WEBVTT - Data Center Dynamics: AI and Energy Demand 0:00:00.280 --> 0:00:03.040 This is Dana Perkins and you're listening to Switched on 0:00:03.279 --> 0:00:06.960 the BNAF podcast. Today we bring you a recording from 0:00:06.960 --> 0:00:10.040 our BNAF summit in San Francisco, which took place on 0:00:10.080 --> 0:00:13.440 the fourth and fifth of February. The panel was titled 0:00:13.680 --> 0:00:17.800 Data Dynamics. Now certainly a hot topic in energy circles 0:00:17.880 --> 0:00:20.560 has been the growth in data centers and how AI 0:00:20.880 --> 0:00:23.680 has led to a rise in demand for power. In 0:00:23.800 --> 0:00:26.360 order to meet this, big tech companies are looking for 0:00:26.440 --> 0:00:30.240 solutions and firms like Google are signing power purchase agreements 0:00:30.320 --> 0:00:33.920 in an array of technologies like nuclear, long duration energy storage, 0:00:33.960 --> 0:00:38.520 and even geothermal. While these data centers are undeniably energy intensive, 0:00:38.600 --> 0:00:42.479 efficiencies can be found when it comes to reducing load requirements. 0:00:42.760 --> 0:00:45.480 This is seen with the recent release of Deepseek's AI 0:00:45.560 --> 0:00:48.680 model in January twenty twenty five. On today's show, the 0:00:48.720 --> 0:00:52.839 panelists discuss the increasing capital expenditure on data centers, as 0:00:52.880 --> 0:00:55.400 well as what has been driving the decision making of 0:00:55.440 --> 0:00:58.120 big tech companies when it comes to this spend and 0:00:58.240 --> 0:01:01.720 how potential bottlenecks might slow them down. The panelists include 0:01:01.760 --> 0:01:06.080 Stephen Carlini, Chief Advocate Data Centers and AI at Schneider Electric, 0:01:06.400 --> 0:01:09.600 Will Conkling, head of Data Center Energy for the America's 0:01:09.600 --> 0:01:13.280 in Amia at Google, Kleiber Costa, chief commercial officer for 0:01:13.480 --> 0:01:17.160 AS Corporation, and Darwesh Singh, the founder and CEO of 0:01:17.200 --> 0:01:21.160 Bold Graphics. The panel was moderated by Mark Daily, BNF's 0:01:21.200 --> 0:01:25.319 head of Technology and Innovation. For more information about BNF 0:01:25.400 --> 0:01:27.920 Summit's taking place around the world, as well as our 0:01:28.000 --> 0:01:30.120 upcoming event in New York on the twenty ninth and 0:01:30.160 --> 0:01:32.760 thirtieth of April, an to view recordings from this and 0:01:32.840 --> 0:01:36.399 other previous events, head to about dot BNF dot com 0:01:36.400 --> 0:01:39.320 Forward Slash Summit. Right now, let's hear from our panel 0:01:39.440 --> 0:01:48.440 regarding power demand and data centers. 0:01:52.800 --> 0:01:55.160 Thank you everyone very much for joining us here today. 0:01:56.880 --> 0:02:00.480 We've heard an awful lot about the energy transition, like 0:02:00.600 --> 0:02:02.880 at all be enough events, and I imagine a lot 0:02:02.880 --> 0:02:05.240 of events that everyone here goes to. It's very energy 0:02:05.240 --> 0:02:08.600 focused group, transport focused group, but there's lots of big 0:02:08.600 --> 0:02:11.720 transitions happening in the global economy. We're here today to 0:02:11.760 --> 0:02:14.799 talk about the intersection of the energy transition with one 0:02:14.840 --> 0:02:18.480 of the other really big transformations in the economy, maybe 0:02:18.480 --> 0:02:21.000 the biggest one of our lifetime. If you believe some people. 0:02:21.680 --> 0:02:24.560 So we're here to talk about data centers, energy demand, 0:02:24.720 --> 0:02:29.360 artificial intelligence. I'm joined here today by Stephen Carlini, Chief 0:02:29.400 --> 0:02:33.080 Advocate Data Centers and AI at Schneider Electric, Will Conkling, 0:02:33.240 --> 0:02:37.400 head of Data Center Energy America's Animea at Google, Clever Costa, 0:02:38.639 --> 0:02:42.080 chief commercial Officer at the AEES Corporation, and Darbush Sing, 0:02:42.080 --> 0:02:46.960 Founder and CEO of Bolt Graphics. So the first thing 0:02:47.000 --> 0:02:49.560 that I want to talk about is really why has 0:02:49.560 --> 0:02:54.240 there been so much interest in data centers this year? Like, 0:02:54.320 --> 0:02:56.240 let's just start from the beginning. What's causing out go 0:02:56.240 --> 0:02:58.320 to you first, Star Resh, and if everyone can actually 0:02:58.360 --> 0:03:01.120 just introduce themselves and kind of their view on the 0:03:01.160 --> 0:03:03.280 industry in their first answer, that'd be great. 