WEBVTT 1 00:00:04.000 --> 00:00:07.559 With Laurent's Segeleen from London and Gerard read from Berlin. 2 00:00:08.080 --> 00:00:10.759 This is redefining energy today. 3 00:00:10.880 --> 00:00:14.800 On redefinding energy Jihad, it's a very hot topic. We're 4 00:00:14.800 --> 00:00:17.800 going to talk about AI data centers and the cost 5 00:00:17.839 --> 00:00:19.120 of speed. Yeah. 6 00:00:19.120 --> 00:00:20.640 I like the way I say that the cost of 7 00:00:20.719 --> 00:00:23.000 speeder on because it definitely does say that we're in 8 00:00:23.039 --> 00:00:26.960 an AI race and the winner seems to be the 9 00:00:27.000 --> 00:00:29.679 one that's the fastest. At least that's what the industry 10 00:00:29.719 --> 00:00:30.320 thinks anyway. 11 00:00:30.719 --> 00:00:34.399 The first award from our new partner. Today's show is 12 00:00:34.439 --> 00:00:38.960 supported by the BMW Foundation, Harvard quant The BMW Foundation 13 00:00:39.200 --> 00:00:43.520 unites leaders from diverse sectors to develop solutions that foster 14 00:00:43.719 --> 00:00:47.920 an innovative economy and a future proof society. A key 15 00:00:48.000 --> 00:00:52.159 focus is the energy transition and climate change, where the 16 00:00:52.159 --> 00:00:57.679 Foundation drives international collaboration to accelerate the energy transition. With 17 00:00:57.840 --> 00:01:02.560 rising energy demands from AI and data centers, new partnerships, 18 00:01:02.679 --> 00:01:06.959 effective collaboration, and the exchange of science based solutions and 19 00:01:07.000 --> 00:01:11.519 strategies are essential. That's why the BMW Foundation supports this 20 00:01:11.640 --> 00:01:15.799 podcast and bring these discussions to global stages by hosting 21 00:01:15.799 --> 00:01:19.920 the Energy Security Hub at the Munich Security conference in 22 00:01:19.920 --> 00:01:25.120 twenty twenty six, streaming live from February twelve to the fourteenth. 23 00:01:25.519 --> 00:01:28.400 You can learn more at the BMW Foundation dot org. 24 00:01:28.599 --> 00:01:31.719 Back to the show. Yeah, and of course the greed 25 00:01:32.159 --> 00:01:35.760 being on decade long planning, and all of a sudden 26 00:01:35.799 --> 00:01:38.799 you get three mega what's coming in one giga? What's 27 00:01:38.840 --> 00:01:43.599 coming there? And the grid simply cannot follow, which means 28 00:01:43.640 --> 00:01:46.359 that it's all in on deck and there's a lot 29 00:01:46.359 --> 00:01:50.120 of things happening behind the meta. And you know, we're 30 00:01:50.120 --> 00:01:54.280 going to talk in details about those OCGT open cycle 31 00:01:54.359 --> 00:01:59.359 gastar binds or maybe fuel cells and batteries. Basically they're 32 00:01:59.359 --> 00:02:03.599 throwing everything they can just to get some energy. 33 00:02:04.280 --> 00:02:08.199 It's a really offsy topic to talk about, so and 34 00:02:08.280 --> 00:02:11.240 it's great to actually bring an expert on that knows 35 00:02:11.240 --> 00:02:14.599 the space very well. And we ended up finding Andrew Perry, 36 00:02:14.879 --> 00:02:18.280 who's the director of the Energy Transition and Environmental Business 37 00:02:18.360 --> 00:02:23.520 Unit at Faculty AI, which is leading consultancy businesses in 38 00:02:23.560 --> 00:02:24.879 the AI space in Europe. 39 00:02:25.439 --> 00:02:28.919 Yeah, and the interesting thing is that it's coming from 40 00:02:28.919 --> 00:02:32.159 the AI world and it's not coming from the nuts 41 00:02:32.240 --> 00:02:36.360 and boards world. So he show us the moon, we 42 00:02:36.439 --> 00:02:39.159 look at the finger. We have a very good conversation. 43 00:02:39.360 --> 00:02:42.000 So let's bring on on de Perry on the show 44 00:02:42.560 --> 00:02:44.800 and the welcome to the show. 45 00:02:45.080 --> 00:02:46.680 Pleasure to be here, thank you for having me. 46 00:02:47.199 --> 00:02:50.199 Andy. Maybe let's jump straight in. I mean, what we're 47 00:02:50.240 --> 00:02:52.520 really going to do is talk about AI and energy, 48 00:02:52.719 --> 00:02:54.719 try to get a better understanding of that. But maybe 49 00:02:54.800 --> 00:02:57.400 just talk a little bit about what's going on in AI. 50 00:02:57.919 --> 00:02:59.400 I think that'd be a great starting point. 