1 00:00:00,000 --> 00:00:04,117 And so, does that create that, you know, in more of a bifurcation that a more 2 00:00:04,200 --> 00:00:06,837 polarization? How do we ensure that that 3 00:00:06,920 --> 00:00:09,557 that diffusion is spread evenly? How do 4 00:00:09,640 --> 00:00:14,557 we make sure that we're [clears throat] not leaving major portions of society or 5 00:00:14,640 --> 00:00:18,357 the world behind? Because I think that's that's [clears throat] going to be the big 6 00:00:18,440 --> 00:00:21,717 issue for us going forward. Yeah, so it's it's it's interesting, right? This 7 00:00:21,800 --> 00:00:24,677 is one of those times when 8 00:00:24,760 --> 00:00:28,557 uh by definition and because of the 9 00:00:28,640 --> 00:00:30,477 rails that have been established, you 10 00:00:30,560 --> 00:00:33,277 know, as you said, right? Which is uh 11 00:00:33,360 --> 00:00:37,717 what's happened with mobile as well as what's happened with the 12 00:00:37,800 --> 00:00:39,997 uh broad, you know, essentially connectivity. 13 00:00:40,080 --> 00:00:43,717 >> Right. You have the ability [clears throat] to sort of deliver the 14 00:00:43,800 --> 00:00:46,157 tokens pretty evenly around the world. 15 00:00:46,240 --> 00:00:51,357 Right. Uh a lot more so than, let's say, uh the PC era or even the beginning of 16 00:00:51,440 --> 00:00:52,917 the mobile era, right? Because it took a 17 00:00:53,000 --> 00:00:55,157 long time for even the most smartphone 18 00:00:55,240 --> 00:00:58,597 in particular to penetrate um all of the 19 00:00:58,680 --> 00:01:01,277 world. Whereas now, it's not the case, 20 00:01:01,360 --> 00:01:04,277 right? These models and their outputs 21 00:01:04,360 --> 00:01:06,477 are pretty much available everywhere. 22 00:01:06,560 --> 00:01:09,717 And so, the question to me is 23 00:01:09,800 --> 00:01:14,077 what's the use cases that make sense, right? So, what in fact one of the demos 24 00:01:14,160 --> 00:01:16,037 I always go back to, I think this was 25 00:01:16,120 --> 00:01:18,357 even in the beginning of '23, was a 26 00:01:18,440 --> 00:01:21,197 rural Indian farmer was able to use a 27 00:01:21,280 --> 00:01:23,717 bot built on, I think, a very early 28 00:01:23,800 --> 00:01:26,637 GPT-3 or 2.5 even, 29 00:01:26,720 --> 00:01:29,157 uh essentially to reason over some farm 30 00:01:29,240 --> 00:01:34,157 subsidies that he had heard about in a local language and had it even in that 31 00:01:34,240 --> 00:01:38,077 very early days uh have it even show some agentic 32 00:01:38,160 --> 00:01:39,677 behavior, right? Like, go complete a 33 00:01:39,760 --> 00:01:41,837 form for me. So, in some sense, it took 34 00:01:41,920 --> 00:01:45,397 uh you know, it brought back agency to 35 00:01:45,480 --> 00:01:48,757 someone uh who perhaps didn't have that 36 00:01:48,840 --> 00:01:53,157 because the technology was so much more accessible. So, I I do think it's in the 37 00:01:53,240 --> 00:01:55,637 our hands, even in the global south, to 38 00:01:55,720 --> 00:01:59,557 use it uh to create, I would say, more 39 00:01:59,640 --> 00:02:05,027 of that opportunity where there isn't one. Um but I think what the necessary 40 00:02:05,110 --> 00:02:08,877 [clears throat] conditions still are, do you have uh the 41 00:02:08,960 --> 00:02:11,557 capital investment being put in? 42 00:02:11,640 --> 00:02:14,437 Uh do you even have an environment for capital? And because in an interesting 43 00:02:14,520 --> 00:02:18,117 way, we are, for example, as hyperscalers investing all over, right? 44 00:02:18,200 --> 00:02:22,037 Including the global south. Uh so, as long as there's an environment which 45 00:02:22,120 --> 00:02:24,757 attracts the capital investment >> And you see the demand. 46 00:02:24,840 --> 00:02:27,037 >> And you see And then, yeah, the demand is there. 47 00:02:27,120 --> 00:02:29,637 >> Yeah. Um and so, the question is, how do 48 00:02:29,720 --> 00:02:34,397 you have a set of policies that allow for both the capital to come in, for it 49 00:02:34,480 --> 00:02:37,877 to find nexus where There are certain things, by the way, private capital can 50 00:02:37,960 --> 00:02:43,037 do, certain things that public capital only can do. For example, the grid. 51 00:02:43,120 --> 00:02:47,157 Right? It's not I mean, grid in most countries is sort of fundamentally 52 00:02:47,240 --> 00:02:51,517 driven by governments. >> And public. And so, if you [clears throat] don't have a 53 00:02:51,600 --> 00:02:54,477 sophisticated or rather, if you don't have 54 00:02:54,560 --> 00:02:59,197 a real approach to modernizing the grid, uh that will hold things back. I mean, 55 00:02:59,280 --> 00:03:02,477 there's a lot of talk about behind the meter and so on. And yes, there's some 56 00:03:02,560 --> 00:03:03,997 amount of that we can do ourselves. 57 00:03:04,080 --> 00:03:05,997 >> do that in the US. Many countries can't. 58 00:03:06,080 --> 00:03:07,637 Exactly. And it's not long-term 59 00:03:07,720 --> 00:03:09,477 scalable, right? I mean, like, to me, a 60 00:03:09,560 --> 00:03:13,397 long-term scalable solution is to have 61 00:03:13,480 --> 00:03:17,917 um you know, all of these token factories part of the real economy, 62 00:03:18,000 --> 00:03:22,037 connected to the grid, connected to the telco network, delivering 63 00:03:22,120 --> 00:03:26,597 uh just like we deliver bits, you have to deliver tokens plus bits. 64 00:03:26,680 --> 00:03:30,637 Um and that's kind of what's going to drive an at scale, whether it's in the 65 00:03:30,720 --> 00:03:34,357 global south or in uh on the developed world. So, so many 66 00:03:34,440 --> 00:03:37,837 people talk about there may be an AI bubble. I mean, the most important thing 67 00:03:37,920 --> 00:03:40,797 that we see as an investor is 68 00:03:40,880 --> 00:03:44,117 the the democratization of technology 69 00:03:44,200 --> 00:03:46,317 and the diffusion of that technology 70 00:03:46,400 --> 00:03:49,477 really does then transform the demand. 71 00:03:49,560 --> 00:03:53,637 And the the companies or the countries that diffuse it fastest are going to be 72 00:03:53,720 --> 00:03:56,637 the ultimate winners, not the technology creator. 73 00:03:56,720 --> 00:03:59,677 That's that's you know, it's it's it's 74 00:03:59,760 --> 00:04:00,717 uh 75 00:04:00,800 --> 00:04:04,197 the for this not to be a bubble Yep. by 76 00:04:04,280 --> 00:04:06,837 definition, it requires 77 00:04:06,920 --> 00:04:10,037 that the benefits of this are much more 78 00:04:10,120 --> 00:04:11,637 evenly spread. I mean, I think a 79 00:04:11,720 --> 00:04:15,597 telltale sign of if it's a bubble would 80 00:04:15,680 --> 00:04:18,477 be if all we are talking about are the 81 00:04:18,560 --> 00:04:23,317 tech firms. Uh right? If uh all we talk about is 82 00:04:23,400 --> 00:04:27,437 what's happening to the technology side, that then that's by you know, it's just 83 00:04:27,520 --> 00:04:31,397 purely supply side. >> Right. Uh ultimately, if we are not 84 00:04:31,480 --> 00:04:35,637 talking about, "Wow, here is a drug you know, drug that was sort of uh brought 85 00:04:35,720 --> 00:04:39,917 into the market that's super successful because it was uh AI accelerated the 86 00:04:40,000 --> 00:04:43,437 clinical trial." It's not even the magical molecule, right? It's kind of 87 00:04:43,520 --> 00:04:45,757 even the rest of what is 88 00:04:45,840 --> 00:04:47,957 uh needed in order to make something 89 00:04:48,040 --> 00:04:50,917 much more relevant, right? Um and so, 90 00:04:51,000 --> 00:04:55,437 the more we [clears throat] have uh And by the way, it's happening, right? So, 91 00:04:55,520 --> 00:04:57,117 I'm not sort of saying that that's why 92 00:04:57,200 --> 00:04:59,997 I'm much more confident that this is a 93 00:05:00,080 --> 00:05:06,037 technology that will, in fact, build on the rails of cloud and mobile, diffuse 94 00:05:06,120 --> 00:05:09,317 faster, >> Yep. and bend the productivity [clears throat] 95 00:05:09,400 --> 00:05:12,637 curve, and bring local surplus and 96 00:05:12,720 --> 00:05:15,117 economic growth all around the world, 97 00:05:15,200 --> 00:05:19,797 not just economic growth driven by capital expenses. 98 00:05:19,880 --> 00:05:25,117 Uh right? Because that's it's a narrow point in time Right now, that's what 99 00:05:25,200 --> 00:05:26,957 we're seeing more >> That's what we're seeing in you know, 100 00:05:27,040 --> 00:05:29,757 you know, in in the developed world in particular. 101 00:05:29,840 --> 00:05:34,117 Uh but remember, my capital, like that the one thing that, you know, is 102 00:05:34,200 --> 00:05:37,797 definitely is spending a lot of it in the United States, but a 50% of it is 103 00:05:37,880 --> 00:05:42,157 also all over the world. >> Right. Um and so, interestingly enough, 104 00:05:42,240 --> 00:05:45,517 it depends on uh demand all over the world. And the 105 00:05:45,600 --> 00:05:49,557 demand all over the world will only be there if there is local surplus all over 106 00:05:49,640 --> 00:05:51,437 the world. And so, that's sort of the 107 00:05:51,520 --> 00:05:53,797 way I see the equation. So, let's drill 108 00:05:53,880 --> 00:05:57,317 down a little more. As AI diffuses, 109 00:05:57,400 --> 00:05:59,597 obviously, organizations, companies, 110 00:05:59,680 --> 00:06:01,997 governments are going to have to evolve. 111 00:06:02,080 --> 00:06:04,757 Now, getting into the demand side. So, 112 00:06:04,840 --> 00:06:09,317 how do you think the structure of organizations is changing in an AI 113 00:06:09,400 --> 00:06:10,717 world? 114 00:06:10,800 --> 00:06:14,037 Across roles, across teams, management. 115 00:06:14,120 --> 00:06:18,877 I'm I'm sure um Microsoft has evolved itself. 116 00:06:18,960 --> 00:06:20,677 So, I it'd probably be good to tell the 117 00:06:20,760 --> 00:06:23,357 audience how do you see this diffusion 118 00:06:23,440 --> 00:06:27,997 occur in the utilization at the corporate level or at a government level 119 00:06:28,080 --> 00:06:31,597 that which ultimately then creates that demand, which eliminates any fears and 120 00:06:31,680 --> 00:06:35,957 bubbles. Yeah, now I think it it's probably one of the the big challenges 121 00:06:36,040 --> 00:06:38,277 with all of these new technologies is 122 00:06:38,360 --> 00:06:39,597 when uh 123 00:06:39,680 --> 00:06:44,637 work work artifact and workflow changes, 124 00:06:44,720 --> 00:06:48,917 uh that means we as firms have to change 125 00:06:49,000 --> 00:06:52,517 how we work. I In fact, [clears throat] I remember meeting um 126 00:06:52,600 --> 00:06:56,797 uh the CEO of Generali, you know, a few years back, and he was describing he had 127 00:06:56,880 --> 00:07:01,087 joined um uh the firm, you know, pre-PC 128 00:07:01,170 --> 00:07:04,997 [clears throat] era, and uh uh and he was describing how, for example, they 129 00:07:05,080 --> 00:07:07,837 worked with their agents in the field uh 130 00:07:07,920 --> 00:07:11,277 with faxes, interoffice memos, and 131 00:07:11,360 --> 00:07:15,437 uh and and suddenly the PC showed up, and people would then put a spreadsheet 132 00:07:15,520 --> 00:07:19,717 in an email and send it around, and the entire workflow and the work process 133 00:07:19,800 --> 00:07:22,437 changed. Right? So, similarly, I think 134 00:07:22,520 --> 00:07:24,917 with AI, uh you are going to start 135 00:07:25,000 --> 00:07:29,877 seeing uh actual change in how workflow happens, right? 