1 00:00:05,860 --> 00:00:09,010 Speaker 1: Welcome to COVID time, podcast series on markets and economies 2 00:00:09,010 --> 00:00:12,529 Speaker 1: from DBS Group Research. I'm Tee, chief economist, welcoming you 3 00:00:12,529 --> 00:00:16,889 Speaker 1: to our 161st episode. Today, we'll dive into the world 4 00:00:16,889 --> 00:00:20,599 Speaker 1: of cloud services, and for which we have an industry leader. 5 00:00:20,770 --> 00:00:24,840 Speaker 1: Jeff Johnson is MD and general manager of Amazon Web Services, 6 00:00:25,090 --> 00:00:27,250 Speaker 1: or what is now known as AWS. He has been 7 00:00:27,250 --> 00:00:29,489 Speaker 1: with AWS for almost 8 years, prior to which he 8 00:00:29,489 --> 00:00:32,729 Speaker 1: spent over a decade with Microsoft. Through these jobs, Jeff 9 00:00:32,729 --> 00:00:33,970 Speaker 1: has led teams on core. 10 00:00:34,125 --> 00:00:38,034 Speaker 1: Structure, enterprise, and commercial sales. Jeff Johnson, welcome to 11 00:00:38,034 --> 00:00:40,244 Speaker 2: Kobe. Thank you, Tammer. Thank you so much for having me. 12 00:00:40,314 --> 00:00:41,005 Speaker 2: It's great to be here. 13 00:00:41,115 --> 00:00:42,793 Speaker 1: It's great to have you. I've been looking forward to 14 00:00:42,793 --> 00:00:45,994 Speaker 1: this conversation because it's such a fascinating field right now, Jeff. 15 00:00:46,075 --> 00:00:49,115 Speaker 1: So let's actually begin by talking about AWS. Give us 16 00:00:49,115 --> 00:00:52,915 Speaker 1: a sense of the scale and breadth of your operations globally. 17 00:00:53,240 --> 00:00:55,110 Speaker 1: Yeah, Regionally and also. Yeah, 18 00:00:55,270 --> 00:00:58,659 Speaker 2: sure, so it's, you know, I think an incredible time 19 00:00:58,659 --> 00:01:01,349 Speaker 2: for around innovation and you know our story of innovation 20 00:01:01,349 --> 00:01:06,709 Speaker 2: at AWS just continues in terms of globally now, we 21 00:01:06,709 --> 00:01:12,779 Speaker 2: have over 120 availability zones across 38 different geographic regions 22 00:01:13,309 --> 00:01:14,510 Speaker 2: across Southeast Asia. 23 00:01:14,910 --> 00:01:19,739 Speaker 2: Uh, we have AWS regions in Singapore, in Indonesia, in Malaysia, 24 00:01:20,160 --> 00:01:22,910 Speaker 2: and in Thailand, and we just continue to really try 25 00:01:22,910 --> 00:01:26,269 Speaker 2: and bring our services closest to our customers so they 26 00:01:26,269 --> 00:01:30,190 Speaker 2: can take advantage of cloud services and and AI. I 27 00:01:30,190 --> 00:01:31,989 Speaker 2: think what really kind of sets us apart is also 28 00:01:31,989 --> 00:01:34,589 Speaker 2: how we build these regions, Tama, and you know, one 29 00:01:34,589 --> 00:01:35,989 Speaker 2: of one of the things that we do is when 30 00:01:35,989 --> 00:01:40,010 Speaker 2: we create a region, it's really creating 3 different data 31 00:01:40,010 --> 00:01:43,309 Speaker 2: centers that are physically separate and isolated. 32 00:01:43,760 --> 00:01:47,629 Speaker 2: And what this does is provide just great levels of resilience, uh, 33 00:01:47,650 --> 00:01:51,169 Speaker 2: enabling customers to be able to run mission critical workloads 34 00:01:51,569 --> 00:01:56,449 Speaker 2: on our infrastructure to meet complex regulatory or compliance standards 35 00:01:56,449 --> 00:01:59,330 Speaker 2: as well. I'd say the other area that we really 36 00:01:59,330 --> 00:02:01,809 Speaker 2: focus on a lot is security, so we like to 37 00:02:01,809 --> 00:02:05,330 Speaker 2: say that security, AWS and Amazon is is job zero, 38 00:02:05,449 --> 00:02:08,119 Speaker 2: so security is very much in a built in, uh, 39 00:02:08,130 --> 00:02:10,369 Speaker 2: from the ground up. And so if I take, uh, 40 00:02:10,380 --> 00:02:12,210 Speaker 2: innovations that we've made, um, 41 00:02:12,679 --> 00:02:15,869 Speaker 2: Across our data centers, so for example, we've engineered at 42 00:02:15,869 --> 00:02:19,440 Speaker 2: the at the silicon layer something called nitro, which actually 43 00:02:19,440 --> 00:02:22,080 Speaker 2: takes some of the virtualization capabilities and puts it on 44 00:02:22,080 --> 00:02:25,149 Speaker 2: a separate chip. That means that the surface area is 45 00:02:25,149 --> 00:02:28,500 Speaker 2: reduced from threats and attacks, so we're innovating right down 46 00:02:28,960 --> 00:02:31,960 Speaker 2: to the silicon layer, and I think what this shows 47 00:02:31,960 --> 00:02:33,888 Speaker 2: really is that, you know, we continue to provide, you know, 48 00:02:33,919 --> 00:02:35,929 Speaker 2: breadth and depth of services and. 49 00:02:36,429 --> 00:02:38,990 Speaker 2: We're super proud that actually Gartner have just recognized us 50 00:02:38,990 --> 00:02:42,279 Speaker 2: for the 15th consecutive year as being a leader in 51 00:02:42,279 --> 00:02:47,470 Speaker 2: the cloud infrastructure, magic quadrants, so, uh, our ability to 52 00:02:47,470 --> 00:02:49,750 Speaker 2: execute is, is ranked as the highest, and I think 53 00:02:49,750 --> 00:02:52,110 Speaker 2: this is really about that breadth and depth of service, 54 00:02:52,149 --> 00:02:52,869 Speaker 2: but also. 55 00:02:53,619 --> 00:02:57,008 Speaker 2: The scale, as you mentioned, that we have globally and 56 00:02:57,300 --> 00:03:00,220 Speaker 2: particularly in this region here, and then also our partner 57 00:03:00,220 --> 00:03:04,399 Speaker 2: ecosystem as well with the 200,000 partners supporting our customers 58 00:03:04,399 --> 00:03:07,970 Speaker 2: to help them take advantage of these services and innovation 59 00:03:07,970 --> 00:03:08,179 Speaker 2: as 60 00:03:08,179 --> 00:03:08,490 Speaker 2: well. 61 00:03:08,570 --> 00:03:12,728 Speaker 1: And how central is Singapore to your operations in the region? 