0:03:03.560 --> 0:03:04.440 Yeah. Thanks Mark. 0:03:05.240 --> 0:03:08.240 I'm dar Wesh Funer Ceobo Graphics. We are a semi 0:03:08.240 --> 0:03:11.800 conductory startup focusing on GPUs. My background is in building 0:03:11.840 --> 0:03:14.800 data centers, so I share a lot of the pains 0:03:15.400 --> 0:03:18.560 at the panel and in the industry in general. Mark, 0:03:18.600 --> 0:03:23.760 I think to your question, compute requirements have always been increasing. 0:03:23.800 --> 0:03:25.760 This is I think it's not new to anyone. We 0:03:25.800 --> 0:03:29.799 have phones now that are more powerful than PlayStation fours 0:03:29.880 --> 0:03:33.160 from twelve years ago. What is new, though, is the 0:03:33.280 --> 0:03:38.400 demand for ten years in the future computing power right now. 0:03:38.680 --> 0:03:41.680 And a lot of these AI companies, whether they're training models, 0:03:42.080 --> 0:03:45.880 whether they're building data centers to host these models, or 0:03:45.880 --> 0:03:48.520 whether they're making phones that can run inference on these things, 0:03:48.520 --> 0:03:51.160 they want it right now. And so I think that 0:03:51.200 --> 0:03:53.040 creates a lot of interest, a lot of hype. It 0:03:53.080 --> 0:03:57.680 also creates an opportunity for new players in the market 0:03:57.680 --> 0:03:59.720 to come in, whether they're DC builders that can build 0:03:59.760 --> 0:04:03.480 data centers in six months instead of four years, small 0:04:04.160 --> 0:04:07.240 modular nuclear reactor companies that are building these and seven 0:04:07.320 --> 0:04:09.760 years instead of twenty five years. So I think it's 0:04:09.800 --> 0:04:11.640 just like a timeline. Let's shift all this left. I 0:04:11.720 --> 0:04:13.720 want to do right now and what does that really enable? 0:04:14.680 --> 0:04:17.840 Okay, and so Stephen, could you give us a primer 0:04:17.880 --> 0:04:21.120 on how Schneider Electric relates to this conversation and when 0:04:21.200 --> 0:04:24.920 did your job change in this kind of this new 0:04:25.360 --> 0:04:26.680 hype cycle that we're going through. 0:04:27.520 --> 0:04:30.640 Yeah, Kinder Electric, if you're not familiar with this largest 0:04:31.040 --> 0:04:34.960 power and cooling solution provider for data centers in the world, 0:04:35.279 --> 0:04:37.960 and as you said in an earlier panel, you know 0:04:38.000 --> 0:04:40.719 AI is not new AI has been around for a while. 0:04:40.760 --> 0:04:43.440 We've been talking to a lot of the hyperscalers. We 0:04:43.520 --> 0:04:47.400 have staff of people signed to each hyperscaler, each large COLO, 0:04:47.520 --> 0:04:52.760 each large enterprise accounts. And we you know, we noticed 0:04:53.080 --> 0:04:55.800 five or six years ago, you know, these data centers 0:04:55.800 --> 0:04:57.880 can't be built overnight. And five or six years ago, 0:04:57.920 --> 0:04:59.680 a lot of the hyper scalers were coming to us, 0:04:59.800 --> 0:05:02.200 not only talking to us about higher densities, but higher 0:05:02.279 --> 0:05:05.120 densities and scales that we've never heard of. You know, 0:05:05.200 --> 0:05:07.800 twenty twenty megawat data centers. Back then it was a 0:05:07.800 --> 0:05:10.600 big data center. They're talking one hundred, one hundred and 0:05:10.640 --> 0:05:13.880 fifty three hundred megawat campuses back then, and we're like, wow, 0:05:13.920 --> 0:05:17.240 this is this is a big change. So the densities, 0:05:17.720 --> 0:05:22.039 you know, started to really change when Nvidia came out 0:05:22.080 --> 0:05:24.680 with you know, the A one hundreds and the. 0:05:24.600 --> 0:05:25.400 A one hundreds. 