136 00:07:29,960 --> 00:07:32,517 In [clears throat] fact, for me, coming to Davos, you know, 137 00:07:32,600 --> 00:07:37,237 whatever 50 bilateral meetings I have, preparing for those 138 00:07:37,320 --> 00:07:39,037 uh had a particular workflow, right? 139 00:07:39,120 --> 00:07:43,517 Which is uh there used to be my field team would prepare notes, and that would 140 00:07:43,600 --> 00:07:45,357 come to my HQ, and that would get 141 00:07:45,440 --> 00:07:47,517 further refined. And nothing had really 142 00:07:47,600 --> 00:07:50,597 changed, right? Since I joined in '92 to 143 00:07:50,680 --> 00:07:52,637 essentially even a few years back. 144 00:07:52,720 --> 00:07:56,597 Whereas now, I just go to Copilot and say, "Hey, I'm meeting Larry. Please 145 00:07:56,680 --> 00:08:01,077 give me a brief." And it comes back and gives me By the way, the one nice thing 146 00:08:01,160 --> 00:08:05,477 is, it gives me a 360, right? It knows what we're doing with you as a client, 147 00:08:05,560 --> 00:08:08,547 what we're doing as a client of yours, and everything in between 148 00:08:08,630 --> 00:08:13,237 [clears throat] as an investment. It's so, it captures even information unlike 149 00:08:13,320 --> 00:08:17,437 anything else. In fact, what I do is I take that and immediately share that 150 00:08:17,520 --> 00:08:21,117 back with all my colleagues across all the functions, right? Think about it. 151 00:08:21,200 --> 00:08:25,597 It's a complete inversion of how information is flowing in the 152 00:08:25,680 --> 00:08:29,957 organization. It's not like this classic, we have an organization, we 153 00:08:30,040 --> 00:08:33,677 have departments, we have these specializations, and the information 154 00:08:33,760 --> 00:08:36,757 trickles up. No, no, no. It actually It 155 00:08:36,840 --> 00:08:40,077 flattens the entire information flow. 156 00:08:40,160 --> 00:08:42,957 So, once you start having that, you have 157 00:08:43,040 --> 00:08:45,797 to redesign structurally. Uh so, the 158 00:08:45,880 --> 00:08:49,717 current structure may not make sense. 159 00:08:49,800 --> 00:08:52,797 Um because you want people to be able to 160 00:08:52,880 --> 00:08:56,557 work in a way that allows them to have 161 00:08:56,640 --> 00:08:59,437 this information flow freely. So, what 162 00:08:59,520 --> 00:09:03,637 all this leads me to, if I have to sort of say what's the formula, the formula I 163 00:09:03,720 --> 00:09:05,637 think it starts with the mindset. So, 164 00:09:05,720 --> 00:09:09,837 the mindset we as leaders should have is 165 00:09:09,920 --> 00:09:13,237 we need to think about changing the work 166 00:09:13,320 --> 00:09:16,397 the workflow with the technology. Then 167 00:09:16,480 --> 00:09:20,477 that needs skill set. So, you can't sort of talk about this in the abstract. You 168 00:09:20,560 --> 00:09:24,077 got to use it. Like, so if I'm not using the >> You have to trust it. 169 00:09:24,160 --> 00:09:28,277 >> You have to trust it, you have to use it, you have to learn even how to put 170 00:09:28,360 --> 00:09:31,837 the guardrails to trust it, right? So, you can't again, you can't just be 171 00:09:31,920 --> 00:09:35,117 afraid of it. Uh it's going to it's going to be 172 00:09:35,200 --> 00:09:40,117 diffused. So, the question is, as a firm, you have to use it to learn how to 173 00:09:40,200 --> 00:09:43,197 even uh put the guardrails that allow you to 174 00:09:43,280 --> 00:09:46,437 be able to trust it. So, skills, uh so, 175 00:09:46,520 --> 00:09:48,717 mindset, skills. The other big 176 00:09:48,800 --> 00:09:52,237 consideration um really is how do you 177 00:09:52,320 --> 00:09:55,557 make sure you have the data set that 178 00:09:55,640 --> 00:09:58,597 you're feeding? Like, context. Like it's 179 00:09:58,680 --> 00:10:00,957 kind of like you have a new intelligence 180 00:10:01,040 --> 00:10:03,317 layer, but the intelligence layer is 181 00:10:03,400 --> 00:10:07,157 only as good as the context you give it. 182 00:10:07,240 --> 00:10:12,317 So, people describe it even as context engineering, but that is what firms do. 183 00:10:12,400 --> 00:10:15,997 Right? If you think about what do firms do, it's all about the tacit knowledge 184 00:10:16,080 --> 00:10:19,077 we have by working as people in various 185 00:10:19,160 --> 00:10:23,157 departments and moving paper and information. So, the question is how do 186 00:10:23,240 --> 00:10:25,157 you really have this AI also have that 187 00:10:25,240 --> 00:10:27,357 context? So, these are sort of some of 188 00:10:27,440 --> 00:10:32,557 the new things that have to percolate throughout an organization 189 00:10:32,640 --> 00:10:36,037 to take advantage. In fact, that's why I think you're you're going to see that 190 00:10:36,120 --> 00:10:40,797 challenge of why am I not seeing immediate results in productivity? 191 00:10:40,880 --> 00:10:44,197 Because you have to do the hard work. In fact, that's why it's not going to be at 192 00:10:44,280 --> 00:10:47,517 some, you know, it's going to they're going to be firm-wide differences. 193 00:10:47,600 --> 00:10:49,437 They're going to they could [clears throat] be sector-wide 194 00:10:49,520 --> 00:10:53,197 differences, but it's going to fundamentally be because of the 195 00:10:53,280 --> 00:10:56,397 leadership will in an organization. 196 00:10:56,480 --> 00:11:01,077 Do you see the applications being used across large companies and medium and 197 00:11:01,160 --> 00:11:04,677 small companies or is it still the domain of mostly the large companies at 198 00:11:04,760 --> 00:11:08,157 this moment? >> I think that what you're seeing is it's 199 00:11:08,240 --> 00:11:13,917 easier the because if you have a green sort of 200 00:11:14,000 --> 00:11:16,437 you know, you if you start fresh, it's 201 00:11:16,520 --> 00:11:18,397 easier to adopt these tools and you 202 00:11:18,480 --> 00:11:21,597 construct your organization knowing that 203 00:11:21,680 --> 00:11:25,597 these tools exist. So, I would say >> then? It is a barbell. So, small 204 00:11:25,680 --> 00:11:28,517 companies that are just starting use that platform. 205 00:11:28,600 --> 00:11:31,617 >> 100%. And I think in fact, I would say even for large organizations, there's 206 00:11:31,700 --> 00:11:34,677 [clears throat] a fundamental challenge, right? Because unless and until your 207 00:11:34,760 --> 00:11:38,237 rate of change keeps up with Right. uh 208 00:11:38,320 --> 00:11:42,757 with with what is possible, uh you're you're going to get schooled by someone 209 00:11:42,840 --> 00:11:47,557 small being able to achieve scale because of these tools. So, but I think 210 00:11:47,640 --> 00:11:50,637 scale large organizations have an inherent strength. You have the 211 00:11:50,720 --> 00:11:53,997 relationships, you have the data, you have 212 00:11:54,080 --> 00:11:56,437 you have know-how, but the bottom line 213 00:11:56,520 --> 00:12:01,197 is if you don't translate that with a new production function, 214 00:12:01,280 --> 00:12:04,917 uh then you really will be stuck. And so, therefore, the change management 215 00:12:05,000 --> 00:12:08,237 challenge for large organizations is going to be bigger. 216 00:12:08,320 --> 00:12:12,077 The structural challenge for small organizations of how to overcome scale 217 00:12:12,160 --> 00:12:15,277 issues are going to be harder. So, it's sort of the two sides. In an interesting 218 00:12:15,360 --> 00:12:19,637 way, it's going to be a very competitively intense world 219 00:12:19,720 --> 00:12:21,997 where neither side, like whether you're 220 00:12:22,080 --> 00:12:27,357 a new entrant or an incumbent, can't take it as like I I can just coast. What 221 00:12:27,440 --> 00:12:30,757 about country to country? Are you seeing big 222 00:12:30,840 --> 00:12:32,477 differences in how the applications are 223 00:12:32,560 --> 00:12:35,637 being used? Is it Is AI still the domain 224 00:12:35,720 --> 00:12:40,117 of developed countries or is it becoming rapidly a domain of all countries? 225 00:12:40,200 --> 00:12:45,157 >> I I I'm seeing that two things I'd say, Larry. As I travel around the world, the 226 00:12:45,240 --> 00:12:48,237 quality of 227 00:12:48,320 --> 00:12:50,597 whether it's the know-how, the software 228 00:12:50,680 --> 00:12:54,037 developers, the startups, 229 00:12:54,120 --> 00:12:57,997 or even large or large organizations, it's not that 230 00:12:58,080 --> 00:13:01,117 different. It's fascinating. You can show up in Jakarta, you can show up in 231 00:13:01,200 --> 00:13:05,517 Istanbul, you can show up in Mexico City. It's not that different than 232 00:13:05,600 --> 00:13:08,037 showing up even in say Seattle or San 233 00:13:08,120 --> 00:13:13,317 Francisco, right? It's not for the first time just because access to what's 234 00:13:13,400 --> 00:13:14,557 happening 235 00:13:14,640 --> 00:13:17,637 uh is there. That said, 236 00:13:17,720 --> 00:13:21,277 at [clears throat] scale, the commitment 237 00:13:21,360 --> 00:13:24,277 to using this, the risk capital being 238 00:13:24,360 --> 00:13:26,837 there, the large companies pushing it 239 00:13:26,920 --> 00:13:30,317 hard. I mean, I you know, again, the US, 240 00:13:30,400 --> 00:13:33,757 you know, is In fact, if I compare it, 241 00:13:33,840 --> 00:13:36,157 uh take financial sector. 242 00:13:36,240 --> 00:13:38,277 If in financial sector's adoption of the 243 00:13:38,360 --> 00:13:41,957 cloud versus AI, night and day, right? 