51 00:03:00.120 --> 00:03:03.479 The technology development is, if anything, progressing quicker than what 52 00:03:03.599 --> 00:03:04.840 people have been expecting. 53 00:03:05.479 --> 00:03:05.919 Recently. 54 00:03:06.120 --> 00:03:08.439 There's something called the International Mass Olympiad, which is like 55 00:03:08.479 --> 00:03:11.319 the premiere high school competition for math. And if you 56 00:03:11.319 --> 00:03:13.000 looked at the bedding markets, so like the kind of 57 00:03:13.039 --> 00:03:17.120 predicted markets, the prediction was abound twenty percent that AI 58 00:03:17.280 --> 00:03:20.960 could could win that competition in this year, so twenty 59 00:03:21.000 --> 00:03:24.039 twenty five, And actually it won that competition in July, 60 00:03:24.680 --> 00:03:27.120 and then a month later it then achieved a perfect 61 00:03:27.159 --> 00:03:31.199 score in the ICPC, which is the International Collegiate Programming Contest, 62 00:03:31.240 --> 00:03:33.680 which is basically where top university teams come together and 63 00:03:33.680 --> 00:03:37.319 solve like complex algorithmic problems. This was GPT five. I 64 00:03:37.360 --> 00:03:39.560 thinkam GEM and I both were able to achieve a 65 00:03:39.560 --> 00:03:40.439 perfect score of. 66 00:03:40.680 --> 00:03:44.280 Out twelve, which none of the human teams managed that. 67 00:03:44.159 --> 00:03:47.280 Pace of change listening incredibly quick and so it's beating 68 00:03:47.360 --> 00:03:50.960 benchmark every which way. I think the benchmarks effectively meaningless 69 00:03:51.000 --> 00:03:53.840 now because because all the benchmarks effectively and decent, And 70 00:03:53.879 --> 00:03:55.719 so when you look at the rate of progress and 71 00:03:55.719 --> 00:03:58.280 you look at it over time, it is possible actually 72 00:03:58.319 --> 00:04:01.319 to understand it in a way that doesn't always seem possible. 73 00:04:01.639 --> 00:04:04.360 So it sometimes feels like it's just this exponential thing, 74 00:04:04.439 --> 00:04:06.280 like how do we get our hands around it? But 75 00:04:06.319 --> 00:04:08.280 if you look at the last of six years, there's 76 00:04:08.280 --> 00:04:10.479 a really interesting study by a Meter, which is a 77 00:04:10.520 --> 00:04:14.439 nonprofit research organization, which tried to look at how long 78 00:04:14.439 --> 00:04:17.199 it took humans to complete tasks that AI could now 79 00:04:17.279 --> 00:04:19.560 do and how that's developed over time. 80 00:04:20.199 --> 00:04:21.639 If you look way back. 81 00:04:21.480 --> 00:04:23.480 Sort of six years ago, with GPT two, we were 82 00:04:23.519 --> 00:04:26.439 talking about tasks that took a few seconds that AI 83 00:04:26.560 --> 00:04:28.560 was able to complete, and we were quite impressed by 84 00:04:28.560 --> 00:04:31.920 that back in twenty nineteen, and then GVT three a 85 00:04:31.959 --> 00:04:33.600 couple of years later, was able to do sort of 86 00:04:33.600 --> 00:04:37.319 ten second task. Then GVT four, which you guys will 87 00:04:37.360 --> 00:04:39.480 remember that was in twenty and twenty three, you know, 88 00:04:39.600 --> 00:04:41.639 was able to do sort of five minute tasks, but 89 00:04:41.680 --> 00:04:43.399 we're now up to the point where we're able to 90 00:04:43.439 --> 00:04:45.800 do sort of one to two hour tasks with AI 91 00:04:46.319 --> 00:04:48.879 with a fifty percent success rate. So what you then 92 00:04:48.920 --> 00:04:50.199 see when you look at it that way is that 93 00:04:50.240 --> 00:04:53.720 you have this sort of doubling of ability every seven 94 00:04:53.759 --> 00:04:54.839 months or so. 95 00:04:54.920 --> 00:04:57.639 What you're actually saying is that AI what could take 96 00:04:57.639 --> 00:04:59.920 the human two hours AI is doing in five se 97 00:05:00.319 --> 00:05:01.480 That's what you're sort of saying. 