62 00:03:13,258 --> 00:03:16,500 Speaker 2: Yeah, so, so Singapore is, is uh what is the 63 00:03:16,500 --> 00:03:20,380 Speaker 2: APJ headquarters of our APJ team and business here. 64 00:03:20,788 --> 00:03:23,130 Speaker 2: But it's part of, you know, collection of regions that 65 00:03:23,130 --> 00:03:25,910 Speaker 2: we have, as I mentioned across ASEAN, we have 4 66 00:03:25,910 --> 00:03:30,579 Speaker 2: regions here in Indonesia and Malaysia, and Thailand, and each 67 00:03:30,580 --> 00:03:34,339 Speaker 2: of these regions represents significant investments that we've we've made 68 00:03:34,339 --> 00:03:37,710 Speaker 2: in the region. So actually just last year at a 69 00:03:37,710 --> 00:03:40,289 Speaker 2: summit here in Singapore, we announced a further 70 00:03:40,779 --> 00:03:44,979 Speaker 2: $12 million US dollars of investment into Singapore in terms 71 00:03:44,979 --> 00:03:49,460 Speaker 2: of our cloud infrastructure build out in Malaysia. It's about 72 00:03:49,460 --> 00:03:52,300 Speaker 2: a $6 billion investment that we announced there through to 73 00:03:52,300 --> 00:03:56,600 Speaker 2: 2038 and in Indonesia and Thailand a further $5 billion 74 00:03:56,600 --> 00:03:58,419 Speaker 2: each in those regions, so. 75 00:03:58,850 --> 00:04:02,779 Speaker 2: So again, we continue to invest holistically really across the 76 00:04:02,779 --> 00:04:05,289 Speaker 2: region and, and really our focus is on really trying to, 77 00:04:05,300 --> 00:04:08,339 Speaker 2: you know, bring, bring these services as close as we 78 00:04:08,339 --> 00:04:11,080 Speaker 2: can to our customers. And I think what excites me 79 00:04:11,080 --> 00:04:13,259 Speaker 2: probably the most is, you know, how we're helping our 80 00:04:13,259 --> 00:04:16,260 Speaker 2: customers transform using these services. So if I think about 81 00:04:16,649 --> 00:04:20,100 Speaker 2: customers right here in Singapore like Grab, uh, which, uh, 82 00:04:20,220 --> 00:04:23,200 Speaker 2: you know, is used by over 30 million active users 83 00:04:23,200 --> 00:04:25,170 Speaker 2: on a on a monthly basis now. 84 00:04:25,589 --> 00:04:28,789 Speaker 2: Uh, and really AWS are powering and and the the 85 00:04:28,790 --> 00:04:31,750 Speaker 2: platform underneath all of, all of the services that that 86 00:04:31,750 --> 00:04:34,750 Speaker 2: Grab provide and and that's enabled Grab to also launch 87 00:04:34,750 --> 00:04:38,390 Speaker 2: new products quickly like digital banks and so on as well. Uh, 88 00:04:38,428 --> 00:04:40,109 Speaker 2: in the banking space, I know we're here in the 89 00:04:40,109 --> 00:04:43,500 Speaker 2: wonderful DBS studio, but, um, some of the digital banks 90 00:04:43,500 --> 00:04:45,820 Speaker 2: that have started up like uh Trust Bank, for example, 91 00:04:46,040 --> 00:04:49,510 Speaker 2: in the last 18 months, uh, again, built a digital 92 00:04:49,510 --> 00:04:52,420 Speaker 2: bank on top of AWS infrastructure. They're able to onboard 93 00:04:52,420 --> 00:04:54,790 Speaker 2: new customers in less than 3 minutes. 94 00:04:55,570 --> 00:04:58,700 Speaker 2: So this ability to move fast and be really, really 95 00:04:58,700 --> 00:05:02,659 Speaker 2: agile is important. We serve other large customers like Petronas 96 00:05:02,660 --> 00:05:06,769 Speaker 2: in Malaysia. We're helping them modernize their operations or Bank 97 00:05:06,769 --> 00:05:09,500 Speaker 2: Islam in Malaysia as well, where we're helping them bring 98 00:05:09,500 --> 00:05:15,899 Speaker 2: their digital banks to market quickly. Siam commercial Bank in Thailand, so. 99 00:05:16,329 --> 00:05:21,029 Speaker 2: So both from enterprises right through to to startups, you know, 100 00:05:21,109 --> 00:05:23,519 Speaker 2: there's there's a lot happening in that space as well, 101 00:05:23,549 --> 00:05:27,950 Speaker 2: just with startups creating, you know, many new generative AI solutions, 102 00:05:28,149 --> 00:05:34,070 Speaker 2: great examples like Botnoy in Thailand who provide translation services 103 00:05:34,070 --> 00:05:37,350 Speaker 2: used by over 10 million customers now or users using 104 00:05:37,350 --> 00:05:40,980 Speaker 2: that service. So, so incredible innovation. I think what we, um, 105 00:05:40,988 --> 00:05:42,709 Speaker 2: you know, what we really try and do is just 106 00:05:42,709 --> 00:05:45,149 Speaker 2: democratize access to this technology so. 107 00:05:45,920 --> 00:05:47,929 Speaker 2: If you have an idea and you have an ambition, 108 00:05:47,980 --> 00:05:51,140 Speaker 2: you can get started on AWS and build and obviously 109 00:05:51,140 --> 00:05:53,570 Speaker 2: the pay as you go cloud model just enables. 110 00:05:53,880 --> 00:05:57,630 Speaker 2: Businesses to try and experiment quickly but also scale very 111 00:05:57,630 --> 00:06:01,390 Speaker 2: quickly within regional or more broadly as well with the 112 00:06:01,390 --> 00:06:02,420 Speaker 2: global infrastructure 113 00:06:04,010 --> 00:06:06,670 Speaker 1: footprint, absolutely fascinating. So I was doing a quick mental 114 00:06:06,670 --> 00:06:10,230 Speaker 1: summation when you were mentioning those country specific investments, so 115 00:06:10,230 --> 00:06:12,750 Speaker 1: about 20 to $30 billion worth of investments in the 116 00:06:12,750 --> 00:06:16,109 Speaker 1: coming years just in this ASEAN area. 117 00:06:16,459 --> 00:06:19,039 Speaker 1: Um, and toward the end of your response, you mentioned 118 00:06:19,040 --> 00:06:21,329 Speaker 1: something about innovation and helping startups. So let's talk a 119 00:06:21,329 --> 00:06:23,399 Speaker 1: little bit about that in the context of Singapore. You 120 00:06:23,399 --> 00:06:24,440 Speaker 1: have an innovation hub here. 121 00:06:24,519 --> 00:06:27,000 Speaker 2: Yes, yeah, no, thanks for mentioning that. So just a 122 00:06:27,000 --> 00:06:29,920 Speaker 2: couple of months back we launched our innovation hub here, 123 00:06:29,928 --> 00:06:32,570 Speaker 2: which again we think is, you know, just a really, 124 00:06:32,760 --> 00:06:38,480 Speaker 2: you know, unique experience to take customers and partners and 125 00:06:38,480 --> 00:06:42,079 Speaker 2: students through to really just showcase some of this technology 126 00:06:42,079 --> 00:06:43,760 Speaker 2: that I think we're all reading about often, you know, 127 00:06:43,799 --> 00:06:44,238 Speaker 2: in the 128 00:06:44,579 --> 00:06:47,000 Speaker 2: In the press on a daily basis, but, um, so 129 00:06:47,000 --> 00:06:50,808 Speaker 2: we've built this environment, it's really immersive, it's over 8000 130 00:06:50,809 --> 00:06:55,079 Speaker 2: square feet. There's over 50 different demonstrations across industries ranging 131 00:06:55,079 --> 00:06:57,279 Speaker 2: from AI smart farming to. 132 00:06:57,670 --> 00:07:02,119 Speaker 2: Uh, to smart manufacturing, to blockchain, to areas like Kuiper, 133 00:07:02,149 --> 00:07:05,309 Speaker 2: which is an area where um Kuiper is our low 134 00:07:05,309 --> 00:07:09,709 Speaker 2: Earth orbit satellite solution that we've we've launched and we'll 135 00:07:09,709 --> 00:07:13,190 Speaker 2: be providing actually full coverage around to Southeast Asia by 136 00:07:13,190 --> 00:07:16,829 Speaker 2: the end of 2026. Um, so anyway, we, we, we 137 00:07:16,829 --> 00:07:19,309 Speaker 2: showcase all of these technologies in a way that really 138 00:07:19,309 --> 00:07:20,190 Speaker 2: helps customers. 