244 00:13:42,040 --> 00:13:45,357 Because in an interesting way, it's much faster 245 00:13:45,440 --> 00:13:49,957 uh when it comes to AI versus it was with the cloud. And cloud because for a 246 00:13:50,040 --> 00:13:52,917 variety of reasons, regulatory >> issues, too. 247 00:13:53,000 --> 00:13:56,117 Until the regulators allowed banks to 248 00:13:56,200 --> 00:14:00,117 bring their data off out out of campus, that was a big issue. Yeah. So, I would 249 00:14:00,200 --> 00:14:02,837 say I think [clears throat] wherever, you know, so 250 00:14:02,920 --> 00:14:06,077 in the West, in particular in the US, 251 00:14:06,160 --> 00:14:10,637 uh there is clearly a real I would say more of an energy 252 00:14:10,720 --> 00:14:13,877 around it in terms of going and using it. 253 00:14:13,960 --> 00:14:17,237 Uh but it's sort of a lot more uniformly 254 00:14:17,320 --> 00:14:21,317 spreading around the world than any technology at least I've seen. Are you 255 00:14:21,400 --> 00:14:25,117 Are you you mentioned [clears throat] about power, the grid. 256 00:14:25,200 --> 00:14:28,477 Is that going to be one of the determinants of of [snorts] 257 00:14:28,560 --> 00:14:30,957 the accessibility? If you do not have 258 00:14:31,040 --> 00:14:32,597 cheap power, 259 00:14:32,680 --> 00:14:35,797 it the demand is costly. 100%. So, if 260 00:14:35,880 --> 00:14:40,907 you sort of look at the tokens per dollar per watt, right? Which I think in 261 00:14:40,990 --> 00:14:45,677 [clears throat] some sense, I would claim that GDP growth in any place will 262 00:14:45,760 --> 00:14:49,357 be directly correlated. Like you if you sort of buy my entire argument that 263 00:14:49,440 --> 00:14:54,397 look, we've got a new commodity, it it's tokens. Right. And the job of every 264 00:14:54,480 --> 00:14:57,797 economy and every firm in the economy is to 265 00:14:57,880 --> 00:14:59,877 translate these tokens into economic 266 00:14:59,960 --> 00:15:02,437 growth. Then, if you have a cheaper 267 00:15:02,520 --> 00:15:07,277 commodity, it's better. Uh and so, that's sort of what why this 268 00:15:07,360 --> 00:15:10,877 tokens per dollar per watt. And by the way, there are many, many elements to 269 00:15:10,960 --> 00:15:14,077 this, right? Which is it's not just um 270 00:15:14,160 --> 00:15:18,397 the production side. That's why I think even having the grid is important. 271 00:15:18,480 --> 00:15:20,717 Construction costs, right? So, if you 272 00:15:20,800 --> 00:15:25,837 like if you think about the total TCO, uh everything, it's like the how are you 273 00:15:25,920 --> 00:15:27,677 a cheap producer of energy? Can you 274 00:15:27,760 --> 00:15:30,277 build the data centers? Uh then what's 275 00:15:30,360 --> 00:15:34,797 the cost curve of the silicon in the systems? 276 00:15:34,880 --> 00:15:36,557 And by the way, look at the token 277 00:15:36,640 --> 00:15:39,557 pricing, right? Token pricing basically 278 00:15:39,640 --> 00:15:43,397 drops by you know, a half every 3 279 00:15:43,480 --> 00:15:46,597 months. Uh I mean, this is a So, how So, 280 00:15:46,680 --> 00:15:49,397 that that's why I think you can sort of 281 00:15:49,480 --> 00:15:53,037 really plot how you use the tokens to 282 00:15:53,120 --> 00:15:57,237 create surplus knowing that you have a commodity that's whose prices are just 283 00:15:57,320 --> 00:16:00,717 going to monotonically come down in a pretty fast curve. We're sitting in 284 00:16:00,800 --> 00:16:05,517 Europe uh and there is a real fear 285 00:16:05,600 --> 00:16:09,477 because of the Europe does not have its own power. 286 00:16:09,560 --> 00:16:12,077 It has to import mostly of its power. 287 00:16:12,160 --> 00:16:15,997 Um Do you have any messages for Europe 288 00:16:16,080 --> 00:16:19,437 related to this? Yeah, I mean, I think so, there are two sets of things, right? 289 00:16:19,520 --> 00:16:21,997 One is, you know, here we are in in 290 00:16:22,080 --> 00:16:24,117 Switzerland and I look at 291 00:16:24,200 --> 00:16:26,797 the the pharma or the financial sector, 292 00:16:26,880 --> 00:16:32,077 you know, obviously Swiss they're they do a big job in in in this country 293 00:16:32,160 --> 00:16:34,877 as in in Europe, but they're also international brands and international 294 00:16:34,960 --> 00:16:39,277 operations. So, one thing that whenever I think about Europe is the 295 00:16:39,360 --> 00:16:43,197 Europeans are producing products and services that actually are going 296 00:16:43,280 --> 00:16:47,597 everywhere in the world. And so, therefore, uh European competitiveness 297 00:16:47,680 --> 00:16:51,837 is about the competitiveness of their output globally, not just in Europe. I 298 00:16:51,920 --> 00:16:53,637 think sometimes when you come to Europe, 299 00:16:53,720 --> 00:16:55,517 there's a lot of conversation about just 300 00:16:55,600 --> 00:16:59,357 Europe. Uh but European economy is uh 301 00:16:59,440 --> 00:17:03,357 thrives and has, you know, thrived in the last whatever 200 years, 300 years, 302 00:17:03,440 --> 00:17:07,637 the miracle of the West is fundamentally because of what has happened in Europe 303 00:17:07,720 --> 00:17:11,197 uh is because they were able to produce things that the world needed. And so, I 304 00:17:11,280 --> 00:17:15,437 would say that's number one. And in order to do that, you again I go back to 305 00:17:15,520 --> 00:17:19,157 the human capital here is just fantastic and world-class. 306 00:17:19,240 --> 00:17:21,877 Uh you have to absolutely invest in 307 00:17:21,960 --> 00:17:26,597 producing uh you know, having the energy and the tokens here, which again, you're 308 00:17:26,680 --> 00:17:30,757 attracting like as I said, we are investing and others are investing uh 309 00:17:30,840 --> 00:17:35,317 the data centers here. So, the question is what's that next generation of output 310 00:17:35,400 --> 00:17:38,797 that comes from here, right? I always think about the German Mittelstand. 311 00:17:38,880 --> 00:17:43,237 Whenever I go to a jeweler or a dentist in the United States, I'm surrounded by 312 00:17:43,320 --> 00:17:44,637 German Mittelstand, right? 313 00:17:44,720 --> 00:17:47,237 >> Yes, totally. Um it's just unbelievable 314 00:17:47,320 --> 00:17:50,517 engineering prowess of that country. 315 00:17:50,600 --> 00:17:52,277 Uh and now the question And by the way, 316 00:17:52,360 --> 00:17:54,837 that's the point. They are producing 317 00:17:54,920 --> 00:17:58,077 industrial products, which today are 318 00:17:58,160 --> 00:18:01,357 built in into it all the intelligence as 319 00:18:01,440 --> 00:18:03,917 well. That data, right? So, I know 320 00:18:04,000 --> 00:18:05,757 whenever we come to Europe, everyone's 321 00:18:05,840 --> 00:18:08,397 like talking about sovereignty and data 322 00:18:08,480 --> 00:18:11,237 this, data that. Guess what? Europe 323 00:18:11,320 --> 00:18:14,277 actually should be much more concerned 324 00:18:14,360 --> 00:18:17,117 about access to their industrial 325 00:18:17,200 --> 00:18:19,637 companies, their financial services 326 00:18:19,720 --> 00:18:23,037 companies of data from you US and the 327 00:18:23,120 --> 00:18:27,237 rest of the world. Uh as opposed to just thinking that somehow by protecting 328 00:18:27,320 --> 00:18:28,837 Europe, you're going to be competitive. 329 00:18:28,920 --> 00:18:30,997 You are only going to be competitive if 330 00:18:31,080 --> 00:18:33,557 the products coming out of Europe are 331 00:18:33,640 --> 00:18:36,317 globally competitive. 332 00:18:36,400 --> 00:18:38,757 And so, that's I think what needs to 333 00:18:38,840 --> 00:18:40,877 change. I know Europe has led in 334 00:18:40,960 --> 00:18:43,637 privacy, that's fantastic, has led in 335 00:18:43,720 --> 00:18:48,437 many aspects of even safety around AI and what have you, and that's a feature 336 00:18:48,520 --> 00:18:52,757 that's great. But you also have to complement it with by building locally 337 00:18:52,840 --> 00:18:55,117 and then also thinking globally, what's 338 00:18:55,200 --> 00:18:57,477 the contribution this continent will 339 00:18:57,560 --> 00:19:02,277 make to the rest of the world, which it has historically been a leader in. 340 00:19:02,360 --> 00:19:05,557 >> A leader. So, do you think the whole idea around 341 00:19:05,640 --> 00:19:09,077 sovereignty of data, is that being misunderstood? 342 00:19:09,160 --> 00:19:10,597 I think that 343 00:19:10,680 --> 00:19:13,717 when people talk about sovereignty, 344 00:19:13,800 --> 00:19:17,917 first of all, it's very important, clearly in Who owns the data? And in a 345 00:19:18,000 --> 00:19:20,557 week like this, it's more important. 346 00:19:20,640 --> 00:19:24,357 Uh but that said, uh 347 00:19:24,440 --> 00:19:28,477 it is you have to kind of think about how what 348 00:19:28,560 --> 00:19:32,717 is sovereignty mean? Like for example, in the AI era, the topic that's least 349 00:19:32,800 --> 00:19:35,117 talked about, but I feel will be most 350 00:19:35,200 --> 00:19:38,317 talked about in in this this calendar 351 00:19:38,400 --> 00:19:41,117 year will be the sovereignty of a firm. 352 00:19:41,200 --> 00:19:43,837 Just imagine, if your firm, 353 00:19:43,920 --> 00:19:46,717 you're not able to embed the tacit 354 00:19:46,800 --> 00:19:49,757 knowledge of the firm in a set of 355 00:19:49,840 --> 00:19:53,797 weights in a model that you control, 356 00:19:53,880 --> 00:19:56,397 by definition, you have no sovereignty. 357 00:19:56,480 --> 00:19:58,197 That means you're leaking enterprise 358 00:19:58,280 --> 00:20:01,517 value to some model company somewhere. 359 00:20:01,600 --> 00:20:04,997 In fact, that it's sort of fascinating that nobody's talking about that, right? 360 00:20:05,080 --> 00:20:08,317 It's like everybody's talking about everything else that is sort of, you 361 00:20:08,400 --> 00:20:11,717 know, outside of that. Whereas the most 362 00:20:11,800 --> 00:20:15,437 important thing is, it really doesn't matter if you have In fact, the data 363 00:20:15,520 --> 00:20:19,237 center where it runs is the least important thing, quite honestly. But 364 00:20:19,320 --> 00:20:22,797 like even there, first of all, the data centers all are all over just because 365 00:20:22,880 --> 00:20:27,037 speed of light is a real constraint. Uh and so therefore, the data centers will 366 00:20:27,120 --> 00:20:29,677 be spread. >> Yes. Uh you will have digital, you'll be 367 00:20:29,760 --> 00:20:33,357 able to encrypt everything, you'll be able to have the keys with you. All of 368 00:20:33,440 --> 00:20:35,477 these are much more technically solved 369 00:20:35,560 --> 00:20:38,357 problems. But the one problem that will 370 00:20:38,440 --> 00:20:41,557 only be solved is by you having much 371 00:20:41,640 --> 00:20:46,317 more sovereignty over the, you know, tacit knowledge uh and control over the 372 00:20:46,400 --> 00:20:50,797 models, and it's not a one-way enterprise value transfer. Um and so to 373 00:20:50,880 --> 00:20:53,277 me, I think sovereignty requires real 374 00:20:53,360 --> 00:20:55,797 thought on what is it. 375 00:20:55,880 --> 00:20:58,277 You know, control of destiny means that 376 00:20:58,360 --> 00:21:00,957 you are your ability to produce 377 00:21:01,040 --> 00:21:03,597 something that is unique is preserved. 