0:05:26.360 --> 0:05:29.720 It was about three years ago and and the A 0:05:29.839 --> 0:05:32.760 one hundreds were kind of the first you know, at 0:05:32.760 --> 0:05:35.680 scale deployments for a lot of the hyperscalers, and a 0:05:35.720 --> 0:05:37.280 lot of those were air cool they were twenty five 0:05:37.360 --> 0:05:40.680 kilowats parac Then you saw the grace hoppers, which were 0:05:41.800 --> 0:05:45.040 last year, which were thirty six kilowats are acting seventy 0:05:45.040 --> 0:05:48.080 two kilowats parak and now we have the Blackwells. The 0:05:48.120 --> 0:05:50.640 black Wells are one hundred and thirty two kilowats per rack, 0:05:51.160 --> 0:05:54.240 which is really pushing the limit of what we can 0:05:54.279 --> 0:05:58.000 get powered to these racks and cooling to these racks. Next, 0:05:58.240 --> 0:06:02.000 they're working on Ruben, which is the next generation. They're 0:06:02.000 --> 0:06:04.680 already we're already working on the designs for that at 0:06:04.680 --> 0:06:08.240 two hundred and forty kilowats per rack. So it's just exponentially. 0:06:08.320 --> 0:06:11.440 For years, it was just two socket x eighty six 0:06:11.680 --> 0:06:14.760 pizza box servers in these data centers, and we were 0:06:14.760 --> 0:06:17.320 doing ten kilo wats per rack. And in the last 0:06:17.400 --> 0:06:20.599 four or five years, it's just exponentially, you know, increased, 0:06:20.920 --> 0:06:23.240 and not just increased in small scale, but it's a 0:06:23.320 --> 0:06:24.599 large scale that we talked about. 0:06:25.040 --> 0:06:28.359 Okay, so will you're you're coming from Google, the company 0:06:28.360 --> 0:06:30.320 that started the all it was Google's R and D 0:06:30.400 --> 0:06:34.080 Live a transformer and Google's obviously been a huge energy 0:06:34.080 --> 0:06:38.520 procurement company for years and years though and when did 0:06:38.560 --> 0:06:40.840 when did this translate into changes in your job? The 0:06:41.000 --> 0:06:42.880 recent technology advances. 0:06:43.680 --> 0:06:47.080 Yeah, so the story for me is like I've been 0:06:47.080 --> 0:06:49.480 at Google just over ten years in some way to 0:06:49.560 --> 0:06:53.400 performed buying energy for data centers and buying renewal energy 0:06:53.400 --> 0:06:56.080 for data centers, and working with utilities and working energy supply. 0:06:56.600 --> 0:06:59.320 And you know, Google has been an AI first company 0:06:59.440 --> 0:07:01.680 since about like twenty sixteen, twenty seventeen. Our CEO has 0:07:01.720 --> 0:07:03.360 been saying that for a long time, right, And we've 0:07:03.400 --> 0:07:06.400 been infusing AI and machine learning into our products since 0:07:06.400 --> 0:07:09.679 then in various various ways that you know, we probably 0:07:09.720 --> 0:07:12.240 don't notice as users, but it's been there. And then 0:07:12.560 --> 0:07:14.320 in the last you know, eighteen months or so, it 0:07:14.360 --> 0:07:16.160 see the emergence of you know, sort of more consumer 0:07:16.240 --> 0:07:20.400 facing pure mL and AI products you know before so, 0:07:20.480 --> 0:07:23.520 like i'd say, mid twenty twenty three was kind of 0:07:23.840 --> 0:07:27.920 a watershed moment for us, but we saw the green 0:07:28.000 --> 0:07:30.640 suits of this before that, when you know, the grid 0:07:30.720 --> 0:07:36.760 started to see demand growth, you know, generally across a 0:07:36.840 --> 0:07:39.400 number of sectors and industries, right from more manufacturing and 0:07:39.440 --> 0:07:43.360 more batteries and manufacturing and more car manufacturing and more 0:07:43.400 --> 0:07:45.280 s reshoring of a lot of stuff back to the US, 0:07:46.120 --> 0:07:49.080 and that started to manifest itself, right, and utilities coming 0:07:49.120 --> 0:07:50.840 to us and saying, you know, they have to buil 0:07:50.840 --> 0:07:53.400 out their transmission system in ways that they hadn't anticipated 0:07:53.440 --> 0:07:55.920 in order to continue to serve load. And that's sort 0:07:55.920 --> 0:07:57.600 of you know, power for the course as business as 0:07:57.680 --> 0:08:02.320 usual for us. You couple that sort of like growth 0:08:02.400 --> 0:08:05.120 of going from twenty years ago or sorry for the 0:08:05.200 --> 0:08:07.920 last twenty years twenty twenty three of zero point five 0:08:07.960 --> 0:08:11.200 percent annual growth on the grid to to three four 0:08:11.360 --> 0:08:13.200 now four and a half five percent annual growth from 0:08:13.200 --> 0:08:15.960 the grid, and then you have an emergence of a 0:08:16.000 --> 0:08:18.320 new sort of large you know, data center demand with 0:08:18.400 --> 0:08:20.800 machine learning chips to what Steve has talked about with 0:08:20.800 --> 0:08:23.920 with nvidious chips and then our own internal tens of 0:08:23.960 --> 0:08:27.760 processing units. We saw in mid twenty twenty three the 0:08:27.880 --> 0:08:32.280 sort of crossover point from you know, being able to 0:08:32.360 --> 0:08:34.640 sort of generally source power where when we needed it 0:08:34.760 --> 0:08:37.040 on a timescal that made sense to us, to an 0:08:37.080 --> 0:08:40.880 acceleration of demand to what Darbush just said around you know, 0:08:41.080 --> 0:08:43.680 needing it sooner, and then a push out of lead 0:08:43.679 --> 0:08:45.439 times on some of the grid buildout and some of 0:08:45.720 --> 0:08:48.679 and serving that demand by utilities, and it's a it 0:08:48.800 --> 0:08:52.439 was kind of a collision of technology sort of product 0:08:52.920 --> 0:08:59.040 development curves and infrastructure timelines that don't always mesh well. 0:08:59.240 --> 0:09:01.200 And we've been kind of been in that that soup 0:09:01.280 --> 0:09:03.200 ever since. And it's I like to say, it's been 0:09:03.240 --> 0:09:04.800 one of the more dynamic times in my career. And 0:09:04.800 --> 0:09:06.959 I don't thpect I'll see anything like I would again, So. 0:09:08.080 --> 0:09:10.880 And clever, let's get the energy company perspective on this. 0:09:11.559 --> 0:09:13.680 When did you start noticing a big change from what 0:09:13.800 --> 0:09:14.280 was happening. 0:09:14.640 --> 0:09:18.080 Look, I think if you hear my my my colleagues 0:09:18.120 --> 0:09:21.280 here on the panel, it has been about a lot 0:09:21.280 --> 0:09:24.280 of changes over a very short period of time, a 0:09:24.320 --> 0:09:27.440 lot of volatility. I would say, you know, I think, 0:09:28.000 --> 0:09:32.440 I think you'll see, uh, these these massive growth over 0:09:32.440 --> 0:09:37.880 the past few years projections and then one day opened 0:09:38.040 --> 0:09:41.600 the papers and there's deep, deep seek there changing everything 0:09:41.640 --> 0:09:43.280 and you don't know what is true what is not. 0:09:43.520 --> 0:09:47.240 So everybody's trying to figure figure that out. Look the industry, 0:09:47.320 --> 0:09:51.480 I'm with AES corporation. I've been at AS for for 0:09:51.520 --> 0:09:53.679 about seven years there, but I've been in the energy 0:09:53.800 --> 0:09:56.880 business for about twenty five years. I started my career 0:09:56.960 --> 0:10:00.439 right when air Ron was the greatest thing on earth. 0:10:00.960 --> 0:10:04.280 Everybody wanted to work for and run not not after 0:10:04.320 --> 0:10:08.640 that and Ron, we all know what happened. And then UH, 0:10:08.679 --> 0:10:11.400 there's the boom and bust on the gas cycle. Hainesville 0:10:11.440 --> 0:10:17.760 Economics showed up with Shell Gas UH. A lot of 0:10:17.800 --> 0:10:21.680 companies also went bankrupt, and volatility in the in the 0:10:21.800 --> 0:10:24.640 energy sector disappeared for a long period of time, with 0:10:24.880 --> 0:10:29.840 very small growth over the past five years or maybe 0:10:29.840 --> 0:10:33.720 a little a little more, as Will said, is explosive demand. 