98 00:05:01.720 --> 00:05:03.560 Yeah, fairly instantly, it's a now. 99 00:05:03.639 --> 00:05:05.839 So what's interesting about that is like you've basically had 100 00:05:05.879 --> 00:05:07.839 a model back in twenty nineteen that was like a 101 00:05:08.279 --> 00:05:12.319 preschooler in its intelligence, that then developed into more like 102 00:05:12.360 --> 00:05:15.240 a primary schooler and then became like a high schooler 103 00:05:15.319 --> 00:05:18.279 with GVC four and with GVD five is kind of 104 00:05:18.279 --> 00:05:20.839 like a genius high schooler that's now winning the Mass 105 00:05:20.839 --> 00:05:24.160 Olympiad and is able to complete these coding and programming 106 00:05:24.199 --> 00:05:28.360 tasks that the university level graduates are struggling with. You're 107 00:05:28.399 --> 00:05:32.040 seeing this evolution of the models that it's kind of 108 00:05:32.040 --> 00:05:33.800 progressing at about three times the rate of what a 109 00:05:33.920 --> 00:05:36.800 child would in its development, which is really impressive looking 110 00:05:36.839 --> 00:05:39.399 backwards when you start looking forwards and you think, well, okay, 111 00:05:39.439 --> 00:05:42.600 if you take that progression that has been pretty consistent 112 00:05:42.600 --> 00:05:44.839 for the last six years and play it forwards, you 113 00:05:44.879 --> 00:05:50.120 start getting into some very interesting, expiking gary types of 114 00:05:50.120 --> 00:05:51.920 scenarios of what can happen in the. 115 00:05:51.879 --> 00:05:53.360 Next three or five years. 116 00:05:53.600 --> 00:05:55.600 And it helps to explain a bit some of the 117 00:05:56.079 --> 00:06:00.240 what seemed like fairly crazy investments and predictions that being 118 00:06:00.279 --> 00:06:02.959 made right now in the market of like where things could. 119 00:06:02.800 --> 00:06:05.759 Go, because at the end of the day, it's probably 120 00:06:05.800 --> 00:06:08.759 a bit of a winner take hold. Yeah, and I 121 00:06:08.800 --> 00:06:12.439 know that there are probably five or ten big guys 122 00:06:12.639 --> 00:06:17.000 literally putting everything they have hoping to be or do 123 00:06:17.000 --> 00:06:20.560 not know the Microsoft or the Google of tomorrow factor ten, 124 00:06:21.279 --> 00:06:25.360 But of course there's more applicants than final winner. How 125 00:06:25.399 --> 00:06:28.680 do you see even the structure of the industry like 126 00:06:28.839 --> 00:06:31.959 new player, more of the same or some which you 127 00:06:32.040 --> 00:06:35.560 throw the towel and say, okay, fine, what's the feeling 128 00:06:35.639 --> 00:06:36.160 right now? 129 00:06:36.920 --> 00:06:39.199 I think you're right that through the winner takes All 130 00:06:39.319 --> 00:06:42.879 dynamics is definitely a play part of that is because 131 00:06:43.120 --> 00:06:45.439 there's a perception there's a sort of a three to 132 00:06:45.519 --> 00:06:48.600 five year window looking to like twenty twenty seven, eight 133 00:06:48.800 --> 00:06:52.680 twenty thirty, where if you get the same levels of 134 00:06:52.720 --> 00:06:56.000 development that we've seen so far, we will achieve some 135 00:06:56.040 --> 00:07:00.240 sort of AGI, so artificial general intelligence effectively AI being 136 00:07:00.279 --> 00:07:03.240 on a general basis at the level of the smartest humans. 137 00:07:03.759 --> 00:07:06.639 If you get to that point, then you start moving 138 00:07:06.639 --> 00:07:10.759 into a position of potential platform for superintelligence. So one 139 00:07:10.759 --> 00:07:12.879 of the reasons why you see this dynamic right now 140 00:07:12.920 --> 00:07:16.480 is because everyone's chasing this dragon, racing to be the 141 00:07:16.519 --> 00:07:18.279 first to get there or to be one of the 142 00:07:18.279 --> 00:07:21.439 first to get there, and that explains probably two things. 143 00:07:21.480 --> 00:07:25.319 Explains the massive investments in compute, and it explains the 144 00:07:25.360 --> 00:07:29.079 massive investment into talent, and METSA is probably the best 145 00:07:29.120 --> 00:07:32.120 example of people who are doing this. So the hundreds 146 00:07:32.160 --> 00:07:37.120 of billions going into computational infrastructure and hundreds of millions 147 00:07:37.199 --> 00:07:40.439 and billions being spent on talent. The reason for that 148 00:07:40.560 --> 00:07:43.319 is because if you look at what drives development and 149 00:07:43.319 --> 00:07:45.720 what will drive these improvements over the coming years, and 150 00:07:45.879 --> 00:07:48.360 the sort of order of magnitude improvements that we've seen 151 00:07:48.360 --> 00:07:53.000 over the last six is primarily compute and investment into that, 152 00:07:53.079 --> 00:07:55.240 and we've seen as sort of a five X on 153 00:07:55.800 --> 