378 00:21:03,680 --> 00:21:05,637 Uh David Ricardo was not wrong, there's 379 00:21:05,720 --> 00:21:08,557 comparative advantage in countries. Uh 380 00:21:08,640 --> 00:21:11,597 there is comparative advantage in firms. 381 00:21:11,680 --> 00:21:15,837 That needs to be preserved even in the AI era. That's what'll give you real 382 00:21:15,920 --> 00:21:20,357 sovereignty. One last question, I know we're running out of time. 383 00:21:20,440 --> 00:21:24,357 Um in 5 years or 10 years, is there going to be one dominant model that 384 00:21:24,440 --> 00:21:28,757 we're all going to be using, or or and how is Microsoft preparing for this? Are 385 00:21:28,840 --> 00:21:30,637 you going to be Are we going to be using 386 00:21:30,720 --> 00:21:33,597 one model for for enterprise, one model 387 00:21:33,680 --> 00:21:36,677 for other other trades? You know, even 388 00:21:36,760 --> 00:21:38,757 in the last, whatever, four three years, 389 00:21:38,840 --> 00:21:41,677 four years that we've been at it, um the 390 00:21:41,760 --> 00:21:44,837 the reality at this point is it's a 391 00:21:44,920 --> 00:21:50,157 multi-model world, right? I mean, the In fact, if you think about it, 392 00:21:50,240 --> 00:21:52,237 the in 393 00:21:52,320 --> 00:21:54,797 both, there are going to be 394 00:21:54,880 --> 00:21:58,957 multiple models, and the trick is really 395 00:21:59,040 --> 00:22:02,637 how do you take advantage of these multiple models, 396 00:22:02,720 --> 00:22:05,077 and in fact, build your own model by 397 00:22:05,160 --> 00:22:08,077 distilling these, right? Uh so think of 398 00:22:08,160 --> 00:22:11,277 these models uh that you orchestrate to 399 00:22:11,360 --> 00:22:15,637 build your own model, and more importantly, you do what is described as 400 00:22:15,720 --> 00:22:17,637 hard orchestration or harness 401 00:22:17,720 --> 00:22:20,477 engineering. So, the IP of any 402 00:22:20,560 --> 00:22:23,717 application or any firm is how do you 403 00:22:23,800 --> 00:22:28,757 use all these models with context engineering or your data? Yes. Right? 404 00:22:28,840 --> 00:22:32,757 So, it's not three parts. So, can I bring in all the models, by the way, uh 405 00:22:32,840 --> 00:22:35,317 which is close source, open source, 406 00:22:35,400 --> 00:22:38,597 build my own model, orchestrate them, 407 00:22:38,680 --> 00:22:41,957 and feed it my data to change the 408 00:22:42,040 --> 00:22:44,597 trajectory of some outcome that I care 409 00:22:44,680 --> 00:22:49,237 about. That's it. That's the entire picture. So, you can do it in like, oh, 410 00:22:49,320 --> 00:22:53,717 I produce a particular product or service. Uh first, I got to do better 411 00:22:53,800 --> 00:22:58,037 better job in sales, or better job in R&D, or better job in finance, or what 412 00:22:58,120 --> 00:23:02,317 have you, and you take that outcome, and then you say, can I use all the models, 413 00:23:02,400 --> 00:23:03,997 orchestrate them, and feed it my 414 00:23:04,080 --> 00:23:06,997 context, and then in as a result of it, 415 00:23:07,080 --> 00:23:09,757 the reasoning traces are really leading 416 00:23:09,840 --> 00:23:12,597 to some capability and models that I 417 00:23:12,680 --> 00:23:17,477 control as my IP. As long as firms can answer that question, they're going to 418 00:23:17,560 --> 00:23:19,477 be getting ahead. 419 00:23:19,560 --> 00:23:21,517 Ladies and Gentlemen, let's 420 00:23:21,600 --> 00:23:25,557 thank Satya, my friend, Thank you. 421 00:23:25,640 --> 00:23:29,437 for >> [applause] 422 00:23:29,520 --> 00:23:34,317 >> And hopefully, this is the beginning of many great dialogue and conversations here 423 00:23:34,400 --> 00:23:38,840 at the World Economic Forum. Thank you, everyone. Thank you. 424 00:24:15,200 --> 00:24:17,237 It's getting better and better. 425 00:24:17,320 --> 00:24:19,677 And interestingly enough, you look at 426 00:24:19,760 --> 00:24:24,717 it, uh the software developer still is got a lot of agency in it, right? So, 427 00:24:24,800 --> 00:24:28,597 that's why I kind of still think that, you know, going and thinking of these as 428 00:24:28,680 --> 00:24:32,917 somehow living outside of the realm of human agency is probably not the right 429 00:24:33,000 --> 00:24:37,157 way to think about it. In fact, the way to perhaps conceive it like if let's say 430 00:24:37,240 --> 00:24:40,917 in early '80s, if somebody had come to us and said, 431 00:24:41,000 --> 00:24:45,437 whatever, 4 billion people are going to wake up every morning and start typing. 432 00:24:45,520 --> 00:24:50,037 You would've said, why, right? You know, we have a like we have a typist pool 433 00:24:50,120 --> 00:24:53,477 that's good enough, we don't need 4 billion people. But we That's what 434 00:24:53,560 --> 00:24:55,797 happened. Like we invented this entire 435 00:24:55,880 --> 00:24:58,477 class of thing called knowledge work, 436 00:24:58,560 --> 00:25:00,797 where people started really using 437 00:25:00,880 --> 00:25:04,197 computers uh to go amplify what we were 438 00:25:04,280 --> 00:25:07,117 trying to achieve uh using software. I 439 00:25:07,200 --> 00:25:09,117 think in the context of AI, that same 440 00:25:09,200 --> 00:25:11,677 thing is going to happen. Um it's not 441 00:25:11,760 --> 00:25:16,597 like, you know, what is hardcore coding is going to remain hardcore coding 442 00:25:16,680 --> 00:25:20,637 forever. It's just that the levels of abstraction are going to change. Uh but 443 00:25:20,720 --> 00:25:24,997 we also are going to have code as output, just like documents. In fact, 444 00:25:25,080 --> 00:25:29,677 one of Bill's things at Microsoft from the day I joined in '92 always was, 445 00:25:29,760 --> 00:25:31,117 what's the real difference between a 446 00:25:31,200 --> 00:25:34,317 document, a website, and an application, 447 00:25:34,400 --> 00:25:37,037 right? It's the lack of sort of software 448 00:25:37,120 --> 00:25:41,597 that can transform itself. Interestingly enough, AI finally gives us that, right? 449 00:25:41,680 --> 00:25:45,437 Which is I can write a document, I can just say, no, I don't want it as a 450 00:25:45,520 --> 00:25:49,007 document, I want it as a website, it'll just transform that document using 451 00:25:49,090 --> 00:25:52,317 [clears throat] code into a website. I say, oh, I don't like the website, I 452 00:25:52,400 --> 00:25:54,117 want an app, it'll write more code to 453 00:25:54,200 --> 00:25:57,317 transform it. So, that a reasoning and 454 00:25:57,400 --> 00:25:59,477 capab- reasoning capabilities, that 455 00:25:59,560 --> 00:26:01,717 prediction capabilities, that ability to 456 00:26:01,800 --> 00:26:04,957 take action, remain long-term coherent, 457 00:26:05,040 --> 00:26:08,957 is all improving. Um and our job, 458 00:26:09,040 --> 00:26:12,317 though, is to parlay this. Like take even what 459 00:26:12,400 --> 00:26:15,397 you at BlackRock are doing, right? When you're 460 00:26:15,480 --> 00:26:18,957 taking something like, say, Co-pilot plus Aladdin and bringing those things 461 00:26:19,040 --> 00:26:21,877 together [clears throat] Right. to improve the productivity in 462 00:26:21,960 --> 00:26:26,397 the firm for the decisions you want to make, right? With your data. I could 463 00:26:26,480 --> 00:26:31,557 just tell you from at our firm, things that would take 12 hours to compute now 464 00:26:31,640 --> 00:26:35,597 takes minutes. For us processing 14 trillion dollars of 465 00:26:35,680 --> 00:26:40,957 other people's money with hundreds of thousands of different um mandates, um 466 00:26:41,040 --> 00:26:44,157 we could do that instantaneously. And we, you know, that To me, if it wasn't 467 00:26:44,240 --> 00:26:47,997 for the technology and AI today, we would not be able to function to the 468 00:26:48,080 --> 00:26:49,437 scale that we're operating. That's 469 00:26:49,520 --> 00:26:52,797 right. And so, to me, that one firm at a 470 00:26:52,880 --> 00:26:55,277 time, one country at a time, if we can 471 00:26:55,360 --> 00:27:00,557 really take these tokens and bend the curve of productivity, then there is 472 00:27:00,640 --> 00:27:03,557 surplus everywhere, and that's really the goal. 473 00:27:03,640 --> 00:27:07,277 Well, surplus could be scary, too. Does Does surplus mean 474 00:27:07,360 --> 00:27:11,517 fewer workers? What do we mean by surplus? And so, you know, the I'm going 475 00:27:11,600 --> 00:27:15,037 to tie that into my second question about AI diffusion. 476 00:27:15,120 --> 00:27:18,637 >> Yeah. To me, the the whole realization of AI 477 00:27:18,720 --> 00:27:20,717 for any society, 478 00:27:20,800 --> 00:27:23,877 and also for a more balanced world, is 479 00:27:23,960 --> 00:27:28,197 making sure that it's diffused and accessible and available across the 480 00:27:28,280 --> 00:27:30,237 world. So, what, you know, can you 481 00:27:30,320 --> 00:27:33,957 describe how this process of diffusion 482 00:27:34,040 --> 00:27:36,277 across economies, across companies, 483 00:27:36,360 --> 00:27:38,117 across people and countries, how does 484 00:27:38,200 --> 00:27:43,557 that play out? Yeah, I think that this is the real question, right? 485 00:27:43,640 --> 00:27:44,877 Because one of the 486 00:27:44,960 --> 00:27:47,197 things right now, the zeitgeist is a 487 00:27:47,280 --> 00:27:50,597 little bit about the admiration for AI 488 00:27:50,680 --> 00:27:56,077 in its abstract form, or as as as technology. Uh 489 00:27:56,160 --> 00:28:00,077 >> [clears throat] >> but I think we as even a global community um 490 00:28:00,160 --> 00:28:02,397 have to get to a point where we're using 491 00:28:02,480 --> 00:28:05,997 this to do something useful uh that 492 00:28:06,080 --> 00:28:10,197 changes uh the outcomes of people and communities and countries and 493 00:28:10,280 --> 00:28:14,357 industries, right? Otherwise, I don't think uh this makes much sense, right? 