GivaNova has 25,000 employees here in Europe.
23:39Beautiful. Thank you, Philippe.
23:40Thank you. Thank you, Tim.
Comments