0:10:33.880 --> 0:10:37.520 I think it's been one of the most dynamics periods 0:10:37.559 --> 0:10:39.360 of times of my career as well. 0:10:39.920 --> 0:10:41.040 But I guess the point here. 0:10:40.960 --> 0:10:45.280 Is that the energy the industry is not UH is 0:10:45.440 --> 0:10:49.400 very familiar with challenges and how to overcome those challenges. 0:10:49.720 --> 0:10:53.640 So I think my job really changed when when we 0:10:53.720 --> 0:10:59.559 moved from being providers of projects of or or or 0:10:59.679 --> 0:11:04.840 or or technology to partnering with some of these hyperscalers, 0:11:05.120 --> 0:11:11.440 some of these large data center companies and start putting 0:11:11.440 --> 0:11:15.200 together solutions with them. Not only at AS, not only 0:11:15.240 --> 0:11:20.120 we're one of the largest according to bn EF. Actually 0:11:20.120 --> 0:11:25.440 we're the largest provider of renewable energy to corporates over 0:11:25.480 --> 0:11:28.480 the past three years, so we know how that works. 0:11:28.520 --> 0:11:31.800 But we also own utility companies. We own utilities in 0:11:31.800 --> 0:11:36.400 Ohio and Indiana, and we work very closely with those 0:11:36.800 --> 0:11:39.720 hyper scalers and data center companies to meet their needs 0:11:39.760 --> 0:11:41.120 at the utility level as well. 0:11:41.480 --> 0:11:43.440 But again, I think the biggest takeaway. 0:11:44.559 --> 0:11:49.880 It feels very very uncertain in terms of demand projections. 0:11:49.880 --> 0:11:53.559 We're going to talk about that later. But whatever projections 0:11:53.600 --> 0:11:58.280 you look, however you slice and dice there. The growth 0:11:58.360 --> 0:12:02.719 is here, is real, and I think the industry is 0:12:04.320 --> 0:12:08.719 everybody said to meet those challenges and meet solutions for 0:12:09.600 --> 0:12:10.120 that growth. 0:12:10.440 --> 0:12:13.000 So this growth is real, we need to deal with it. 0:12:13.520 --> 0:12:16.320 What's been a big question that I've been asked actually 0:12:16.360 --> 0:12:20.320 a lot in my role is does this look different 0:12:20.400 --> 0:12:23.120 than the data center growth that came before? With the 0:12:23.160 --> 0:12:27.000 idea being data centers have been cited close to populations 0:12:27.120 --> 0:12:29.440 so that latency is low and your Netflix loads really quick. 0:12:30.160 --> 0:12:31.960 But that might not be the case for a new 0:12:32.000 --> 0:12:35.360 AI applications where actually training doesn't need to take place 0:12:35.520 --> 0:12:37.600 close to people. Maybe it'll take place in a far 0:12:37.640 --> 0:12:42.840 away place. Starsh what's your sense of this kind of 0:12:42.920 --> 0:12:44.120 regional dynamic. 0:12:45.240 --> 0:12:50.720 Yeah, definitely, AI is changing the requirements for where you 0:12:50.720 --> 0:12:53.360 build data centers, where you can source power and also 0:12:53.400 --> 0:12:56.160 the conversation. Like I think four years ago, if I 0:12:56.200 --> 0:12:58.959 had a conversation with someone about sourcing like hydrogen power 0:12:58.960 --> 0:13:00.640 for a data center, be laughed at. 0:13:01.080 --> 0:13:03.400 And now I'd be laughed at, but like a lot less. 0:13:03.600 --> 0:13:07.200 I think I'd be pointed towards other areas maybe in 0:13:07.240 --> 0:13:07.720 this direction. 0:13:08.720 --> 0:13:10.920 But definitely. 0:13:12.760 --> 0:13:15.240 The workload that that's running in the data center does 0:13:15.360 --> 0:13:18.120 impact really heavily what those requirements are where I can 0:13:18.160 --> 0:13:20.480 put that data center. And now with training and with 0:13:20.920 --> 0:13:25.160 let's say training is a batch workload, all right, HPEC 0:13:25.320 --> 0:13:28.000 supercomputing is also a batch workload that can also be 0:13:28.080 --> 0:13:32.440 an maybe choose Jensen's common It can be an Antarctica somewhere, right, 0:13:32.480 --> 0:13:34.839 it doesn't be very close to me. So yeah, that 0:13:35.240 --> 0:13:38.040 definitely does change, But I don't think it changes the 0:13:38.320 --> 0:13:41.480 economics that much because you still have to deliver a 0:13:41.520 --> 0:13:43.040 lot of power to it. You still need a really 0:13:43.080 --> 0:13:46.840 fat gigabit multi hunter gigabit network link, and these requirements 0:13:46.880 --> 0:13:50.400 like haven't really changed, I think, to be honest, Like, 0:13:50.400 --> 0:13:51.920 if you're going to build a lot of data centers, 0:13:51.960 --> 0:13:54.520 you're already going in the direction of, hey, I need 0:13:54.520 --> 0:13:57.400 to build these further and further away from large metropolitan 0:13:57.440 --> 0:13:58.920 areas because I can't get power anyway. 