139 00:07:20,839 --> 00:07:23,269 Speaker 2: Unlock maybe some of the aspiration they have around innovation, 140 00:07:23,359 --> 00:07:26,440 Speaker 2: helps them accelerate how they can really start to begin 141 00:07:26,440 --> 00:07:29,720 Speaker 2: to make that happen and and take action. So we're 142 00:07:29,720 --> 00:07:33,660 Speaker 2: actually using AI technologies to help customers maybe translate some 143 00:07:33,660 --> 00:07:36,250 Speaker 2: of that vision immediately into an action plan so they 144 00:07:36,250 --> 00:07:37,980 Speaker 2: can come through the experience and 145 00:07:38,390 --> 00:07:41,390 Speaker 2: Leave with a very sort of actionable set of activities 146 00:07:41,390 --> 00:07:43,709 Speaker 2: to go and execute. And and so we plan to 147 00:07:43,709 --> 00:07:49,239 Speaker 2: take over 1000 customers through the experience across the next year. We, 148 00:07:49,390 --> 00:07:52,399 Speaker 2: we're already, um, it's already kind of running hot every 149 00:07:52,399 --> 00:07:54,790 Speaker 2: day with customers running through there and a couple 100 150 00:07:54,790 --> 00:07:58,019 Speaker 2: students as well. And, and I think for me, uh, 151 00:07:58,309 --> 00:08:00,269 Speaker 2: it really sort of demonstrates one of the things I 152 00:08:00,269 --> 00:08:02,750 Speaker 2: really love about Amazon and AWS is our culture of 153 00:08:02,750 --> 00:08:03,869 Speaker 2: innovation and 154 00:08:04,549 --> 00:08:07,149 Speaker 2: You know, in short, I suppose that really gets us 155 00:08:07,149 --> 00:08:09,750 Speaker 2: to really sort of anchor around thinking every day is 156 00:08:09,750 --> 00:08:13,869 Speaker 2: day one and thinking like a startup, and, and so, uh, as, 157 00:08:13,910 --> 00:08:16,589 Speaker 2: as customers go through that innovation center, you know, one 158 00:08:16,589 --> 00:08:19,268 Speaker 2: of the sayings I like is we like to think big, 159 00:08:19,670 --> 00:08:23,829 Speaker 2: start small, scale fast, and, and really the innovation hub 160 00:08:23,829 --> 00:08:25,790 Speaker 2: really kind of captures that as a way to help 161 00:08:25,790 --> 00:08:28,940 Speaker 2: customers think big, see that ambition, but also be able 162 00:08:28,940 --> 00:08:31,109 Speaker 2: to start small and innovate and then think about how 163 00:08:31,109 --> 00:08:32,549 Speaker 2: to scale as well, so. 164 00:08:33,039 --> 00:08:35,299 Speaker 2: Uh, so we're very excited. You're super welcome to to 165 00:08:35,299 --> 00:08:37,239 Speaker 2: come visit soon and and just see some of the 166 00:08:37,239 --> 00:08:38,710 Speaker 2: technologies we're by all means. 167 00:08:38,719 --> 00:08:38,729 Speaker 1: So 168 00:08:38,729 --> 00:08:42,000 Speaker 1: Jeff, you also mentioned students. Can you tell me a 169 00:08:42,000 --> 00:08:43,919 Speaker 1: little bit about how students fit into this whole 170 00:08:43,919 --> 00:08:44,348 Speaker 1: thing? 171 00:08:44,719 --> 00:08:47,039 Speaker 2: Yeah, so, so, uh, so we just plan to, you know, 172 00:08:47,280 --> 00:08:50,988 Speaker 2: ensure that some of the students across universities as well will, will, uh, 173 00:08:51,000 --> 00:08:53,409 Speaker 2: target and take through the experience as well, just to, 174 00:08:53,479 --> 00:08:56,919 Speaker 2: to inspire, uh, the student populations. There's a, there's a, 175 00:08:57,039 --> 00:08:58,218 Speaker 2: you know, there's a lot that we 176 00:08:58,900 --> 00:09:02,359 Speaker 2: Also do just more broadly around, you know, empowerment of 177 00:09:02,359 --> 00:09:07,020 Speaker 2: skills and students through AWS digital skills and learning that 178 00:09:07,020 --> 00:09:08,229 Speaker 2: you know I can talk about in a bit. 179 00:09:09,000 --> 00:09:12,520 Speaker 1: So I'm actually a big beneficiary of one of the 180 00:09:12,520 --> 00:09:15,109 Speaker 1: capabilities of AWS so we used to have this. 181 00:09:16,640 --> 00:09:18,710 Speaker 1: What is it called, the speed racer or the cyber racer? 182 00:09:19,000 --> 00:09:22,789 Speaker 1: Deep racer exactly, uh, and, and the whole idea of 183 00:09:22,789 --> 00:09:24,199 Speaker 1: you can sort of come up with your own little 184 00:09:24,200 --> 00:09:27,239 Speaker 1: algorithm or solutions to make the car go as fast 185 00:09:27,239 --> 00:09:30,510 Speaker 1: or as efficiently as possible with all these, you know, 186 00:09:30,679 --> 00:09:33,479 Speaker 1: like a dozen if hundreds of parameters that you can 187 00:09:33,479 --> 00:09:34,799 Speaker 1: play with, uh. 188 00:09:35,390 --> 00:09:38,069 Speaker 1: At an employee level, like, you know, hundreds of devious 189 00:09:38,070 --> 00:09:40,940 Speaker 1: staff could model train these things and try it out 190 00:09:40,940 --> 00:09:42,090 Speaker 1: and go for this virtual race. 191 00:09:42,190 --> 00:09:43,169 Speaker 2: I love that you played with that. 192 00:09:43,469 --> 00:09:46,270 Speaker 1: Yes, it didn't do very well, but there were others 193 00:09:46,270 --> 00:09:48,270 Speaker 1: in the technology and operation side who I think were 194 00:09:48,270 --> 00:09:51,349 Speaker 1: more dedicated than I was. It wasn't trivial. It was dating. 195 00:09:51,465 --> 00:09:53,814 Speaker 1: Intensive it was demanding, which I really loved. It wasn't 196 00:09:53,815 --> 00:09:56,974 Speaker 1: some sort of a trivial pursuit. It was pretty substantial. Yeah, 197 00:09:57,354 --> 00:09:59,215 Speaker 2: it's interesting. I mean, we keep trying to find ways 198 00:09:59,215 --> 00:10:01,414 Speaker 2: to sort of gamify some of this learning and you know, 199 00:10:01,455 --> 00:10:04,085 Speaker 2: a couple of other examples. Actually something that we've built 200 00:10:04,085 --> 00:10:06,974 Speaker 2: is called the League of Large Language models, which is 201 00:10:06,974 --> 00:10:07,655 Speaker 2: a way of 202 00:10:08,020 --> 00:10:10,849 Speaker 2: Uh, in the same way that we gamified deep races 203 00:10:10,849 --> 00:10:14,250 Speaker 2: to tune, you know, a racing sort of car analogy, 204 00:10:14,309 --> 00:10:16,539 Speaker 2: but in this case, like how to tune a large 205 00:10:16,539 --> 00:10:19,729 Speaker 2: language model and so we've been running these game like 206 00:10:19,729 --> 00:10:22,559 Speaker 2: large scale sessions where people are in a competition to 207 00:10:22,890 --> 00:10:25,659 Speaker 2: create the most effective fine tuning of a large language 208 00:10:25,659 --> 00:10:26,690 Speaker 2: model and 209 00:10:27,070 --> 00:10:29,699 Speaker 2: So, uh, yeah, and actually just while we're on the 210 00:10:29,700 --> 00:10:32,728 Speaker 2: kind of topic, but just actually just this weekend in Malaysia, 211 00:10:32,979 --> 00:10:38,130 Speaker 2: we launched what is officially Malaysia's largest ever uh national hackathon, 212 00:10:38,820 --> 00:10:42,539 Speaker 2: so with Minister Gobind supporting as well as a wide 213 00:10:42,539 --> 00:10:46,020 Speaker 2: number of universities, but so over the the coming weeks, 214 00:10:46,099 --> 00:10:49,890 Speaker 2: we have a number of students and organizations all participating, uh, 215 00:10:50,260 --> 00:10:54,179 Speaker 2: so there's almost 2000 individuals participating in this, in this 216 00:10:54,179 --> 00:10:55,718 Speaker 2: large scale hackathon across. 