494 00:28:14,440 --> 00:28:16,517 In fact, I would say we would quickly 495 00:28:16,600 --> 00:28:20,317 lose even the social permission uh to 496 00:28:20,400 --> 00:28:24,877 actually take something like uh energy, which is a scarce resource, and use it 497 00:28:24,960 --> 00:28:28,797 to generate these tokens if these tokens are not improving health outcomes, 498 00:28:28,880 --> 00:28:32,757 education outcomes, public sector efficiency, private sector 499 00:28:32,840 --> 00:28:34,797 competitiveness across all sectors, 500 00:28:34,880 --> 00:28:37,357 small and large, right? And that to me 501 00:28:37,440 --> 00:28:40,237 is ultimately the goal. So, therefore, I 502 00:28:40,320 --> 00:28:43,477 think really diffusion is everything. 503 00:28:43,560 --> 00:28:47,837 And so, the way it happens is, let's sort of unpack this. 504 00:28:47,920 --> 00:28:50,357 Uh on the supply side, what needs to 505 00:28:50,440 --> 00:28:53,397 happen in each country is the tokens per 506 00:28:53,480 --> 00:28:57,837 dollar per watt have to sort of monotonically get more 507 00:28:57,920 --> 00:29:02,037 efficient and better, right? So, to some degree, even what we are trying to do 508 00:29:02,120 --> 00:29:06,837 with investments the two firms are doing around the world is to just say that 509 00:29:06,920 --> 00:29:08,877 like let's make sure that the supply is 510 00:29:08,960 --> 00:29:11,317 there, which is uh everything from the 511 00:29:11,400 --> 00:29:16,557 chips on down ultimately to these token factories that get deployed everywhere. 512 00:29:16,640 --> 00:29:20,517 By the way, there's not one token factory. This token factory is the first 513 00:29:20,600 --> 00:29:23,757 thing that's going to be diffused all around the world. It's just like 514 00:29:23,840 --> 00:29:26,597 electricity, right? You just need a 515 00:29:26,680 --> 00:29:31,717 ubiquitous grid of uh energy and tokens 516 00:29:31,800 --> 00:29:35,917 that then will power the rest of the economy, right? So, that's I think one 517 00:29:36,000 --> 00:29:39,717 side of it. Then the demand side of this is a little 518 00:29:39,800 --> 00:29:42,877 bit like every firm has to start by 519 00:29:42,960 --> 00:29:46,837 using it. If I look back even, you know, when the 520 00:29:46,920 --> 00:29:51,237 PCs first came out or the personal computing era started, I loved, you 521 00:29:51,320 --> 00:29:53,197 know, I think Jobs had a nice metaphor. 522 00:29:53,280 --> 00:29:55,677 He called it the bicycle for the mind. 523 00:29:55,760 --> 00:29:59,717 Uh Bill had a metaphor which I remember was like information at your fingertips, 524 00:29:59,800 --> 00:30:03,037 right? These two metaphors were great, like which you know, allowed us to say, 525 00:30:03,120 --> 00:30:06,677 "That's what it is. It's a tool that I will use to get information at my 526 00:30:06,760 --> 00:30:08,677 fingertips. I'll use it as a cognitive 527 00:30:08,760 --> 00:30:12,157 amplifier." Now, I think that that's 528 00:30:12,240 --> 00:30:15,837 what we have, but you know, 10x, 100x, right? So, in some 529 00:30:15,920 --> 00:30:18,557 sense, you as a as every knowledge 530 00:30:18,640 --> 00:30:21,797 worker, you now have access to infinite 531 00:30:21,880 --> 00:30:26,157 minds. That's the way I think about it, right? So, there's a Turing Award winner 532 00:30:26,240 --> 00:30:31,117 uh called Raj Reddy who had this nice metaphor of AI and he had this long before 533 00:30:31,200 --> 00:30:35,157 uh even generative AI. He said he either either it's a cognitive amplifier or 534 00:30:35,240 --> 00:30:37,397 it's a guardian angel, right? So, if you 535 00:30:37,480 --> 00:30:39,997 think of AI as that, 536 00:30:40,080 --> 00:30:43,237 uh then in the global workforce, right? 537 00:30:43,320 --> 00:30:47,397 When a doctor can get to a patient, spend more time 538 00:30:47,480 --> 00:30:49,557 with the patient because the AI is doing 539 00:30:49,640 --> 00:30:54,597 the transcription and entering the records in the EMR system, entering the 540 00:30:54,680 --> 00:30:57,557 right billing code so that the healthcare 541 00:30:57,640 --> 00:30:59,797 uh you know, industry is better served 542 00:30:59,880 --> 00:31:02,717 across the payer, the provider, and uh 543 00:31:02,800 --> 00:31:04,797 and the patient ultimately, right? 544 00:31:04,880 --> 00:31:07,397 That's an an outcome that I think all of 545 00:31:07,480 --> 00:31:12,317 us can benefit from. So, I feel ultimately, it's going to require real 546 00:31:12,400 --> 00:31:14,277 leadership on the private sector and the 547 00:31:14,360 --> 00:31:17,597 public sector to ensure the diffusion 548 00:31:17,680 --> 00:31:21,797 happens. And the one thing I thing I'll mention, Larry, is skilling. 549 00:31:21,880 --> 00:31:23,877 Right? So, in some sense, 550 00:31:23,960 --> 00:31:27,437 the thing that diffusion is very 551 00:31:27,520 --> 00:31:30,557 strongly correlated to one thing alone, 552 00:31:30,640 --> 00:31:33,757 which is how broadly are people skilled 553 00:31:33,840 --> 00:31:37,397 in using this. Um and interestingly 554 00:31:37,480 --> 00:31:41,157 enough, I think if mobile has taught us one thing 555 00:31:41,240 --> 00:31:45,637 is it it's actually distinct from what happened in the PC, right? 556 00:31:45,720 --> 00:31:50,117 Uh I remember even growing up in the global south, uh there used to be a real 557 00:31:50,200 --> 00:31:52,597 relationship between learning Excel 558 00:31:52,680 --> 00:31:56,197 skills or Word skills and getting a job. 559 00:31:56,280 --> 00:31:59,117 Um you know, right now, um 560 00:31:59,200 --> 00:32:03,237 what's the model in in mobile? It's kind of created the same opportunity, but 561 00:32:03,320 --> 00:32:05,437 it's been a lot more consumption-led. 562 00:32:05,520 --> 00:32:09,997 It's these creator economy and what have you. But it's not been about sort of oh 563 00:32:10,080 --> 00:32:13,677 wow, here's how you get a healthcare job or here is how you get a finance job or 564 00:32:13,760 --> 00:32:18,277 here is how you get you know, you get ahead uh professionally. 565 00:32:18,360 --> 00:32:21,277 Um and that needs to come back, right? 566 00:32:21,360 --> 00:32:22,797 People need to say, "Oh, I pick up this 567 00:32:22,880 --> 00:32:25,837 AI skill and now I'm a better provider 568 00:32:25,920 --> 00:32:30,537 of some product or service in the real economy." So, it's it's very easy 569 00:32:30,620 --> 00:32:34,117 [clears throat] to see how mobile and the diffusion of mobile, how it 570 00:32:34,200 --> 00:32:37,997 transformed economies, especially in the global south. 571 00:32:38,080 --> 00:32:42,277 Um how does how does this, you know, to me, I I I just read a research report 572 00:32:42,360 --> 00:32:46,957 that said um the applications for AI so far are 573 00:32:47,040 --> 00:32:51,637 heavily weighted toward those who are educated or educated economies. 574 00:32:51,720 --> 00:32:55,597 And so, does that create that, you know, more of a bifurcation and a more 575 00:32:55,680 --> 00:32:58,277 polarization? How do we ensure that that 576 00:32:58,360 --> 00:33:02,627 that diffusion is spread evenly? How do we make sure that we're not 577 00:33:02,710 --> 00:33:07,597 [clears throat] leaving major portions of society or the world behind? 578 00:33:07,680 --> 00:33:10,357 Because I think that's that's going [clears throat] to be the big issue for 579 00:33:10,440 --> 00:33:13,597 us going forward. Yeah, so it's it's it's interesting, right? This is one of 580 00:33:13,680 --> 00:33:16,197 those times when 581 00:33:16,280 --> 00:33:20,077 uh by definition and because of the 582 00:33:20,160 --> 00:33:23,837 rails that have been established, you know, I you said it, right? Which is 583 00:33:23,920 --> 00:33:29,277 uh what's happened with mobile as well as what's happened with the 584 00:33:29,360 --> 00:33:31,517 uh bro you know, essentially connectivity. 585 00:33:31,600 --> 00:33:35,237 >> Right. You have the ability [clears throat] to sort of deliver the 586 00:33:35,320 --> 00:33:37,677 tokens pretty evenly around the world. 587 00:33:37,760 --> 00:33:42,837 Right. Uh a lot more so than, let's say, uh the PC era or even the beginning of 588 00:33:42,920 --> 00:33:44,437 the mobile era, right? Because it took a 589 00:33:44,520 --> 00:33:46,837 long time for even the smartphone in 590 00:33:46,920 --> 00:33:50,077 particular to penetrate um all of the 591 00:33:50,160 --> 00:33:52,757 world. Whereas now, it's not the case, 592 00:33:52,840 --> 00:33:55,797 right? These models and their outputs 593 00:33:55,880 --> 00:33:57,957 are pretty much available everywhere. 594 00:33:58,040 --> 00:34:01,237 And so, the question to me is 595 00:34:01,320 --> 00:34:05,237 what's the use cases that make sense, right? So, what In fact, one of the 596 00:34:05,320 --> 00:34:07,397 demos I always go back to, I think this 597 00:34:07,480 --> 00:34:09,877 is even in the beginning of '23, was a 598 00:34:09,960 --> 00:34:12,717 rural Indian farmer was able to use a 599 00:34:12,800 --> 00:34:15,237 bot built on, I think, a very early 600 00:34:15,320 --> 00:34:18,117 GPT-3 or 2.5 even, 601 00:34:18,200 --> 00:34:20,677 uh essentially to reason over some farm 602 00:34:20,760 --> 00:34:25,677 subsidies that he had heard about in a local language and had it even in that 603 00:34:25,760 --> 00:34:29,597 very early days uh have it even show some agentic 604 00:34:29,680 --> 00:34:31,557 behavior, right? Like go complete a form 605 00:34:31,640 --> 00:34:33,557 for me. So, in some sense, it took uh 606 00:34:33,640 --> 00:34:36,917 you know, it brought back agency to 607 00:34:37,000 --> 00:34:40,237 someone uh who perhaps didn't have that 608 00:34:40,320 --> 00:34:41,997 because the technology was so much more 609 00:34:42,080 --> 00:34:44,957 accessible. So, I I do think it's in the 610 00:34:45,040 --> 00:34:48,397 hands even in the global south to use it 611 00:34:48,480 --> 00:34:51,357 uh to create, I would say, more of that 612 00:34:51,440 --> 00:34:56,597 opportunity where there isn't one. Um but I think what the necessary 613 00:34:56,680 --> 00:35:00,157 [clears throat] conditions still are, do you have 614 00:35:00,240 --> 00:35:03,037 uh the capital investment being put in? 615 00:35:03,120 --> 00:35:05,917 Uh do you even have an environment for capital? And because in an interesting 616 00:35:06,000 --> 00:35:09,637 way, we are, for example, as hyperscalers investing all over, right? 617 00:35:09,720 --> 00:35:13,357 Including the global south. Uh so, as long as there's an environment 618 00:35:13,440 --> 00:35:16,277 which attracts the capital investment And you see the demand. 619 00:35:16,360 --> 00:35:18,517 >> And you see And then yeah, the demand is there. 