00:07:59.800 Moore's law in terms of computational improvement and algorithmic efficiency, 154 00:08:00.079 --> 00:08:01.480 and that's where the talent comes in, that's where the 155 00:08:01.480 --> 00:08:04.439 brain power comes in. Where you have a similar impact 156 00:08:04.480 --> 00:08:07.240 to computes. Probably it's probably underestimated in the main, but 157 00:08:07.240 --> 00:08:11.199 it's actually probably simlar to compute where algorithmic improvements are 158 00:08:11.279 --> 00:08:13.560 improving the performance of the models, improving the efficiency of 159 00:08:13.600 --> 00:08:15.720 the models, improving their ability to use the compute, and 160 00:08:15.759 --> 00:08:18.680 therefore improving their performance. So right now you have this 161 00:08:18.839 --> 00:08:21.839 fight for those two things to move as quickly as possible, 162 00:08:22.360 --> 00:08:24.720 and because of the trend line, if we can pinue 163 00:08:24.759 --> 00:08:26.279 on that trend line, then it's a three to five 164 00:08:26.319 --> 00:08:29.120 year window to get something that people would expect to 165 00:08:29.160 --> 00:08:32.639 look like an AGI level. That's really interesting in itself 166 00:08:32.679 --> 00:08:35.759 as well, because then you get this platform twowards well. 167 00:08:36.000 --> 00:08:37.840 If you're at that level, and especially if AI is 168 00:08:37.879 --> 00:08:39.840 good enough to do AI research as well as the 169 00:08:39.840 --> 00:08:43.120 best human then AI can start becoming self replicating, it 170 00:08:43.159 --> 00:08:45.600 can start producing the next AI model that will be 171 00:08:45.600 --> 00:08:48.840 beyond a level of human capability or sort of understanding. 172 00:08:48.960 --> 00:08:51.120 So I'll say it's very exciting and very scaryly very 173 00:08:51.159 --> 00:08:53.840 quickly when you go down that road. It could also pluitow, 174 00:08:53.879 --> 00:08:56.320 and there's reasons why it might not develop at that pace. 175 00:08:56.879 --> 00:08:59.000 I can see that your head is in the stars, 176 00:09:00.080 --> 00:09:03.240 our boots are in the mud, and we are the 177 00:09:03.320 --> 00:09:06.360 nuts and balls guy. And a lot of our listeners 178 00:09:06.399 --> 00:09:09.159 are from the infrastructure world, and of course they're going 179 00:09:09.200 --> 00:09:13.519 to invest into those data centers, providing equity or that, 180 00:09:14.360 --> 00:09:17.519 and hopefully the principle of infrastructures is like they like 181 00:09:17.559 --> 00:09:20.440 to see their money back now. Of course I hear, 182 00:09:20.720 --> 00:09:23.600 it's fine with German paddocks. We build all of that center, 183 00:09:23.799 --> 00:09:27.519 we'll have use for them. And by the way, the railways, 184 00:09:27.600 --> 00:09:31.320 the internet, YadA, YadA, YadA. But when you would build 185 00:09:31.320 --> 00:09:34.200 a railway, the alway will be there for the next 186 00:09:34.240 --> 00:09:37.200 twenty thirty years. So even if the trains the first 187 00:09:37.240 --> 00:09:40.120 three four years were not there, they would come over time. 188 00:09:40.840 --> 00:09:44.279 The second problem I have is the obsolescence, because I 189 00:09:44.399 --> 00:09:48.080 see the chips moving so fast in terms of improvement 190 00:09:48.559 --> 00:09:51.480 that I hear that the data center with chips which 191 00:09:51.519 --> 00:09:54.559 are three or four years old, you can literally scrape 192 00:09:54.600 --> 00:09:57.919 it and almost go back from zero. So you've got 193 00:09:57.960 --> 00:10:01.679 the obsolescence factor in the technology that you didn't have 194 00:10:01.759 --> 00:10:05.519 in the previous let's say, crazy cycle of bubbles of infrastructure. 195 00:10:06.120 --> 00:10:08.679 What's your take on my thinking right now? 196 00:10:09.120 --> 00:10:11.279 I think you're right when you look at the GPU 197 00:10:11.600 --> 00:10:15.159 side of this, there's definite obsolescence risk. These things get 198 00:10:15.200 --> 00:10:18.039 replaced every two or three years, so hard to reteep 199 00:10:18.120 --> 00:10:21.399 that capital spend if you're not using it. What's interesting though, 200 00:10:21.480 --> 00:10:23.879 right now, I think with ai and I'm not a 201 00:10:23.879 --> 00:10:27.240 financial expert, but when you look at the annual run 202 00:10:27.320 --> 00:10:32.320 rates that Opening Eye is operating, that Anthropic is operating, 203 00:10:32.399 --> 00:10:36.039 and the speed of improvement of those run rates, you 204 00:10:36.120 --> 00:10:38.600 do see some level of justification even for the numbers 205 00:10:38.639 --> 00:10:41.679 that they are being valued at and are investing. So 206 00:10:42.000 --> 00:10:44.919 I think open Aiye was at a twelve billion ARR 207 00:10:45.200 --> 00:10:48.240 quite recently and was increasing at about a billion a month. 