0:13:59.000 --> 0:14:00.520 So this is a problem like three four years ago, 0:14:00.840 --> 0:14:01.640 it's just worse now. 0:14:02.240 --> 0:14:05.079 Well, actually, interested in you, you're the energy part of 0:14:05.120 --> 0:14:08.719 the equation at Google. Can you give us a bit 0:14:08.760 --> 0:14:12.040 of information? How does that factor into decision making rom 0:14:12.080 --> 0:14:15.120 where to put data centers? Is it it's decided where 0:14:15.160 --> 0:14:16.960 it goes and then we need to find energy or 0:14:17.080 --> 0:14:18.600 is it part of the decision making process? 0:14:18.720 --> 0:14:20.000 Is that the first thing you decide on? 0:14:21.440 --> 0:14:26.840 Yes? And yes it you know historically so before you know, 0:14:26.960 --> 0:14:28.840 before the growth and the grid that I talked about 0:14:29.040 --> 0:14:31.520 kind of started. You know, it's not quite true, but 0:14:31.520 --> 0:14:33.000 you could almost like throw a dart on a map 0:14:33.040 --> 0:14:35.520 and you know, if you were somewhere within a reasonable 0:14:35.560 --> 0:14:38.240 distance of a metropolitan area, you could probably find power 0:14:38.240 --> 0:14:39.840 and a utility would at least build something for you 0:14:39.880 --> 0:14:41.680 in a couple of years, right, And and that was 0:14:41.720 --> 0:14:46.680 generally okay. You know, today, available grid and available generation 0:14:46.880 --> 0:14:49.360 on that grid are are more scarce, at least for 0:14:49.440 --> 0:14:51.520 the you know, short to medium term, right, And so 0:14:51.560 --> 0:14:53.920 you have to be sort of smarter and better at 0:14:53.920 --> 0:14:57.080 picking locations that have available power in the time scale 0:14:57.120 --> 0:15:00.000 you're looking for and and and sort of act quick 0:15:00.080 --> 0:15:04.720 to go and and reserve it and use it, right. So, 0:15:04.720 --> 0:15:08.640 so yet, yes, it's a it's a more important factor 0:15:09.320 --> 0:15:12.320 for us than maybe it was in the past. But 0:15:12.440 --> 0:15:15.200 it doesn't I think divers to your last point there. 0:15:15.200 --> 0:15:17.120 It's like if you go in the middle of nowhere, 0:15:17.360 --> 0:15:19.320 like there isn't power, there isn't fiber, there isn't there 0:15:19.320 --> 0:15:21.880 aren't people to build things, right, There's just these things 0:15:21.880 --> 0:15:26.040 require a certain amount of infrastructure and civilization around them 0:15:26.080 --> 0:15:28.040 to support them, right, and no one wants to live 0:15:28.040 --> 0:15:29.240 next to it. You have to have people that work 0:15:29.240 --> 0:15:31.960 there all the time, right like so so so there's 0:15:32.000 --> 0:15:34.960 there's that factor. And and then our products also like 0:15:35.520 --> 0:15:36.960 don't want to be in the middle of nowhere if 0:15:37.000 --> 0:15:39.320 they can help it. Because if you build a building 0:15:39.480 --> 0:15:42.600 for a data center, and and and so I'll back up. 0:15:42.640 --> 0:15:44.840 There's there's a few different ways that or its fuveent 0:15:44.840 --> 0:15:46.640 things that happen in an mL data center, right or 0:15:46.800 --> 0:15:49.080 a data center. One, you could be training at Google 0:15:49.080 --> 0:15:51.480 at least an internal model too. You could have a 0:15:51.520 --> 0:15:54.040 customer paying you to be able to train their model, right, 0:15:54.360 --> 0:15:56.480 or Three, you could be