620 00:35:18,600 --> 00:35:21,117 >> Yeah. Um and so, the question is how do 621 00:35:21,200 --> 00:35:25,917 you have a set of policies that allow for both the capital to come in, for it 622 00:35:26,000 --> 00:35:29,357 to find nexus with There are certain things, by the way, private capital can 623 00:35:29,440 --> 00:35:31,517 do, certain things that public capital 624 00:35:31,600 --> 00:35:34,557 only can do. For example, the grid. 625 00:35:34,640 --> 00:35:38,637 Right? It's not I mean, grid in may most countries is sort of fundamentally 626 00:35:38,720 --> 00:35:40,397 driven by governments. >> And public. 627 00:35:40,480 --> 00:35:43,037 >> And public. And so, if you don't have a 628 00:35:43,120 --> 00:35:44,917 sophisticated or rather, if you don't 629 00:35:45,000 --> 00:35:47,917 have a real approach to modernizing the 630 00:35:48,000 --> 00:35:52,077 grid, uh that will hold things back. I mean, there's a lot of talk about behind 631 00:35:52,160 --> 00:35:55,517 the meter and so on. And yes, there's some amount of that we can do ourselves. 632 00:35:55,600 --> 00:35:57,557 >> do that in the US. Many countries can't. 633 00:35:57,640 --> 00:35:59,797 Exact And it's not long-term scalable, 634 00:35:59,880 --> 00:36:01,557 right? I mean, like to me, a long-term 635 00:36:01,640 --> 00:36:05,117 scalable solution is to have uh you 636 00:36:05,200 --> 00:36:08,077 know, all of these token factories part 637 00:36:08,160 --> 00:36:12,197 of the real economy, connected to the grid, connected to the telco network, 638 00:36:12,280 --> 00:36:15,557 delivering uh just like we delivered bits, you have 639 00:36:15,640 --> 00:36:20,477 to deliver tokens plus bits. Um and that's kind of what's going to drive an 640 00:36:20,560 --> 00:36:25,317 at scale, whether it's in the global south or in uh on the developed world. 641 00:36:25,400 --> 00:36:28,717 So, so many people talk about there may be an AI bubble. I mean, the most 642 00:36:28,800 --> 00:36:30,037 important thing that we see as an 643 00:36:30,120 --> 00:36:34,117 investor is the the the democratization 644 00:36:34,200 --> 00:36:37,157 of technology and the diffusion of that 645 00:36:37,240 --> 00:36:41,797 technology really does then transform the demand. And the 646 00:36:41,880 --> 00:36:45,797 companies or the countries that diffuse it fastest are going to be the ultimate 647 00:36:45,880 --> 00:36:48,197 winners, not the technology creator. 648 00:36:48,280 --> 00:36:51,197 That's that's you know, it's it's it's 649 00:36:51,280 --> 00:36:54,837 uh uh this for this not to be a bubble, 650 00:36:54,920 --> 00:36:58,357 Yep. by definition, it requires 651 00:36:58,440 --> 00:37:01,517 that the benefits of this are much more 652 00:37:01,600 --> 00:37:03,117 evenly spread. I mean, I I think a 653 00:37:03,200 --> 00:37:07,117 telltale sign of if it's a bubble would 654 00:37:07,200 --> 00:37:09,957 be if all we're talking about are the 655 00:37:10,040 --> 00:37:14,837 tech firms. Uh right? If uh all we talk about is 656 00:37:14,920 --> 00:37:18,957 what's happening to the technology side, that then that's by you know, it's just 657 00:37:19,040 --> 00:37:22,877 purely supply side. >> Right. Uh ultimately, if we are not 658 00:37:22,960 --> 00:37:26,877 talking about wow, here is a drug comp you know, drug that was sort of uh 659 00:37:26,960 --> 00:37:30,517 brought into the market that's super successful because it was uh AI 660 00:37:30,600 --> 00:37:34,557 accelerated the clinical trial. It's not even the magical molecule, right? It's 661 00:37:34,640 --> 00:37:37,317 kind of even the rest of what is uh 662 00:37:37,400 --> 00:37:39,797 needed in order to make something much 663 00:37:39,880 --> 00:37:45,037 more relevant, right? Um and so, the more we have [clears throat] 664 00:37:45,120 --> 00:37:47,917 uh And by the way, it's happening, right? So, that I'm not sort of saying 665 00:37:48,000 --> 00:37:50,717 that that's why I'm much more confident 666 00:37:50,800 --> 00:37:55,637 that this is a technology that will in fact build on the rails of cloud and 667 00:37:55,720 --> 00:37:59,997 mobile, diffuse faster, >> Yep. and bend the productivity 668 00:38:00,080 --> 00:38:04,117 [clears throat] curve, and bring local surplus and 669 00:38:04,200 --> 00:38:08,797 economic growth all around the world, not just economic growth driven by 670 00:38:08,880 --> 00:38:11,117 capital expenses. 671 00:38:11,200 --> 00:38:15,877 Uh right? Because that's it's a narrow point in time uh calculation. 672 00:38:15,960 --> 00:38:17,437 >> Right now, that's what we're seeing more of. 673 00:38:17,520 --> 00:38:21,237 >> we're seeing in you know, in in in in in in the developed world in particular. 674 00:38:21,320 --> 00:38:25,637 Uh but remember, my capital, like that the one thing that, you know, is 675 00:38:25,720 --> 00:38:29,597 definitely we're spending a lot of it in the United States, but 50% of it is also 676 00:38:29,680 --> 00:38:33,637 all over the world. >> Right. Um and so, interestingly enough, 677 00:38:33,720 --> 00:38:37,037 it depends on uh demand all over the world. And the 678 00:38:37,120 --> 00:38:41,077 demand all over the world will only be there if there is local surplus all over 679 00:38:41,160 --> 00:38:42,917 the world. And so, that's sort of the 680 00:38:43,000 --> 00:38:45,317 way I see the equation. So, let's drill 681 00:38:45,400 --> 00:38:48,837 down a little more. As AI diffuses, 682 00:38:48,920 --> 00:38:51,077 obviously, organizations, companies, 683 00:38:51,160 --> 00:38:53,517 governments are going to have to evolve. 684 00:38:53,600 --> 00:38:56,277 Now getting to the demand side. So, 685 00:38:56,360 --> 00:38:58,157 how do you think the structure of 686 00:38:58,240 --> 00:39:02,237 organizations is changing in an AI world 687 00:39:02,320 --> 00:39:05,517 across roles, across teams, management? 688 00:39:05,600 --> 00:39:10,397 I'm I'm sure um Microsoft has evolved itself. 689 00:39:10,480 --> 00:39:12,197 So, what I probably be good to tell the 690 00:39:12,280 --> 00:39:14,917 audience how do you see this diffusion 691 00:39:15,000 --> 00:39:19,477 occur and the utilization at the corporate level or at a government level 692 00:39:19,560 --> 00:39:23,117 that which ultimately then creates that demand which eliminates any fears of 693 00:39:23,200 --> 00:39:26,837 bubbles. Yeah, now I think it it's probably one of the the the big 694 00:39:26,920 --> 00:39:28,357 challenges with all of these new 695 00:39:28,440 --> 00:39:31,117 technologies is when uh 696 00:39:31,200 --> 00:39:36,157 work work artifact and workflow changes, 697 00:39:36,240 --> 00:39:40,437 uh that means we as firms have to change 698 00:39:40,520 --> 00:39:44,037 how we work. In fact, I remember meeting um 699 00:39:44,120 --> 00:39:48,317 uh the CEO of Generali, you know, a few years back and he was describing he had 700 00:39:48,400 --> 00:39:52,657 joined um the firm, you know, pre-PC era 701 00:39:52,740 --> 00:39:56,917 [clears throat] and uh and he was describing how, for example, they worked 702 00:39:57,000 --> 00:39:59,557 with their agents in the field with 703 00:39:59,640 --> 00:40:03,037 faxes, interoffice memos and 704 00:40:03,120 --> 00:40:07,077 and and suddenly the PC showed up and people would then put a spreadsheet in 705 00:40:07,160 --> 00:40:11,197 an email and send it around and the entire workflow and the work process 706 00:40:11,280 --> 00:40:13,957 changed. Right? So, similarly, I think 707 00:40:14,040 --> 00:40:17,437 with AI, you are going to start seeing 708 00:40:17,520 --> 00:40:20,317 actual change in how workflow happens. 709 00:40:20,400 --> 00:40:23,917 Right? I mean, even >> [clears throat] >> in fact, for me coming to Davos, you 710 00:40:24,000 --> 00:40:28,957 know, whatever 50 bilateral meetings I have, preparing for those 711 00:40:29,040 --> 00:40:33,637 had a particular workflow, right? Which is there is to be my field team would 712 00:40:33,720 --> 00:40:37,997 prepare notes and that would come to my HQ and that would get further refined 713 00:40:38,080 --> 00:40:40,117 and nothing had really changed right? 714 00:40:40,200 --> 00:40:45,437 Since I joined in '92 to essentially even a few years back. Whereas now, I 715 00:40:45,520 --> 00:40:49,197 just go to Copilot and say, "Hey, I'm meeting Larry, please give me a brief." 716 00:40:49,280 --> 00:40:53,437 And it comes back and gives me By the way, the one nice thing is it gives me a 717 00:40:53,520 --> 00:40:57,837 360, right? It knows what we are doing with you as a client, what we are doing 718 00:40:57,920 --> 00:41:00,637 as a client of yours and everything in between [clears throat] as an 719 00:41:00,720 --> 00:41:05,477 investment. It's so it captures even information unlike anything else. In 720 00:41:05,560 --> 00:41:09,997 fact, what I do is I take that and immediately share that back with all of 721 00:41:10,080 --> 00:41:11,957 my colleagues across all the functions. 722 00:41:12,040 --> 00:41:16,837 Right? Think about it. It's a complete inversion of how information is flowing 723 00:41:16,920 --> 00:41:18,877 in the organization. It's not like this 724 00:41:18,960 --> 00:41:21,677 classic we have an organization, we have 725 00:41:21,760 --> 00:41:24,157 departments, we have the specializations 726 00:41:24,240 --> 00:41:26,477 and the information trickles up. No, no, 727 00:41:26,560 --> 00:41:29,997 no, it actually it flattens the entire 728 00:41:30,080 --> 00:41:33,037 information flow. So, once you start 729 00:41:33,120 --> 00:41:35,237 having that, you have to redesign 730 00:41:35,320 --> 00:41:39,117 structurally. So, the current structure 731 00:41:39,200 --> 00:41:42,797 may not make sense. Um because you want 732 00:41:42,880 --> 00:41:46,677 people to be able to work in a way that 733 00:41:46,760 --> 00:41:48,917 allows them to have this information 734 00:41:49,000 --> 00:41:52,077 flow freely. So, what all this leads me 735 00:41:52,160 --> 00:41:53,717 to if I had to sort of say what's the 736 00:41:53,800 --> 00:41:55,997 formula, the formula I think it starts 737 00:41:56,080 --> 00:41:59,117 with the mindset. So, the mindset we as 738 00:41:59,200 --> 00:42:02,357 leaders should have is we need to think 739 00:42:02,440 --> 00:42:06,037 about changing the work the workflow 740 00:42:06,120 --> 00:42:10,277 with the technology. Then that needs skill set. So, you can't sort of talk 741 00:42:10,360 --> 00:42:14,517 about this in the abstract. You got to use it. Like so, if I'm not using the 742 00:42:14,600 --> 00:42:17,837 You have to trust it. You have to trust it, you have to use it, you have to 743 00:42:17,920 --> 00:42:20,797 learn even how to put the guardrails to 744 00:42:20,880 --> 00:42:25,037 trust it, right? So, you can't again you can't just be afraid of it. 745 00:42:25,120 --> 00:42:27,237 It's going to It's going to be diffused. 