208 00:10:48.919 --> 00:10:52.440 Anthropic was similar, so they'd achieved like a five billion 209 00:10:53.080 --> 00:10:56.720 ARR and again was improving about a billion a month. 210 00:10:57.120 --> 00:10:59.360 So when you compare to say. 211 00:10:59.240 --> 00:11:02.080 When people talk about terms of the AI bubble competulate 212 00:11:02.120 --> 00:11:05.000 the dot com bubble, it's like there's very real revenues 213 00:11:05.039 --> 00:11:07.639 being associated with a lot of the activity that we're seeing, 214 00:11:08.039 --> 00:11:10.360 which I think it makes it fairly unique. The pace 215 00:11:10.399 --> 00:11:14.320 of development that's being evident in actual usage. When you 216 00:11:14.320 --> 00:11:16.960 look at then the levels of investment that they're putting in. 217 00:11:17.559 --> 00:11:20.559 If you believe in the basic kind of optimistic scenario, 218 00:11:21.240 --> 00:11:23.720 or you think that optimistic scenario is likely enough that 219 00:11:23.919 --> 00:11:26.519 there's a decent chance it could happen, then if it's 220 00:11:26.519 --> 00:11:29.120 someone like Mark Zuckerberg, it makes sense to, as you said, 221 00:11:29.200 --> 00:11:32.559 drop potentially two hundred billion in investment, even if that 222 00:11:32.679 --> 00:11:35.240 may not come through, because the risk of not being 223 00:11:35.240 --> 00:11:38.000 the winner in it is too great. There's all kinds 224 00:11:38.039 --> 00:11:40.799 of financial shannagans going on around this. There's the fact 225 00:11:40.799 --> 00:11:42.960 that in Vidia is investing in companies that then use 226 00:11:42.960 --> 00:11:45.720 this infrastructure, and you've got this kind of circular capex economy. 227 00:11:46.320 --> 00:11:48.159 You've got things that you guys will understand. But the 228 00:11:48.200 --> 00:11:49.720 me to be honest, I think in terms of the 229 00:11:49.799 --> 00:11:53.159 investment vehicles being used, but there does seem to be 230 00:11:53.200 --> 00:11:57.000 a stronger base to this than other capical investment programs 231 00:11:57.080 --> 00:12:00.240 that we've seen. What's really interesting for me, I guess 232 00:12:00.320 --> 00:12:03.960 for this podcast is like the implications through energy and 233 00:12:04.279 --> 00:12:06.360 the fact that energy really is a bottleneck to this 234 00:12:06.440 --> 00:12:10.879 development and will be absolutely key to the success of 235 00:12:10.919 --> 00:12:13.600 being able to build these eighty centers and the way 236 00:12:13.639 --> 00:12:16.080 that these set companies are now becoming major energy players 237 00:12:16.440 --> 00:12:18.919 and also quite independent energy players, maybe even separate from 238 00:12:18.960 --> 00:12:19.279 the grid. 239 00:12:19.960 --> 00:12:22.440 And it's good to talk about energy. So what I'd 240 00:12:22.480 --> 00:12:25.080 like to get your view on and just explain what's 241 00:12:25.120 --> 00:12:27.919 going on here. If I take my iPhone, it uses 242 00:12:27.960 --> 00:12:30.039 maybe two killer one hours of power over a course 243 00:12:30.080 --> 00:12:31.720 of a year, and I'm charging it every day, right, 244 00:12:31.759 --> 00:12:33.240 and I use it over the whole cost of the year. 245 00:12:33.799 --> 00:12:37.360 And then I go and look at Video's new blackweld 246 00:12:37.440 --> 00:12:41.440 chip and this uses fifteen kile of what hours of 247 00:12:41.559 --> 00:12:45.360 power per day and actually the size of it is 248 00:12:45.399 --> 00:12:48.600 the size of a credit card. So just explain to 249 00:12:48.679 --> 00:12:52.360 me what's going on there, because obviously if I think 250 00:12:52.559 --> 00:12:55.080 in terms of relation there, I just can't fathom that. 251 00:12:55.159 --> 00:12:58.919 And obviously that explains then why there's massive need for 252 00:12:59.039 --> 00:13:01.759 power going for Explain that, right, what's going on there? 