217 00:10:56,409 --> 00:10:58,809 Speaker 2: Malaysia, uh, that we kicked off at the weekend. So, 218 00:10:59,169 --> 00:11:01,210 Speaker 2: you know, we're, um, you know, we love really just 219 00:11:01,210 --> 00:11:03,750 Speaker 2: trying to find ways to, um, really make it fun to, 220 00:11:03,849 --> 00:11:06,289 Speaker 2: to learn about these new technologies and bring it to 221 00:11:06,289 --> 00:11:08,770 Speaker 2: life in different ways and, you know, in the context 222 00:11:08,770 --> 00:11:12,169 Speaker 2: of an organization or at a, at a more nationwide level, right. 223 00:11:12,820 --> 00:11:16,400 Speaker 1: Um, speaking of AI, yeah, what is Amazon Nova? 224 00:11:16,940 --> 00:11:21,770 Speaker 2: Yeah, so Nova is, is a first party model that 225 00:11:21,770 --> 00:11:26,130 Speaker 2: we've built. So our approach around generative AI and AI 226 00:11:26,130 --> 00:11:28,690 Speaker 2: is to provide a choice of models. We believe one 227 00:11:28,690 --> 00:11:31,329 Speaker 2: model won't rule them all, and so we want customers 228 00:11:31,330 --> 00:11:33,809 Speaker 2: to have the widest range of choice. We think many 229 00:11:33,809 --> 00:11:36,699 Speaker 2: tasks will be performed by, you know, touching a number 230 00:11:36,700 --> 00:11:37,650 Speaker 2: of different models. 231 00:11:38,140 --> 00:11:40,978 Speaker 2: But Nova is a family of models that we launched 232 00:11:40,979 --> 00:11:46,030 Speaker 2: reinvent last December, and this really enables customers to, to 233 00:11:46,030 --> 00:11:48,190 Speaker 2: have you know, the best of access to first party 234 00:11:48,190 --> 00:11:52,549 Speaker 2: models through Amazon and AWS as well. And there are some, 235 00:11:52,630 --> 00:11:55,460 Speaker 2: you know, some great examples already that we're seeing in 236 00:11:55,469 --> 00:11:59,219 Speaker 2: in ASEAN of customers using this, so first light productions in, 237 00:11:59,270 --> 00:12:03,669 Speaker 2: in the Philippines who are an education production studio producing. 238 00:12:03,979 --> 00:12:06,919 Speaker 2: Uh, a lot of their school and educational content now 239 00:12:06,919 --> 00:12:10,479 Speaker 2: using some of the media capabilities of, of Nova, but yeah, 240 00:12:10,570 --> 00:12:13,080 Speaker 2: there are thousands of customers that are adopting this, and, 241 00:12:13,440 --> 00:12:15,409 Speaker 2: but I would, I would also just position it in 242 00:12:15,409 --> 00:12:18,250 Speaker 2: the context of, you know, our view is that customers 243 00:12:18,250 --> 00:12:20,770 Speaker 2: want choice. Customers want to be, and, and we've seen, 244 00:12:20,849 --> 00:12:23,349 Speaker 2: I think, just the proliferation of models that's, you know, 245 00:12:23,450 --> 00:12:25,978 Speaker 2: that's happened over the last couple of years. So that's 246 00:12:25,979 --> 00:12:28,469 Speaker 2: very much at the core of our strategy, but I can, 247 00:12:28,530 --> 00:12:30,770 Speaker 2: I can, I can tell you a bit more about where. 248 00:12:31,320 --> 00:12:33,710 Speaker 2: Uh, how we're seeing AI and just the direction if 249 00:12:33,710 --> 00:12:34,340 Speaker 2: that's helpful. 250 00:12:34,450 --> 00:12:34,989 Speaker 1: By all means. 251 00:12:35,099 --> 00:12:39,380 Speaker 1: So like GBD 5 and Cloud and Gemini, Nova is 252 00:12:39,380 --> 00:12:44,819 Speaker 1: your sort of all-purpose LLM, which is available in the 253 00:12:44,820 --> 00:12:47,819 Speaker 1: public domain. I can go to the website or download 254 00:12:47,820 --> 00:12:49,500 Speaker 1: the app and use it just the way I use 255 00:12:49,500 --> 00:12:50,390 Speaker 1: any other LLMs. 256 00:12:50,500 --> 00:12:52,179 Speaker 2: Yeah, you can use it. Well, I mean, we, we 257 00:12:52,179 --> 00:12:56,789 Speaker 2: really encourage customers to use models through Amazon Bedrock. There's 258 00:12:56,789 --> 00:13:00,059 Speaker 2: there's you know, reasons for doing that and so Bedrock is. 259 00:13:00,349 --> 00:13:05,289 Speaker 2: Platform that enables customers to very quickly build generative applications, 260 00:13:05,299 --> 00:13:08,539 Speaker 2: but it also comes with a number of other services 261 00:13:08,539 --> 00:13:14,179 Speaker 2: such as guardrail, security features, automated reasoning to uh to 262 00:13:14,179 --> 00:13:16,419 Speaker 2: remove hallucinations and so on. So we, you know, we 263 00:13:16,419 --> 00:13:17,349 Speaker 2: think that actually 264 00:13:17,630 --> 00:13:19,330 Speaker 2: Uh, there's a lot of value when you're using these 265 00:13:19,330 --> 00:13:21,130 Speaker 2: models that we can help just customers with a lot 266 00:13:21,130 --> 00:13:23,530 Speaker 2: of that kind of infrastructure and those key kind of 267 00:13:23,530 --> 00:13:27,960 Speaker 2: enterprise grade, you know, capabilities as well. But, uh, but yeah, there's, uh, 268 00:13:28,049 --> 00:13:29,900 Speaker 2: I mean, you know, other ways that we're just helping, 269 00:13:30,049 --> 00:13:32,848 Speaker 2: you know, customers get really easy access to these kind 270 00:13:32,849 --> 00:13:35,728 Speaker 2: of capabilities as well is, uh, through we launched a 271 00:13:35,729 --> 00:13:38,320 Speaker 2: product called Kiro actually at New York summit in July, 272 00:13:38,369 --> 00:13:40,619 Speaker 2: which is really aimed at the, um, 273 00:13:41,039 --> 00:13:45,830 Speaker 2: Uh, uh, vibe coding, uh, type capabilities as, as, you know, 274 00:13:46,039 --> 00:13:48,780 Speaker 2: a term that's becoming, I think familiar to many people now, but, 275 00:13:48,789 --> 00:13:51,919 Speaker 2: you know, just this very low bar to entry ability to, 276 00:13:51,950 --> 00:13:54,789 Speaker 2: you know, create a loose specification of a task and 277 00:13:54,789 --> 00:13:57,549 Speaker 2: then for the tool to create a plan