serving a model or serve 0:15:56.560 --> 0:15:59.720 doing inference and serving AI to customers, right. Only that 0:15:59.800 --> 0:16:02.480 first thing really is that flexible because our customers still 0:16:02.480 --> 0:16:04.480 want to be like within spitting distance of their of 0:16:04.480 --> 0:16:07.080 their footprint on the cloud. And then for inference, we 0:16:07.120 --> 0:16:09.120 still want to be having little latency service to our 0:16:09.160 --> 0:16:12.440 to our consumers, right. And so the internal training, yes, 0:16:12.480 --> 0:16:14.320 it's more flexible. But if you build a data in 0:16:14.360 --> 0:16:15.880 the middle of nowhere just for that and then you 0:16:15.920 --> 0:16:18.120 finish that job, you actually have a Strandard asset, right. 0:16:18.120 --> 0:16:19.720 And so if you think about efficiency of capital and 0:16:19.760 --> 0:16:21.520 how you want to be able to reuse your capital 0:16:21.560 --> 0:16:24.720 and recycle it, you actually don't necessarily always want to 0:16:24.760 --> 0:16:27.520 be just going far afield to to you know, the 0:16:28.080 --> 0:16:30.520 antarcticas or the deserts. We're might to be power, but 0:16:30.640 --> 0:16:33.240 no people or no no users or no customers. So 0:16:33.240 --> 0:16:35.160 so we tend to still have to follow where our 0:16:35.200 --> 0:16:38.080 products want to be. And and then within those you 0:16:38.120 --> 0:16:40.920 know regions or those uber regions we have to we 0:16:40.960 --> 0:16:42.680 have to go then you know, find power availability. 0:16:42.920 --> 0:16:46.960 So okay, great, and Stephen, you you have great insight 0:16:47.040 --> 0:16:50.520 into the entire kind of data center supply chain in 0:16:50.560 --> 0:16:53.920 your ola Schneider, I'm interested, do you have any kind 0:16:53.920 --> 0:16:56.480 of sense of how much of this new data build, 0:16:56.560 --> 0:17:00.680 data center build that we're seeing is specifically I related. 0:17:00.800 --> 0:17:02.680 Is that even something that makes sense? Is there an 0:17:02.720 --> 0:17:05.280 AI data center versus something else or is it just bits? 0:17:05.440 --> 0:17:08.639 Absolutely, the you know, the servers are completely different, and 0:17:08.880 --> 0:17:10.600 you know, the power and the cooling is different. And 0:17:10.800 --> 0:17:12.440 if you look at a data center, it's an AI 0:17:12.560 --> 0:17:14.920 data center today, say it's a ten megawaut data center. 0:17:15.359 --> 0:17:17.040 You know, five or six years ago, there'll be a 0:17:17.080 --> 0:17:19.600 thousand IT racks and the data hall would be huge. 0:17:19.760 --> 0:17:21.920 Now the data hall has seven d I t racks 0:17:21.960 --> 0:17:24.639 and you have all these chillers outside, you know, to 0:17:24.680 --> 0:17:27.679 support the cooling that so it's much different. It's not 0:17:27.760 --> 0:17:29.600 these Amazon warehouse type data centers. 0:17:29.720 --> 0:17:29.880 More. 0:17:29.920 --> 0:17:33.280 There's smaller, more confined in the IT rooms, a lot 0:17:33.320 --> 0:17:35.159 of power and cooling going through it. It's not a 0:17:35.200 --> 0:17:37.560 place that you know, it used to be walk around 0:17:37.560 --> 0:17:39.439 and you know your place, servers and everyone had a 0:17:39.440 --> 0:17:42.520 good time, but not anymore. It's a it's a business. 0:17:42.840 --> 0:17:45.560 But you know, as we were saying, you know, the 0:17:45.880 --> 0:17:49.919 you know, putting an asset where there's where there's power, 0:17:49.920 --> 0:17:52.639 and everybody in the world right now is saying that 0:17:52.640 --> 0:17:54.200 the data centers are going to go where. 