746 00:42:27,320 --> 00:42:29,677 So, the question is as a firm, you have 747 00:42:29,760 --> 00:42:34,837 to use it to learn how to even put the guardrails that allow you to be 748 00:42:34,920 --> 00:42:39,677 able to trust it. So, skills, so mindset, skills. The 749 00:42:39,760 --> 00:42:41,757 other big consideration 750 00:42:41,840 --> 00:42:45,557 really is how do you make sure you have 751 00:42:45,640 --> 00:42:47,797 the data set that you're feeding? Like 752 00:42:47,880 --> 00:42:51,037 context. Like it's kind of like you have 753 00:42:51,120 --> 00:42:53,517 a new intelligence layer. But the 754 00:42:53,600 --> 00:42:56,637 intelligence layer is only as good as 755 00:42:56,720 --> 00:42:59,157 the context you give it. So, people 756 00:42:59,240 --> 00:43:01,277 describe it even as context engineering. 757 00:43:01,360 --> 00:43:04,117 But that is what firms do. Right? If you 758 00:43:04,200 --> 00:43:05,837 think about what do firms do, it's all 759 00:43:05,920 --> 00:43:08,557 about the tacit knowledge we have by 760 00:43:08,640 --> 00:43:13,717 working as people in various departments and moving paper and information. So, 761 00:43:13,800 --> 00:43:17,637 the question is how do you really have this AI also have that context? So, 762 00:43:17,720 --> 00:43:20,157 these are sort of some of the new things 763 00:43:20,240 --> 00:43:25,237 that have to percolate throughout an organization to take advantage. In fact, 764 00:43:25,320 --> 00:43:29,517 that's why I think you're you're going to see that challenge of why am I not 765 00:43:29,600 --> 00:43:33,197 seeing immediate results in productivity? Because you have to do the 766 00:43:33,280 --> 00:43:36,477 hard work. In fact, that's why it's not going to be at some, you know, it's 767 00:43:36,560 --> 00:43:39,797 going to There're going to be firm-wide differences. There are going to There 768 00:43:39,880 --> 00:43:42,717 could be [clears throat] sector-wide differences, but it's going to 769 00:43:42,800 --> 00:43:47,877 fundamentally be because of the leadership will in an organization. 770 00:43:47,960 --> 00:43:52,557 Do you see the applications being used across large companies and medium and 771 00:43:52,640 --> 00:43:56,157 small companies or is it still the domain of mostly the large companies at 772 00:43:56,240 --> 00:43:59,637 this moment? >> I think that what you're seeing is it's 773 00:43:59,720 --> 00:44:05,397 easier the because if you have a green sort of 774 00:44:05,480 --> 00:44:07,917 you know, you if you start fresh it's 775 00:44:08,000 --> 00:44:10,117 easier to adopt these tools and you 776 00:44:10,200 --> 00:44:15,077 construct your organization knowing that these tools exist. So, is it a barbell 777 00:44:15,160 --> 00:44:19,997 then? It is a barbell. So, small companies that are just starting use that platform. 778 00:44:20,080 --> 00:44:23,177 >> 100% and I think in fact, I would say even for large organizations, there's 779 00:44:23,260 --> 00:44:26,157 [clears throat] a fundamental challenge, right? Because unless and until your 780 00:44:26,240 --> 00:44:30,037 rate of change keeps up with Right. what 781 00:44:30,120 --> 00:44:34,877 with what is possible, you're you're going to get schooled by someone small 782 00:44:34,960 --> 00:44:40,077 being able to achieve scale because of these tools. So, but I think scale Large 783 00:44:40,160 --> 00:44:41,757 organizations have an inherent strength. 784 00:44:41,840 --> 00:44:45,477 You have the relationships, you have the data, you have 785 00:44:45,560 --> 00:44:47,917 you have know-how, but the bottom line 786 00:44:48,000 --> 00:44:50,397 is if you don't translate that with a 787 00:44:50,480 --> 00:44:55,277 new production function then you really will be stuck. And so, therefore, the 788 00:44:55,360 --> 00:44:59,757 change management challenge for large organizations is going to be bigger. 789 00:44:59,840 --> 00:45:03,557 The structural challenge for small organizations of how to overcome scale 790 00:45:03,640 --> 00:45:06,797 issues are going to be harder. So, it's sort of the two sides. In an interesting 791 00:45:06,880 --> 00:45:11,717 way, it's going to be a very competitively intense world where 792 00:45:11,800 --> 00:45:13,797 neither side, like whether you're a new 793 00:45:13,880 --> 00:45:15,957 entrant or an incumbent, can't take it 794 00:45:16,040 --> 00:45:20,397 as like I I can just coast. What about 795 00:45:20,480 --> 00:45:23,957 country to country? Are you seeing big differences in how the applications are 796 00:45:24,040 --> 00:45:27,957 being used? Is it Is AI still the domain of developed 797 00:45:28,040 --> 00:45:31,597 countries or is it becoming rapidly a domain of all countries? 798 00:45:31,680 --> 00:45:35,717 >> I I'm seeing There are two things I'd say, Larry. As I travel around the 799 00:45:35,800 --> 00:45:39,797 world, the quality of 800 00:45:39,880 --> 00:45:42,077 whether it's the know-how, the software 801 00:45:42,160 --> 00:45:46,517 developers, the startups or even large 802 00:45:46,600 --> 00:45:50,797 or large organizations, it's not that different. It's fascinating. You can 803 00:45:50,880 --> 00:45:54,477 show up in Jakarta, you can show up in Istanbul, you can show up in Mexico 804 00:45:54,560 --> 00:45:57,037 City. It's not that different than 805 00:45:57,120 --> 00:45:59,517 showing up even in say Seattle or San 806 00:45:59,600 --> 00:46:02,037 Francisco, right? It's not for the first 807 00:46:02,120 --> 00:46:04,797 time just because access to what's 808 00:46:04,880 --> 00:46:09,157 happening is there. That said 809 00:46:09,240 --> 00:46:12,757 at [clears throat] scale, the commitment 810 00:46:12,840 --> 00:46:15,797 to using this, the risk capital being 811 00:46:15,880 --> 00:46:18,357 there, the large companies pushing it 812 00:46:18,440 --> 00:46:21,797 hard. I mean, I you know, again, the US, 813 00:46:21,880 --> 00:46:25,237 you know, is In fact, if I compare it 814 00:46:25,320 --> 00:46:27,637 take financial sector. 815 00:46:27,720 --> 00:46:29,797 If in financial sector's adoption of the 816 00:46:29,880 --> 00:46:33,477 cloud versus AI night and day, right? 817 00:46:33,560 --> 00:46:36,837 Because in an interesting way, it's much faster 818 00:46:36,920 --> 00:46:41,277 when it comes to AI versus it was with the cloud. And cloud because of a 819 00:46:41,360 --> 00:46:44,477 variety of reasons >> And regulatory issues, too. 820 00:46:44,560 --> 00:46:47,597 Until the regulators allowed banks to 821 00:46:47,680 --> 00:46:51,637 bring their data off out out of campus, that was a big issue. Yeah. So, I would 822 00:46:51,720 --> 00:46:54,357 say I think wherever, [clears throat] you know, so 823 00:46:54,440 --> 00:46:57,557 in the West, in particular in the US, 824 00:46:57,640 --> 00:47:02,077 there is clearly a real, I'd say, more of an energy 825 00:47:02,160 --> 00:47:04,357 around it in terms of going and using 826 00:47:04,440 --> 00:47:07,317 it. But it's sort of a lot more 827 00:47:07,400 --> 00:47:12,237 uniformly spreading around the world than any technology at least I've seen. 828 00:47:12,320 --> 00:47:16,637 Are you >> [clears throat] >> Are you You mentioned about power, the grid. 829 00:47:16,720 --> 00:47:19,997 Is that going to be one of the determinants of of 830 00:47:20,080 --> 00:47:22,437 the accessibility? If you do not have 831 00:47:22,520 --> 00:47:24,077 cheap power 832 00:47:24,160 --> 00:47:27,397 if the demand is costly 100%. So, if you 833 00:47:27,480 --> 00:47:32,477 sort of look at the tokens per dollar per watt, right? Which I think in some 834 00:47:32,560 --> 00:47:37,237 [clears throat] sense, I would claim that GDP growth in any place will be 835 00:47:37,320 --> 00:47:40,877 directly correlated. Like you if you sort of buy my entire argument that 836 00:47:40,960 --> 00:47:45,877 look, we've got a new commodity, it's tokens Right. and the job of every 837 00:47:45,960 --> 00:47:51,397 economy and every firm in the economy is to translate these tokens into economic 838 00:47:51,480 --> 00:47:54,797 growth. Then if you have a cheaper commodity, 839 00:47:54,880 --> 00:47:58,757 it's better. And so, that's sort of what why this 840 00:47:58,840 --> 00:48:02,397 tokens per dollar per watt. And by the way, there are many, many elements to 841 00:48:02,480 --> 00:48:05,677 this, right? Which is it's not just the 842 00:48:05,760 --> 00:48:09,717 production side. That's why I think even having the grid is important. 843 00:48:09,800 --> 00:48:12,077 Uh construction costs, right? So, if you 844 00:48:12,160 --> 00:48:14,717 like if you think about the total TCO 845 00:48:14,800 --> 00:48:17,397 everything, it's like how are you a 846 00:48:17,480 --> 00:48:19,557 cheap producer of energy? Can you build 847 00:48:19,640 --> 00:48:22,397 the data centers? Then what's the cost 848 00:48:22,480 --> 00:48:26,437 curve of the silicon in the systems? And 849 00:48:26,520 --> 00:48:29,077 by the way, look at the token pricing, 850 00:48:29,160 --> 00:48:32,397 right? Token pricing basically drops by, 851 00:48:32,480 --> 00:48:35,997 you know, a half every 3 months. 852 00:48:36,080 --> 00:48:38,397 I mean, this is a So, how So, that 853 00:48:38,480 --> 00:48:40,917 that's why I think you can sort of 854 00:48:41,000 --> 00:48:44,557 really plot how you use the tokens to 855 00:48:44,640 --> 00:48:48,717 create surplus knowing that you have a commodity that's whose prices are just 856 00:48:48,800 --> 00:48:50,397 going to monotonically come down in a 857 00:48:50,480 --> 00:48:52,197 pretty fast curve. We're sitting in 858 00:48:52,280 --> 00:48:56,997 Europe. And there is a real fear 859 00:48:57,080 --> 00:49:00,957 because of the Europe does not have its own power. 860 00:49:01,040 --> 00:49:03,557 It has to import mostly of its power. 861 00:49:03,640 --> 00:49:07,517 Um Do you have any messages for Europe 862 00:49:07,600 --> 00:49:10,957 related to this? Yeah, I mean, I think So, there are two sets of things, right? 