and then 278 00:13:57,549 --> 00:14:00,829 Speaker 2: start to build applications. So, uh, so yeah, we're, we're 279 00:14:00,830 --> 00:14:03,390 Speaker 2: kind of approaching um how we think about AI just 280 00:14:03,390 --> 00:14:05,099 Speaker 2: from sort of multiple layers from 281 00:14:05,460 --> 00:14:08,859 Speaker 2: At the silicon layer as well. So we've, uh, for 282 00:14:08,859 --> 00:14:13,780 Speaker 2: some time now had custom silicon processors to enable model 283 00:14:13,780 --> 00:14:17,260 Speaker 2: training and model inference. So we have ranium and inferentia 284 00:14:17,780 --> 00:14:20,820 Speaker 2: and really these drive down the cost of inference and 285 00:14:20,820 --> 00:14:23,859 Speaker 2: the cost of training models. So ranium 2, which is 286 00:14:23,859 --> 00:14:27,700 Speaker 2: our latest, uh, process, is about 3 times more energy 287 00:14:27,700 --> 00:14:31,500 Speaker 2: efficient and and performant. And so this really helps customers 288 00:14:31,500 --> 00:14:32,809 Speaker 2: train models more efficiently. 289 00:14:33,590 --> 00:14:36,270 Speaker 2: On top of that, then we have a tooling for 290 00:14:36,270 --> 00:14:41,099 Speaker 2: data scientists, so Sagemaker and SageMaker unified studio. So, so 291 00:14:41,099 --> 00:14:45,330 Speaker 2: to enable a data scientists to quickly build trained models, bedrock, 292 00:14:45,349 --> 00:14:48,900 Speaker 2: as I mentioned, then, is really just this aggregation layer 293 00:14:48,900 --> 00:14:52,630 Speaker 2: that helps customers quickly build applications, use any model that 294 00:14:52,630 --> 00:14:57,830 Speaker 2: they want to use, have the benefits of guardrails, automated reasoning, security. 295 00:14:58,390 --> 00:15:01,849 Speaker 2: And then on top of that, the applications that we have. So, uh, Q, 296 00:15:01,929 --> 00:15:04,919 Speaker 2: so queue for Business, which enables business users to just 297 00:15:04,919 --> 00:15:08,650 Speaker 2: really access insights from across their organization. We have Que 298 00:15:08,650 --> 00:15:12,169 Speaker 2: for developer, which has developers build applications quickly and then 299 00:15:12,169 --> 00:15:16,330 Speaker 2: tools like Kiro to also build new applications. So, so 300 00:15:16,330 --> 00:15:18,729 Speaker 2: really the, you know, the, the thought and philosophy that 301 00:15:18,729 --> 00:15:20,169 Speaker 2: we have is that um, 302 00:15:21,000 --> 00:15:24,010 Speaker 2: Look, if, if, if most customers are are like us 303 00:15:24,010 --> 00:15:26,760 Speaker 2: and and like DBS where you're, you know, you're constantly thinking, 304 00:15:26,890 --> 00:15:29,450 Speaker 2: you know, how can I improve the customer experience, what 305 00:15:29,450 --> 00:15:32,090 Speaker 2: can I do to improve that going forward, then if 306 00:15:32,090 --> 00:15:34,010 Speaker 2: you're thinking like that, then you're probably gonna be thinking 307 00:15:34,010 --> 00:15:36,450 Speaker 2: how can I use AI and generative AI to do that, 308 00:15:36,570 --> 00:15:38,010 Speaker 2: and if you're doing that, you're probably going to be 309 00:15:38,010 --> 00:15:39,599 Speaker 2: using a lot of inference and. 310 00:15:40,000 --> 00:15:41,760 Speaker 2: Have a lot of different models that you want to use. 311 00:15:41,840 --> 00:15:44,239 Speaker 2: So it's in that sort of concept that we think 312 00:15:44,239 --> 00:15:47,960 Speaker 2: actually providing bedrock, providing a range of models, making it 313 00:15:47,960 --> 00:15:51,159 Speaker 2: easy for customers to access and switch between these different 314 00:15:51,159 --> 00:15:54,039 Speaker 2: models and, you know, that's the interesting thing I see 315 00:15:54,039 --> 00:15:56,070 Speaker 2: you know, customers that maybe might start on one model 316 00:15:56,070 --> 00:15:59,200 Speaker 2: and change to another or combine it and, and we 317 00:15:59,200 --> 00:16:01,880 Speaker 2: see some, you know, just amazing um impacts I would 318 00:16:01,880 --> 00:16:04,679 Speaker 2: say across, across a range of customers. So if I 319 00:16:04,679 --> 00:16:07,640 Speaker 2: take uh in the in the telco space in Indonesia, 320 00:16:07,719 --> 00:16:08,760 Speaker 2: Telcom sell with. 321 00:16:09,380 --> 00:16:15,340 Speaker 2: over 130 million customers and subscribers, they wanted to kind 322 00:16:15,340 --> 00:16:19,229 Speaker 2: of reimagine some of their customer service capabilities and really 323 00:16:19,229 --> 00:16:22,140 Speaker 2: transformed the kind of IT operations side into a service 324 00:16:22,140 --> 00:16:26,109 Speaker 2: that was really proactive and able to dynamically resolve issues. 325 00:16:26,580 --> 00:16:29,809 Speaker 2: And so building with us, you know, they, they created 326 00:16:29,809 --> 00:16:34,460 Speaker 2: a solution that was able to manage incident analysis 21% 327 00:16:34,460 --> 00:16:38,309 Speaker 2: faster and actually incident resolution 83% faster, so. 328 00:16:38,679 --> 00:16:40,650 Speaker 2: So when you think about that in the context of 329 00:16:40,650 --> 00:16:44,409 Speaker 2: changing uh customer satisfaction and customer journeys, it's, you know, 330 00:16:44,489 --> 00:16:48,570 Speaker 2: it's really powerful, um, or indeed, uh, you know, perhaps 331 00:16:48,570 --> 00:16:52,169 Speaker 2: uh down in Australia, Commonwealth Bank Australia, who we, we 332 00:16:52,169 --> 00:16:57,210 Speaker 2: have a strategic partnership with, but, um, we migrated about 333 00:16:57,210 --> 00:17:02,440 Speaker 2: uh 61,000 of their data pipelines uh onto AWS and then. 334 00:17:02,919 --> 00:17:06,199 Speaker 2: have over 2000 AI models running, and the data point 335 00:17:06,199 --> 00:17:09,688 Speaker 2: that I love is they, uh, there are 55 million, uh, 336 00:17:09,760 --> 00:17:12,839 Speaker 2: AI driven decisions made every day now, uh, by, by 337 00:17:12,839 --> 00:17:16,839 Speaker 2: the platforms. So, so when you kind of think about just, just, just, 338 00:17:16,900 --> 00:17:19,689 Speaker 2: just that in the kind of context of, of using AI, 339 00:17:19,699 --> 00:17:22,469 Speaker 2: I think I'm, you know, I'm seeing kind of organizations 340 00:17:22,469 --> 00:17:25,478 Speaker 2: maybe move from, um, you know, maybe by being you 341 00:17:25,479 --> 00:17:28,239 Speaker 2: obviously moving to cloud and cloud first to thinking about 342 00:17:28,239 --> 00:17:32,000 Speaker 2: um AI first, but also thinking about being AI native and. 343 00:17:32,300 --> 00:17:35,250 Speaker 2: I would say the sort of difference between AI first 344 00:17:35,250 --> 