253 00:13:01.879 --> 00:13:05.159 Why suddenly these chips need so much power. 254 00:13:05.919 --> 00:13:08.840 That's rather than focusing on the chips themselves, maybe thinking 255 00:13:08.840 --> 00:13:11.240 about like the size of the data centers that are 256 00:13:11.279 --> 00:13:14.000 containing them. Obviously, the more power of the chips, the 257 00:13:14.080 --> 00:13:16.399 less you need for your data center and the more 258 00:13:16.440 --> 00:13:19.000 efficient you can operate. But the reason that I think 259 00:13:19.000 --> 00:13:22.240 the data centers are so big, there are free factors 260 00:13:22.480 --> 00:13:26.480 that really drive AI progress. When is this computational capacity 261 00:13:26.879 --> 00:13:29.679 and there are order of magnitude imprevements that we've seen 262 00:13:29.720 --> 00:13:33.120 so far and order of magnitude improvements that can be 263 00:13:33.200 --> 00:13:36.559 unlocked through these investments. So it's just a huge driver 264 00:13:36.679 --> 00:13:39.279 of the power that drives then your ability to move 265 00:13:39.360 --> 00:13:43.639 quickly in AI development. The second is around algorithmic efficiency 266 00:13:43.639 --> 00:13:46.559 and so the application of brain power to it. And 267 00:13:46.639 --> 00:13:48.879 the third is the way that these models operate and 268 00:13:48.919 --> 00:13:53.440 how they become unhobbled by being able to either have memory, 269 00:13:53.639 --> 00:13:55.360 or to be able to use tools, or to be 270 00:13:55.399 --> 00:13:59.200 able to operate on an agentic basis, or to reason 271 00:13:59.240 --> 00:14:03.000 deeply and think the wait that basically the model can 272 00:14:03.240 --> 00:14:06.240 enhance its own capability through that kind of unhobbling of 273 00:14:06.240 --> 00:14:08.840 the way that it operates. When you look at the 274 00:14:08.879 --> 00:14:11.240 power usage of these chips and the power usage of 275 00:14:11.279 --> 00:14:14.519 these data centers, it's purely a reflection of the fact 276 00:14:14.600 --> 00:14:17.480 that there is this raw muscle power that can drive 277 00:14:17.519 --> 00:14:20.799 AI progress. It's the seed at which you're able to 278 00:14:20.799 --> 00:14:24.919 develop and train models, and then the inference power you 279 00:14:24.960 --> 00:14:26.480 need to be able to run them, which is also 280 00:14:26.559 --> 00:14:27.480 massively increasing. 281 00:14:28.200 --> 00:14:30.799 Myself and Ron probably very much agreed. There seems to 282 00:14:30.799 --> 00:14:34.919 be just this crazy frenzy in and or around the 283 00:14:35.000 --> 00:14:38.279 view that we're going to lead not even just gigawats power, 284 00:14:38.320 --> 00:14:41.080 but one hundreds of gigawats of power to fuel this 285 00:14:41.200 --> 00:14:44.720 AI race that we're seeing. And we're obviously seeing CEOs 286 00:14:44.799 --> 00:14:50.240 or businesses, and we're even seeing financial institutions say things like, well, 287 00:14:49.919 --> 00:14:54.200 we'll never meet this AI gap in our lifetime type stuff. Right, So, 288 00:14:54.440 --> 00:14:56.559 how do you view this and how do you try 289 00:14:56.600 --> 00:14:59.159 and help us get to sort of think through and 290 00:14:59.279 --> 00:15:00.879 understand what's going to happen. 291 00:15:01.399 --> 00:15:04.759 The way energy is becoming so key to powering these 292 00:15:04.840 --> 00:15:08.360 data centers is just like transformational, and it's just increased 293 00:15:08.399 --> 00:15:11.000 so quickly in the last couple of years. And so 294 00:15:11.120 --> 00:15:13.440 you say, we're now seeing I think of a move 295 00:15:13.480 --> 00:15:16.559 into the realm of like ken gigawatt plus sort of 296 00:15:16.600 --> 00:15:19.960 data center clusters as we go into the next few years. 