863 00:49:11,040 --> 00:49:13,477 One is you know, here we are in in 864 00:49:13,560 --> 00:49:18,317 Switzerland and I look at the the pharma or the financial sector, 865 00:49:18,400 --> 00:49:22,757 you know, obviously they're they they do do a big job in in 866 00:49:22,840 --> 00:49:25,877 in this country as in in Europe, but they're also international brands and 867 00:49:25,960 --> 00:49:30,597 international operations. So, one thing that whenever I think about Europe is 868 00:49:30,680 --> 00:49:34,717 the Europeans are producing products and services that actually are going 869 00:49:34,800 --> 00:49:37,517 everywhere in the world. And so, therefore, 870 00:49:37,600 --> 00:49:41,717 European competitiveness is about the competitiveness of their output 871 00:49:41,800 --> 00:49:45,197 globally, not just in Europe. I think sometimes when you come to Europe, 872 00:49:45,280 --> 00:49:47,077 there's a lot of conversation about just 873 00:49:47,160 --> 00:49:50,917 Europe. But European economy is 874 00:49:51,000 --> 00:49:54,877 thrives and has, you know, thrived in the last whatever 200 years, 300 years. 875 00:49:54,960 --> 00:49:59,157 The miracle of the West is fundamentally because of what has happened in Europe, 876 00:49:59,240 --> 00:50:02,717 is because they were able to produce things that the world needed. And so I 877 00:50:02,800 --> 00:50:06,357 would say that's number one. And in order to do that, you again, I go back 878 00:50:06,440 --> 00:50:08,317 to the human capital here is just 879 00:50:08,400 --> 00:50:13,437 fantastic, it's world-class. You have to absolutely invest in 880 00:50:13,520 --> 00:50:18,117 producing, you know, having the energy and the tokens here, which again, you're 881 00:50:18,200 --> 00:50:22,317 attracting like as I said, we are investing and others are investing 882 00:50:22,400 --> 00:50:26,357 the data centers here. So the question is, what's that next generation of 883 00:50:26,440 --> 00:50:29,797 output that comes from here, right? I always think about the German 884 00:50:29,880 --> 00:50:33,877 Mittelstand. Whenever I go to a jeweler or a dentist in the United States, I'm 885 00:50:33,960 --> 00:50:35,957 surrounded by German Mittelstand, right? 886 00:50:36,040 --> 00:50:38,757 Yes. Totally. It's just unbelievable 887 00:50:38,840 --> 00:50:42,277 engineering prowess of that country. 888 00:50:42,360 --> 00:50:43,757 And now the question And by the way, 889 00:50:43,840 --> 00:50:46,317 that's the point. They are producing 890 00:50:46,400 --> 00:50:49,557 industrial products, which today are 891 00:50:49,640 --> 00:50:52,877 built in into it all the intelligence as 892 00:50:52,960 --> 00:50:55,437 well. That data, right? So I know 893 00:50:55,520 --> 00:50:57,277 whenever we come to Europe, everyone's 894 00:50:57,360 --> 00:50:59,917 like talking about sovereignty and data 895 00:51:00,000 --> 00:51:02,757 this, data that. Guess what? Europe 896 00:51:02,840 --> 00:51:05,757 actually should be much more concerned 897 00:51:05,840 --> 00:51:08,637 about access to their industrial 898 00:51:08,720 --> 00:51:11,117 companies, their financial services 899 00:51:11,200 --> 00:51:14,517 companies of data from you US and the 900 00:51:14,600 --> 00:51:18,757 rest of the world. As opposed to just thinking that somehow by protecting 901 00:51:18,840 --> 00:51:20,357 Europe, you're going to be competitive. 902 00:51:20,440 --> 00:51:25,077 You are only going to be competitive if the products coming out of Europe are 903 00:51:25,160 --> 00:51:26,677 globally competitive. 904 00:51:26,760 --> 00:51:30,117 >> Right. And so that's I think what needs 905 00:51:30,200 --> 00:51:32,277 to change. And you know, Europe has led 906 00:51:32,360 --> 00:51:35,117 in privacy, that's fantastic, has led in 907 00:51:35,200 --> 00:51:39,797 many aspects of even safety around AI and what have you. And that's a feature 908 00:51:39,880 --> 00:51:44,277 that's great. But you also have to complement it with by building locally 909 00:51:44,360 --> 00:51:48,997 and then also thinking globally. What's the contribution this continent will 910 00:51:49,080 --> 00:51:52,237 make to the rest of the world, which it has 911 00:51:52,320 --> 00:51:54,757 historically been a leader in. 912 00:51:54,840 --> 00:51:58,917 >> So do you think the whole idea around sovereignty of data, is that being 913 00:51:59,000 --> 00:52:02,877 misunderstood? I I think that that when people talk 914 00:52:02,960 --> 00:52:07,677 about sovereignty, first of all, it's very important, 915 00:52:07,760 --> 00:52:09,757 clearly Who owns the data? And in a week 916 00:52:09,840 --> 00:52:12,797 like this, it's more important. Uh 917 00:52:12,880 --> 00:52:15,837 but that said, uh 918 00:52:15,920 --> 00:52:19,997 it is you have to kind of think about how what 919 00:52:20,080 --> 00:52:24,237 is sovereignty mean. Like for example, in the AI era, the topic that's least 920 00:52:24,320 --> 00:52:26,637 talked about, but I feel will be most 921 00:52:26,720 --> 00:52:29,797 talked about in in this this calendar 922 00:52:29,880 --> 00:52:32,637 year, will be the sovereignty of a firm. 923 00:52:32,720 --> 00:52:35,357 Just imagine, if your firm 924 00:52:35,440 --> 00:52:38,237 you're not able to embed the tacit 925 00:52:38,320 --> 00:52:42,597 knowledge of the firm in a set of weights 926 00:52:42,680 --> 00:52:45,237 in a model that you control, 927 00:52:45,320 --> 00:52:47,837 by definition, you have no sovereignty. 928 00:52:47,920 --> 00:52:49,677 That means you're leaking enterprise 929 00:52:49,760 --> 00:52:52,997 value to some model company somewhere. 930 00:52:53,080 --> 00:52:56,437 In fact, that it's sort of fascinating that nobody's talking about that, right? 931 00:52:56,520 --> 00:52:59,797 It's like everybody's talking about everything else that is sort of, you 932 00:52:59,880 --> 00:53:03,157 know, outside of that. Whereas the most 933 00:53:03,240 --> 00:53:06,917 important thing is, it really doesn't matter if you have In fact, the data 934 00:53:07,000 --> 00:53:10,677 center, where it runs, is the least important thing, quite honestly. But 935 00:53:10,760 --> 00:53:14,317 like even there, first of all, the data centers all are all over just because 936 00:53:14,400 --> 00:53:16,797 speed of light is a real constraint. 937 00:53:16,880 --> 00:53:21,077 And so therefore, the data centers will be spread. You will have digital You'll 938 00:53:21,160 --> 00:53:24,837 be able to encrypt everything. You'll be able to have the keys with you. All of 939 00:53:24,920 --> 00:53:26,957 these are much more technically solved 940 00:53:27,040 --> 00:53:29,837 problems. But the one problem that will 941 00:53:29,920 --> 00:53:33,037 only be solved is by you having much 942 00:53:33,120 --> 00:53:37,797 more sovereignty over the you know, tacit knowledge and control over the 943 00:53:37,880 --> 00:53:42,557 models. And it's not a one-way enterprise value transfer. And so to me, 944 00:53:42,640 --> 00:53:44,797 I think sovereignty requires real 945 00:53:44,880 --> 00:53:47,277 thought on what is it? 946 00:53:47,360 --> 00:53:49,717 You know, control of destiny means that 947 00:53:49,800 --> 00:53:52,797 your your ability to produce something 948 00:53:52,880 --> 00:53:57,117 that is unique is preserved. David Ricardo was not wrong. There's 949 00:53:57,200 --> 00:53:59,757 comparative advantage in countries. 950 00:53:59,840 --> 00:54:03,077 There is comparative advantage in firms. 951 00:54:03,160 --> 00:54:07,277 That needs to be preserved even in the AI era. That's what will give you real 952 00:54:07,360 --> 00:54:11,797 sovereignty. One last question. I know we're running out of time. 953 00:54:11,880 --> 00:54:15,837 Um in 5 years or 10 years, is there going to be one dominant model that 954 00:54:15,920 --> 00:54:20,237 we're all going to be using or or And how is Microsoft preparing for this? Are 955 00:54:20,320 --> 00:54:22,117 you going to be Are we going to be using 956 00:54:22,200 --> 00:54:25,077 one model for for enterprise, one model 957 00:54:25,160 --> 00:54:28,157 for other other trades? You know, even 958 00:54:28,240 --> 00:54:30,757 in the last whatever 3 years, 4 years 959 00:54:30,840 --> 00:54:32,677 that we've been at it, 960 00:54:32,760 --> 00:54:36,357 the the reality at this point is it's a 961 00:54:36,440 --> 00:54:41,677 multi-model world, right? I mean, the In fact, if you think about it, 962 00:54:41,760 --> 00:54:46,277 the in both there are going to be 963 00:54:46,360 --> 00:54:48,157 multiple models. 964 00:54:48,240 --> 00:54:51,277 And the trick is really how do you take 965 00:54:51,360 --> 00:54:54,117 advantage of these multiple models 966 00:54:54,200 --> 00:54:56,557 and in fact, build your own model by 967 00:54:56,640 --> 00:54:59,557 distilling these, right? So think of 968 00:54:59,640 --> 00:55:02,757 these models that you orchestrate to 969 00:55:02,840 --> 00:55:04,397 build your own model. And more 970 00:55:04,480 --> 00:55:07,677 importantly, you do what is described as 971 00:55:07,760 --> 00:55:10,597 orchestration or harness engineering. So 972 00:55:10,680 --> 00:55:14,677 the IP of any application or any firm is 973 00:55:14,760 --> 00:55:17,917 how do you use all these models with 974 00:55:18,000 --> 00:55:20,077 context engineering or your data? Yes. 975 00:55:20,160 --> 00:55:24,197 Right? So it's that three parts. So can I bring in all the models, by the way, 976 00:55:24,280 --> 00:55:26,797 which is close source, open source, 977 00:55:26,880 --> 00:55:30,037 build my own model, orchestrate them, 978 00:55:30,120 --> 00:55:33,437 and feed it my data to change the 979 00:55:33,520 --> 00:55:36,077 trajectory of some outcome that I care 980 00:55:36,160 --> 00:55:38,437 about. That's it. That's the entire 981 00:55:38,520 --> 00:55:40,797 picture. So you can do it in like, oh, I 982 00:55:40,880 --> 00:55:43,397 produce a particular product or service. 983 00:55:43,480 --> 00:55:48,557 First, I got to do better better job in sales or better job in R&D or better job 984 00:55:48,640 --> 00:55:52,237 in finance or what have you. And you take that outcome, and then you say, can 985 00:55:52,320 --> 00:55:54,757 I use all the models, orchestrate them, 986 00:55:54,840 --> 00:55:57,197 and feed it my context? And then in as a 987 00:55:57,280 --> 00:56:00,277 result of it, the reasoning traces are 988 00:56:00,360 --> 00:56:03,117 really leading to some capability and 989 00:56:03,200 --> 00:56:06,397 models that I control as my IP. As long 990 00:56:06,480 --> 00:56:10,957 as firms can answer that question, they're going to be getting ahead. 991 00:56:11,040 --> 00:56:15,827 Ladies and gentlemen, let's thank Satya, my friend, 992 00:56:15,910 --> 00:56:19,440 >> [applause] >> for 993 00:56:20,960 --> 00:56:25,797 And hopefully this is the beginning of many great dialogue and conversations here 994 00:56:25,880 --> 00:56:30,280 at the World Economic Forum. Thank you everyone. Thank you. 995 01:09:28,480 --> 01:09:30,640 >> Mhm. 996 01:12:21,240 --> 01:12:23,400 >> Mhm.