0:17:54.040 --> 0:17:54.639 The power is. 0:17:54.680 --> 0:17:57.600 But but what we're what we're seeing and the hyper 0:17:57.600 --> 0:18:00.439 scales are all doing this as you just said, is 0:18:00.560 --> 0:18:05.119 they're they're building these training clusters that are close to 0:18:05.320 --> 0:18:08.480 where people are with the intention of using those to 0:18:08.520 --> 0:18:10.840 make money. There's not a lot of money being made, 0:18:11.320 --> 0:18:13.760 you know, training a model, and the question is how 0:18:13.760 --> 0:18:15.360 many more of these models are we going to need, 0:18:15.840 --> 0:18:17.800 so the money is going to be made and deploying 0:18:17.840 --> 0:18:20.560 them in the field. And we're seeing a big shift 0:18:20.640 --> 0:18:22.960 and this, in my opinion, is kind of the year 0:18:22.960 --> 0:18:26.040 of infernts and and we're starting to see you know 0:18:26.520 --> 0:18:29.640 a lot of these training clusters that were originally deployed 0:18:29.720 --> 0:18:32.640 just to train are now doing infrints, either full time 0:18:32.920 --> 0:18:35.480 or part time. And the other thing that we're seeing 0:18:35.560 --> 0:18:38.760 is you know, with infrints close to the users, optimized 0:18:38.760 --> 0:18:41.840 for different applications. We're not seeing those yet because AI 0:18:41.960 --> 0:18:44.120 is still developing. We're still at the beginning of this. 0:18:44.359 --> 0:18:47.320 We don't know, you know, what it's going to take 0:18:47.640 --> 0:18:50.320 to do an AI agent and a genta AI. You 0:18:50.320 --> 0:18:51.639 know what's that going to take. It's going to be 0:18:51.720 --> 0:18:54.440 multi modal how much processing, how much how much of 0:18:54.480 --> 0:18:56.520 the IT stack is going to be needed and where, 0:18:56.920 --> 0:19:00.040 and as we start inputting more and more video. You 0:19:00.640 --> 0:19:03.400 right now everything's text to text, you know, multimodal, it's 0:19:03.400 --> 0:19:05.280 going to be video, tech, text, image, it's going to 0:19:05.359 --> 0:19:08.680 be all these different things. So we can't optimize the 0:19:09.440 --> 0:19:12.360 data centers close to the users yet for inference, we're 0:19:12.400 --> 0:19:14.520 still going to have to do those in data centers. 0:19:14.800 --> 0:19:17.119 I think that's going to be the case for a 0:19:17.119 --> 0:19:17.960 few years. 0:19:18.480 --> 0:19:21.840 Okay, so things not really changing too dramatically. It's just 0:19:21.880 --> 0:19:23.800 build close to the users, like always. 0:19:25.480 --> 0:19:25.920 So clever. 0:19:25.960 --> 0:19:28.400 Actually, something I want to ask you about. I'm from 0:19:28.400 --> 0:19:30.680 Ireland and so when I hear data center, I think, 0:19:30.720 --> 0:19:34.040 oh my god, it's destroying the power system. There's like 0:19:34.119 --> 0:19:36.399 moratorium and new data centers there because it's such a 0:19:36.440 --> 0:19:38.840 large share of the power system. And there's a couple 0:19:38.880 --> 0:19:40.720 of reasons in Europe where something like this has happened, 0:19:40.920 --> 0:19:43.840 and now there's been conversations about this. This is going to 0:19:43.840 --> 0:19:46.160 happen in the United States. Kind of seems like that's 0:19:46.200 --> 0:19:48.199 calm down in the last few months. But interested to 0:19:48.240 --> 0:19:51.440 hear your thoughts on how big a challenge this will 0:19:51.480 --> 0:19:53.200 be for your business. 0:19:54.080 --> 0:19:56.800 It's hard to talk about that without politicizing things. But 0:19:58.400 --> 0:20:01.879 I think, look, there will be parts of the country 0:20:01.920 --> 0:20:05.359 where there's there would be some resistance uh to data 0:20:05.400 --> 0:20:07.879 center deployment. We are seeing some of that in the 0:20:08.200 --> 0:20:12.800 Southeast and other and other parts. I think I think 0:20:12.840 --> 0:20:15.800 it's the real answer to that is all going to 0:20:15.880 --> 0:20:20.080 depend on the solutions that we bring to to that 0:20:20.240 --> 0:20:21.760 data center load growth. 0:20:21.800 --> 0:20:23.800 When you say we, do you mean as or do 0:20:23.840 --> 0:20:24.480 you mean we. 0:20:24.680 --> 0:20:30.760 The providers of energy and together with the data center 0:20:30.840 --> 0:20:33.840 operators and the and the hyper skaters. I think a 0:20:33.840 --> 0:20:36.080