297 00:15:20.360 --> 00:15:23.519 We already have like one gigawatts in places like Abelene 298 00:15:23.639 --> 00:15:27.679 in Texas where Stargate's getting developed, or the Colossus cluster 299 00:15:27.840 --> 00:15:31.639 in Memphis that the XAI is developing, where you've just 300 00:15:31.720 --> 00:15:35.080 got immense amounts of energy usage that's in the hundreds 301 00:15:35.080 --> 00:15:38.360 of hundreds of megawatz, just incredibly quickly that's been sput up. 302 00:15:38.759 --> 00:15:42.559 The difference in these times between tech progress and what 303 00:15:42.600 --> 00:15:44.720 we know in the energy industry is kind of energy 304 00:15:45.000 --> 00:15:48.919 development progress, where in tech world you're moving quarter by quarter, 305 00:15:49.120 --> 00:15:53.039 and energy infrastructure you're moving in many years or decades 306 00:15:53.120 --> 00:15:55.679 and to get stuff built, and these two things clearly 307 00:15:55.799 --> 00:15:58.200 don't compute when you try and make these two things together. 308 00:15:58.960 --> 00:16:02.120 So we now see companies just taking it into their 309 00:16:02.120 --> 00:16:05.799 own hands to develop their own generation on site completely islanded. 310 00:16:06.399 --> 00:16:09.519 I'm interested to understand maybe what's happening with permitting there 311 00:16:09.960 --> 00:16:12.200 and how they've managed to get around rules around this. 312 00:16:12.279 --> 00:16:14.679 But when you look at the different types of generation 313 00:16:14.799 --> 00:16:16.639 that they have with options click, clearly there's a big 314 00:16:16.639 --> 00:16:20.639 investment in nuclear and SMR development that feels like there's 315 00:16:20.720 --> 00:16:23.759 massive hurdles to cross from a regulatory point of view 316 00:16:23.879 --> 00:16:28.000 to get new generation SMRs developed that will take ten 317 00:16:28.000 --> 00:16:32.559 plus years. Renewables are complicated, and they're cheap and they're quick, 318 00:16:32.559 --> 00:16:35.000 but the complexity of trying to provide base load power 319 00:16:35.000 --> 00:16:38.279 with them complicates matters. When all you want is on 320 00:16:38.360 --> 00:16:41.639 demand power if it's completely reliable, which kind of brings 321 00:16:41.679 --> 00:16:43.600 you back to gas, and obviously at a time when 322 00:16:43.639 --> 00:16:46.799 the US is swimming in gas. But even cggps have 323 00:16:46.960 --> 00:16:49.799 a pipeline of five plus years before you can get 324 00:16:49.799 --> 00:16:53.399 on the list. So they're now reverting to OCGT. And 325 00:16:53.399 --> 00:16:57.159 so I think in a stargate there's something like ten 326 00:16:57.279 --> 00:17:01.960 to fifteen of these operating thirty five megawats each in 327 00:17:02.480 --> 00:17:05.960 XAI facility again talk of thirty to thirty five of 328 00:17:06.000 --> 00:17:11.039 these are reoperating. So you've got like hundreds of megawatts 329 00:17:11.200 --> 00:17:14.920 these three or five hundred megawat's worth of open cycle 330 00:17:14.960 --> 00:17:18.240 gas turbines operating that I guess we'd normally expect to 331 00:17:18.240 --> 00:17:20.599 only see running at like a five to ten percent 332 00:17:20.799 --> 00:17:21.960 max load factor. 333 00:17:22.039 --> 00:17:23.200 That's being on the spaceload. 334 00:17:23.759 --> 00:17:26.960 Yeah, you're one hundred percent right. All those guys they 335 00:17:27.000 --> 00:17:29.240 made like all their climate pledge, and we're going to 336 00:17:29.279 --> 00:17:32.160 be in net Zio and the first thing they do 337 00:17:32.480 --> 00:17:34.799 is they put and I can even tell you the brand. 338 00:17:34.839 --> 00:17:38.599 It's all the same. It's the Titan three fifty by Caterpillar. 339 00:17:39.039 --> 00:17:41.119 And you know what what is great with the your 340 00:17:41.079 --> 00:17:45.640 administration is, you know andamantal law no onenviromuntalo anymore. You 341 00:17:45.640 --> 00:17:50.200 want to pollute, please pollute. It is amazing. For the 342 00:17:50.279 --> 00:17:55.799 sake of speed, the laws don't apply anymore. The only 343 00:17:55.839 --> 00:17:59.000 thing people are complaining about is when you start linking 344 00:17:59.039 --> 00:18:01.319 to the grid that it's kind of mixed with the 345 00:18:01.359 --> 00:18:04.519 great payer and I know there's going to be a backlash. 