00:17:37,329 Speaker 2: thinking I want to adopt and use lots of AI 345 00:17:37,329 --> 00:17:41,050 Speaker 2: across my services, as opposed to AI Native and the 346 00:17:41,050 --> 00:17:44,369 Speaker 2: Commonwealth Bank example is let's let's reimagine our processes from 347 00:17:44,369 --> 00:17:46,489 Speaker 2: scratch and really think about how would we build it 348 00:17:46,489 --> 00:17:49,579 Speaker 2: if we didn't have any of those existing tools and we, we, 349 00:17:49,609 --> 00:17:51,410 Speaker 2: we kind of started with AI so. 350 00:17:52,079 --> 00:17:54,060 Speaker 2: So this is, um, you know, I think it's a 351 00:17:54,060 --> 00:17:57,458 Speaker 2: really kind of interesting, uh, uh, you know, kind of era, 352 00:17:57,500 --> 00:18:00,140 Speaker 2: this kind of probably decade of AI that we're, we're, 353 00:18:00,260 --> 00:18:02,319 Speaker 2: you know, a few years into now and seeing some 354 00:18:02,319 --> 00:18:05,819 Speaker 2: of these transformations. But uh, yeah, we're, we're, you know, 355 00:18:05,900 --> 00:18:08,899 Speaker 2: incredibly excited about it. We think this is probably, you know, 356 00:18:08,939 --> 00:18:11,199 Speaker 2: as big as some of the transformation eras of, of, 357 00:18:11,260 --> 00:18:14,689 Speaker 2: of internet, of cloud computing, you know, maybe of invention 358 00:18:14,689 --> 00:18:16,900 Speaker 2: of of the microprocessor even and so on. 359 00:18:17,369 --> 00:18:19,189 Speaker 2: Uh, in, in terms of that impact, 360 00:18:19,550 --> 00:18:23,619 Speaker 1: right, I particularly resonate with your point, Jeff, that the 361 00:18:24,469 --> 00:18:26,920 Speaker 1: one model would not be ruling all that we would 362 00:18:26,920 --> 00:18:29,109 Speaker 1: have to have platforms which allow us to sort of 363 00:18:29,109 --> 00:18:31,750 Speaker 1: do horse races around that. Now, you are of course 364 00:18:31,750 --> 00:18:34,150 Speaker 1: giving me, you know, very high end corporate examples. I'll 365 00:18:34,150 --> 00:18:36,750 Speaker 1: give you a very mundane personal level example. So my 366 00:18:36,750 --> 00:18:39,669 Speaker 1: 12 year old, uh, he was asked to come up 367 00:18:39,670 --> 00:18:42,379 Speaker 1: with an answer for what sustainability means to you. 368 00:18:42,719 --> 00:18:45,239 Speaker 1: And we asked Cloud and Gemini and Deep Sea and 369 00:18:45,239 --> 00:18:48,010 Speaker 1: a bunch of other models, the same question. It's very interesting, 370 00:18:48,020 --> 00:18:50,410 Speaker 1: the kind of different answers we got, and then the 371 00:18:50,410 --> 00:18:53,209 Speaker 1: winning answer that he resonated with most was an analogy 372 00:18:53,209 --> 00:18:55,369 Speaker 1: where you think of the Earth as a spaceship moving 373 00:18:55,369 --> 00:18:57,890 Speaker 1: through time and you need to have enough food and 374 00:18:57,890 --> 00:19:01,010 Speaker 1: water for the next generation of astronauts. And, and I 375 00:19:01,010 --> 00:19:02,729 Speaker 1: thought that storytelling at first and because it was a 376 00:19:02,729 --> 00:19:05,419 Speaker 1: deep sea, which had the reasoning in there, it was 377 00:19:05,420 --> 00:19:07,930 Speaker 1: a pretty compelling to a 12 year old. Uh, I'm 378 00:19:07,930 --> 00:19:10,369 Speaker 1: sure an automated solution would be better, and that's where 379 00:19:10,369 --> 00:19:11,900 Speaker 1: its platforms come in. 380 00:19:12,560 --> 00:19:16,349 Speaker 1: Um, I want to ask you one question about the 381 00:19:16,349 --> 00:19:16,849 Speaker 1: sort of the 382 00:19:18,010 --> 00:19:23,099 Speaker 1: Distinguishing characteristics between AI strategy and a Gen AI strategy. 383 00:19:23,569 --> 00:19:26,719 Speaker 1: Whether it is DBS or AWS, Gen AI is a 384 00:19:26,719 --> 00:19:29,530 Speaker 1: 345 year journey, but AI has been around for a 385 00:19:29,530 --> 00:19:32,109 Speaker 1: long time and machine learning and so on, and including 386 00:19:32,109 --> 00:19:35,010 Speaker 1: the deep freezer. These are decades old pursuits. 387 00:19:35,369 --> 00:19:39,229 Speaker 1: So when you talk about cloud native versus cloud first, 388 00:19:39,270 --> 00:19:42,069 Speaker 1: I think you've won that battle. That's, that's done in 389 00:19:42,069 --> 00:19:44,750 Speaker 1: the industry. But when we talk about AI first versus 390 00:19:44,750 --> 00:19:47,420 Speaker 1: Gen AI first, are we still talking about two different things? 391 00:19:48,020 --> 00:19:49,750 Speaker 2: Yeah, no, I think it's a really interesting point, and 392 00:19:49,750 --> 00:19:51,510 Speaker 2: you're right. I mean, just, I mean, in terms of, 393 00:19:51,589 --> 00:19:54,229 Speaker 2: you know, Amazon's own journey with with AIML, and I 394 00:19:54,229 --> 00:19:56,589 Speaker 2: mean this began right back at the dawn of Amazon 395 00:19:56,589 --> 00:19:57,550 Speaker 2: really when we were. 396 00:19:57,930 --> 00:20:02,260 Speaker 2: You know, providing, uh, personalized recommendations for books and so on, on, on, 397 00:20:02,300 --> 00:20:04,060 Speaker 2: on the, uh, you know, the very early websites and 398 00:20:04,060 --> 00:20:06,339 Speaker 2: so on. So, so it's something that, you know, we've 399 00:20:06,339 --> 00:20:08,698 Speaker 2: got a couple of decades of experience and, you know, 400 00:20:08,780 --> 00:20:10,899 Speaker 2: even if I think of my own journey, you know, 401 00:20:10,979 --> 00:20:13,020 Speaker 2: kind of 8 years ago on on on joining. 402 00:20:13,140 --> 00:20:16,770 Speaker 2: AWS and we were launching actually uh products back then 403 00:20:16,770 --> 00:20:20,010 Speaker 2: to make AI very accessible to customers. So we took 404 00:20:20,010 --> 00:20:22,889 Speaker 2: things where we'd like built a lot of machine learning 405 00:20:22,890 --> 00:20:25,849 Speaker 2: models around forecasting, for example. Turned out we had a 406 00:20:25,849 --> 00:20:28,010 Speaker 2: lot of experience of that in the retail business, so 407 00:20:28,010 --> 00:20:31,010 Speaker 2: we wrapped that up in some ML models, provided that 408 00:20:31,010 --> 00:20:33,930 Speaker 2: as an AI service that people could very quickly just 409 00:20:33,930 --> 00:20:36,329 Speaker 2: build and call from an application, so. 410 00:20:37,099 --> 00:20:39,869 Speaker 2: So we launched a set of services back then that 411 00:20:39,869 --> 00:20:42,109 Speaker 2: were really as, as you say, more kind of around 412 00:20:42,109 --> 00:20:45,389 Speaker 2: some of these AIML type capabilities, but really again with 413 00:20:45,390 --> 00:20:48,390 Speaker 2: the same mental model in mind actually, let's just really 414 00:20:48,390 --> 00:20:51,630 Speaker 2: just make this very accessible to customers and so we had, 415 00:20:51,670 --> 00:20:57,099 Speaker 2: you know, visual recognition capabilities, speech type capabilities for transcription, translation, 416 00:20:57,140 --> 00:20:59,859 Speaker 2: all those sorts of areas that we, we provided. 417 00:21:00,300 --> 00:21:02,030 Speaker 2: Uh, but yeah, I think as, as we've, you know, 418 00:21:02,339 --> 00:21:05,099 Speaker 2: kind of moved into, into the era of, of, of 419 00:21:05,099 --> 00:21:07,099 Speaker 2: generative AI, I think, I think, you know, probably some 420 00:21:07,099 --> 00:21:09,280 Speaker 2: of the things that have become even more important is, 421 00:21:09,380 --> 00:21:12,060 Speaker 2: you know, is, is again just the focus on data 422 00:21:12,060 --> 00:21:15,419 Speaker 2: and ensuring that, you know, data governance and is, is 423 00:21:15,420 --> 00:21:18,229 Speaker 2: strong and I think it's, you know, it's been interesting actually. 424 00:21:18,380 --> 00:21:21,099 Speaker 2: I think why we've seen relatively fast adoption in regulated 425 00:21:21,099 --> 00:21:23,989 Speaker 2: industries like financial services has been, there's been a lot 426 00:21:23,989 --> 00:21:26,130 Speaker 2: of work over many years done in terms of having, 427 00:21:26,219 --> 00:21:27,939 Speaker 2: you know, really great data governance and 428 00:21:28,589 --> 00:21:31,810 Speaker 2: Data systems in place, but I think, um, you know, 429 00:21:31,849 --> 00:21:35,688 Speaker 2: as organizations are looking at how they move from, you know, 430 00:21:35,729 --> 00:21:39,560 Speaker 2: maybe the, uh, the period of, you know, probably experimenting 431 00:21:39,560 --> 00:21:41,800 Speaker 2: with Gen AI use cases, which was probably the first 432 00:21:41,800 --> 00:21:44,150 Speaker 2: wave that we saw, you know, a couple of a 433 00:21:44,150 --> 00:21:47,290 Speaker 2: couple of years ago, um, obviously the things that we're 434 00:21:47,290 --> 00:21:50,489 Speaker 2: all familiar with like, you know, simple summarization and, and 435 00:21:50,489 --> 00:21:52,489 Speaker 2: so on. I think what we're seeing now is people 436 00:21:52,489 --> 00:21:54,530 Speaker 2: start to really begin to think, how can I, you know, 437 00:21:54,609 --> 00:21:56,389 Speaker 2: kind of industrialize that more and 438 00:21:56,619 --> 00:21:59,630 Speaker 2: And build up almost like a factory model and I think, um, 439 00:21:59,920 --> 00:22:01,959 Speaker 2: you know, uh, and and start to begin to think 440 00:22:01,959 --> 00:22:04,479 Speaker 2: about how to use agents as well. And, uh, you know, 441 00:22:04,560 --> 00:22:06,959 Speaker 2: just a couple of examples. In fact, just last week 442 00:22:06,959 --> 00:22:09,739 Speaker 2: here in Singapore, I was at the Keppel next event, uh, 443 00:22:09,920 --> 00:22:12,280 Speaker 2: which was, you know, again super interesting, but they were 444 00:22:12,280 --> 00:22:17,109 Speaker 2: showcasing uh the KAI platform Keppel AI, that's a platform 445 00:22:17,109 --> 00:22:19,339 Speaker 2: we've we've worked with Keppel to build out, but 446 00:22:20,030 --> 00:22:22,540 Speaker 2: Uh, they just showed some really great demonstrations, I think, of, 447 00:22:22,550 --> 00:22:25,369 Speaker 2: you know, Gen AI and Agentic AI in action where 448 00:22:25,369 --> 00:22:28,469 Speaker 2: they took processes like, you know, maybe they're trying to 449 00:22:28,469 --> 00:22:30,909 Speaker 2: understand whether there's an opportunity in the market to acquire 450 00:22:30,910 --> 00:22:33,389 Speaker 2: a new assets, some land, uh, maybe to build a 451 00:22:33,390 --> 00:22:36,188 Speaker 2: data center, build some infrastructure, you know, next step of 452 00:22:36,189 --> 00:22:39,109 Speaker 2: the process, building proposals, next step of the process, you know, 453 00:22:39,229 --> 00:22:40,989 Speaker 2: thinking about how you actually execute that. 454 00:22:41,500 --> 00:22:44,030 Speaker 2: And they've built a platform that you imagine that first 455 00:22:44,030 --> 00:22:46,189 Speaker 2: step that's just doing all the analysis, you know, just 456 00:22:46,189 --> 00:22:49,390 Speaker 2: crunching together all the reports, finding out maybe the best 457 00:22:49,390 --> 00:22:53,540 Speaker 2: site locations, land, looking at a variety of signals that, 458 00:22:54,030 --> 00:22:56,109 Speaker 2: you know, historically would have taken weeks and months, and 459 00:22:56,109 --> 00:22:58,708 Speaker 2: they're just able to do that far more quickly and 460 00:22:58,709 --> 00:23:01,050 Speaker 2: then sort of package, package that up. So I think. 461 00:23:01,369 --> 00:23:04,569 Speaker 2: You know, very much starting to see organizations think about more, 462 00:23:04,619 --> 00:23:07,119 Speaker 2: you know, these kind of end end to end, uh, 463 00:23:07,219 --> 00:23:09,339 Speaker 2: so rather than a task thinking about actually what's the 464 00:23:09,339 --> 00:23:12,139 Speaker 2: core function of the business and how can I uh 465 00:23:12,140 --> 00:23:16,180 Speaker 2: trans uh translate that into, into uh generative AI and 466 00:23:16,180 --> 00:23:19,339 Speaker 2: agentic AI and then, you know, just, just one more, um, 467 00:23:19,369 --> 00:23:22,219 Speaker 2: example actually just from last week, I know, um, we 468 00:23:22,219 --> 00:23:24,300 Speaker 2: were just chatting that I was, I was in Hanoi 469 00:23:24,300 --> 00:23:27,209 Speaker 2: in Vietnam for a cloud summit and cloud day there 470 00:23:27,209 --> 00:23:28,938 Speaker 2: and um Techcom Bank were. 471 00:23:29,550 --> 00:23:33,689 Speaker 2: Uh, sharing story, so, um, you know, one of Vietnam's 472 00:23:33,689 --> 00:23:38,419 Speaker 2: largest joint stock held banks in Vietnam were sharing, um, 473 00:23:38,829 --> 00:23:41,369 Speaker 2: how whilst they've gone through a cloud journey and migrated, 474 00:23:41,390 --> 00:23:44,479 Speaker 2: you know, 80% of their applications in cloud and are, 475 00:23:44,589 --> 00:23:48,310 Speaker 2: you know, using a lot of personalization already, uh, they 476 00:23:48,310 --> 00:23:52,390 Speaker 2: have over 12,000 features to describe a customer, which going 477 00:23:52,390 --> 00:23:54,310 Speaker 2: back to this sort of point of data and really 478 00:23:54,310 --> 00:23:55,660 Speaker 2: getting hyper personalized. 