346 00:18:04.759 --> 00:18:08.960 I empathize with the necessary speed of development to reach Agi. 347 00:18:10.039 --> 00:18:14.759 There's a thickness in the way infras being built which 348 00:18:15.119 --> 00:18:19.000 you almost need to reverse engine. What is the grid 349 00:18:19.160 --> 00:18:23.400 able to deliver to me? And then start thinking not 350 00:18:23.519 --> 00:18:26.400 about nuclear because nuclear is going to come in ten 351 00:18:26.480 --> 00:18:28.519 years and it's gonna cost you three hundred dollars and 352 00:18:28.559 --> 00:18:30.440 make what I wur now if you're willing to pay, 353 00:18:30.559 --> 00:18:33.039 that's good enough. It's not gonna be now, and it's 354 00:18:33.079 --> 00:18:35.160 not gonna be cheap. It's not going to be cheap. 355 00:18:35.839 --> 00:18:38.880 If you look at the basic numbers. Why do company 356 00:18:39.000 --> 00:18:43.839 like Oaklow or Fermi who literally have zero project, a 357 00:18:43.920 --> 00:18:46.480 bunch of slide, not even a product, and there are 358 00:18:46.519 --> 00:18:49.519 value twenty billion dollar that is for me part of 359 00:18:49.559 --> 00:18:53.279 the frenzy. And there's going to be some tough awakenings 360 00:18:53.920 --> 00:18:57.599 because I have no doubt the Microsoft and Amazon they'll continue. 361 00:18:57.880 --> 00:19:00.640 But there's certain number of things in the supply chain 362 00:19:00.680 --> 00:19:04.519 which are for real and stuff which are ballooney. The 363 00:19:04.640 --> 00:19:09.640 question now is, knowing that the limiting factor is the energy, is, 364 00:19:10.000 --> 00:19:12.519 are they not going to rethink the way they do 365 00:19:12.640 --> 00:19:16.279 data centers or compute our software all this based on 366 00:19:16.319 --> 00:19:17.400 this limiting factor. 367 00:19:18.400 --> 00:19:19.839 I don't think so. Clearly. 368 00:19:19.880 --> 00:19:21.680 There's efficiencies being made all the time. I know there's 369 00:19:21.680 --> 00:19:24.400 been a lot of efficiencies made on water usage by 370 00:19:24.440 --> 00:19:27.279 having close cycle kind of water loops, for example. But 371 00:19:27.319 --> 00:19:29.599 I think it's a matter of, like everything, it won't 372 00:19:29.599 --> 00:19:31.119 be a case of, well, we made the efficiencies, we 373 00:19:31.160 --> 00:19:33.119 need less power. It's just like we've made the efficiencies, 374 00:19:33.160 --> 00:19:35.720 now we can get more out of that power. They 375 00:19:35.759 --> 00:19:38.200 will continue to build more and more because we have 376 00:19:38.279 --> 00:19:40.599 this race that's a three to five year race. I 377 00:19:40.640 --> 00:19:43.400 guess the logic would be, well, look, it's some short 378 00:19:43.480 --> 00:19:45.559 term pain. We would like to do this in a 379 00:19:45.559 --> 00:19:49.039 climate friendly way, but the reality is we can't. We 380 00:19:49.079 --> 00:19:51.279 can't move at the pace we need to, and it's 381 00:19:51.279 --> 00:19:54.759 a national security concern, et cetera, et cetera. But you 382 00:19:54.799 --> 00:19:56.200 know what if we can get to that point in 383 00:19:56.240 --> 00:19:59.440 three to five years, then the machines will fix this 384 00:19:59.559 --> 00:20:02.440 for us. We will find out solutions at that point. 385 00:20:02.519 --> 00:20:04.880 The tech is for the longer term, and this five 386 00:20:04.960 --> 00:20:07.079 year period will just be a short plit up until 387 00:20:07.119 --> 00:20:08.559 the point at which we then have a level of 388 00:20:08.559 --> 00:20:10.759 intelligence that means we can we can solve a lot 389 00:20:10.759 --> 00:20:13.079 of these problems. And I'm not saying I believe that, 390 00:20:13.559 --> 00:20:15.799 but I think that would be a justification that would 391 00:20:15.799 --> 00:20:16.319 get used. 392 00:20:17.559 --> 00:20:21.480 Excellent No, no, sorry, no sorry, I mean Jiard you 393 00:20:21.480 --> 00:20:23.680 say excellent, but me I don't say excellent because you know, 394 00:20:23.720 --> 00:20:26.160 a few take billionaires said you know what, I take 395 00:20:26.200 --> 00:20:28.960 the cash now, and it's a short term paign. But 396 00:20:29.079 --> 00:20:32.319 not worry. The future is going to be better. Leny 397 00:20:32.519 --> 00:20:35.519 used to speak like that, but it was the Communist Party, 398 00:20:35.599 --> 00:20:37.559 it was not the tech bros. But at the end 399 00:20:37.599 --> 00:20:40.480 of the day, it's a minority of people who are 400 00:20:40.559 --> 00:20:45.240 capturing the common good, which is our oxygen, and say, 401 00:20:45.440 --> 00:20:48.200