479 00:23:56,199 --> 00:23:58,849 Speaker 2: 12,000 features for a customer. But what was very interesting 480 00:23:58,849 --> 00:24:02,489 Speaker 2: was how they were thinking about um Agentic AI as 481 00:24:02,489 --> 00:24:06,300 Speaker 2: a kind of extensibility point for their organization. So if 482 00:24:06,300 --> 00:24:08,599 Speaker 2: you think they're a bank that's gone through going from, 483 00:24:08,810 --> 00:24:13,410 Speaker 2: you know, maybe 57 million customers to 16.5 million customers today, 484 00:24:13,449 --> 00:24:16,968 Speaker 2: and they're continuing to grow and uh the affluence of 485 00:24:16,969 --> 00:24:18,569 Speaker 2: their customers continues to grow. 486 00:24:19,060 --> 00:24:21,750 Speaker 2: They realize that, you know, they can't just maybe keep 487 00:24:21,750 --> 00:24:23,869 Speaker 2: continuing to grow their headcount. They need to think about 488 00:24:23,869 --> 00:24:26,429 Speaker 2: how they scale and scale with AI and so, you know, 489 00:24:26,510 --> 00:24:29,188 Speaker 2: they're thinking about the role of a relationship manager in 490 00:24:29,189 --> 00:24:31,629 Speaker 2: the future as they sort of move further up the wealth, 491 00:24:32,069 --> 00:24:34,329 Speaker 2: uh food chain, if I can call it that, how they're, 492 00:24:34,390 --> 00:24:36,630 Speaker 2: you know, going to be using those tools to to 493 00:24:36,630 --> 00:24:37,579 Speaker 2: help them scale. 494 00:24:38,239 --> 00:24:40,930 Speaker 2: And then another example that I thought was very interesting 495 00:24:40,930 --> 00:24:43,329 Speaker 2: they gave was, you know, thinking about an area of 496 00:24:43,329 --> 00:24:45,929 Speaker 2: their market, uh, you know, like, like uh credits and 497 00:24:45,930 --> 00:24:48,879 Speaker 2: loans into SMEs and how historically, 498 00:24:49,650 --> 00:24:51,889 Speaker 2: No, they they'd sort of struggle to get a, you know, 499 00:24:52,010 --> 00:24:54,530 Speaker 2: a really good sort of telemetry around, you know, whether 500 00:24:54,530 --> 00:24:57,079 Speaker 2: it was viable to loan to some of these organizations, 501 00:24:57,089 --> 00:24:59,829 Speaker 2: but they've built a system using generative AI where they 502 00:24:59,829 --> 00:25:03,079 Speaker 2: just take 3 pictures of, of the actual physical location 503 00:25:03,079 --> 00:25:06,520 Speaker 2: and then are able to determine from that the credit risk. 504 00:25:06,890 --> 00:25:11,260 Speaker 2: And this is unlocked 700,000 new SMEs that they can 505 00:25:11,410 --> 00:25:14,130 Speaker 2: uh potentially loan to and so on and and create 506 00:25:14,130 --> 00:25:15,170 Speaker 2: business around. So. 507 00:25:15,910 --> 00:25:18,010 Speaker 2: So I think, I think we've gone from, you know, 508 00:25:18,089 --> 00:25:21,449 Speaker 2: maybe someone doing some perhaps, you know, fraud detection and 509 00:25:21,449 --> 00:25:25,680 Speaker 2: assembling information and summarization into much broader end to end 510 00:25:25,680 --> 00:25:30,050 Speaker 2: use cases, uh, people beginning to organizations thinking about how 511 00:25:30,050 --> 00:25:32,770 Speaker 2: I can scale that, you know, and start to industrialize 512 00:25:32,770 --> 00:25:36,130 Speaker 2: that some more. Um, and then, um, yeah, also thinking 513 00:25:36,130 --> 00:25:39,810 Speaker 2: about agentic AI as this as this scalability point for 514 00:25:39,810 --> 00:25:42,290 Speaker 2: the organization as well. So. 515 00:25:42,859 --> 00:25:44,750 Speaker 2: So I think, yeah, very different from, you know, probably the, 516 00:25:44,790 --> 00:25:46,989 Speaker 2: if I can call it the classical AI period and 517 00:25:46,989 --> 00:25:49,429 Speaker 2: so on, you know, I think we were used to 518 00:25:49,430 --> 00:25:53,430 Speaker 2: seeing customers benefit from, you know, improving their inventory forecasting 519 00:25:53,430 --> 00:25:56,909 Speaker 2: using ML models and so on. So actually beginning to 520 00:25:57,310 --> 00:26:00,909 Speaker 2: think about how we can drive even more significant business 521 00:26:00,910 --> 00:26:03,849 Speaker 2: impact through, you know, scaling or unlocking new areas of 522 00:26:03,849 --> 00:26:04,750 Speaker 2: the market, etc. 523 00:26:05,430 --> 00:26:06,390 Speaker 1: So this is 524 00:26:06,660 --> 00:26:10,640 Speaker 1: You know, potentially sort of historic, to your point that 525 00:26:10,640 --> 00:26:14,359 Speaker 1: as we have seen previous waves of disruptive technology and 526 00:26:14,359 --> 00:26:17,540 Speaker 1: they have been consequential, no question about it, but all 527 00:26:17,540 --> 00:26:20,040 Speaker 1: of these technology rollouts come with their challenges. So let's 528 00:26:20,040 --> 00:26:23,920 Speaker 1: talk a little bit about that. Um, security, something that 529 00:26:23,920 --> 00:26:26,530 Speaker 1: we all worry about, especially in a red space like finance, 530 00:26:26,599 --> 00:26:28,879 Speaker 1: but I think any corporate client that you have would 531 00:26:28,880 --> 00:26:31,520 Speaker 1: want a secure Gen AI or AI system. 532 00:26:31,880 --> 00:26:35,349 Speaker 1: So what are your thoughts on challenges around that and 533 00:26:35,349 --> 00:26:37,780 Speaker 1: the possible solutions that are being worked through in the industry? 534 00:26:37,949 --> 00:26:40,739 Speaker 2: Yeah, sure, for sure, you know, as I touched on earlier, 535 00:26:41,189 --> 00:26:45,670 Speaker 2: absolutely job zero for us. Amazon AWS is security. Everything 536 00:26:45,670 --> 00:26:49,829 Speaker 2: is designed. The first thing that is considered is security and, and, and, 537 00:26:49,849 --> 00:26:52,739 Speaker 2: and so that's, you know, the first point I would say, 538 00:26:52,750 --> 00:26:54,630 Speaker 2: and and that sort of plays through into how we 539 00:26:54,630 --> 00:26:55,149 Speaker 2: think about. 540 00:26:55,560 --> 00:26:58,099 Speaker 2: AI and Gen AI, and really, you know, we're we're 541 00:26:58,099 --> 00:27:01,300 Speaker 2: balancing the we want to of course democratize access and 542 00:27:01,569 --> 00:27:05,260 Speaker 2: provide the broadest capabilities to our customers, but in a 543 00:27:05,260 --> 00:27:07,979 Speaker 2: way that meets the highest safety standards and the highest 544 00:27:07,979 --> 00:27:10,979 Speaker 2: security standards. So, so let me just explain a little 545 00:27:10,979 --> 00:27:12,699 Speaker 2: bit about how we try and think about that and 546 00:27:12,699 --> 00:27:14,180 Speaker 2: how we try and achieve that. So. 547 00:27:14,619 --> 00:27:16,899 Speaker 2: So firstly, I would say, you know, it's, it's awesome 548 00:27:16,900 --> 00:27:20,819 Speaker 2: that that some of the emerging standards around responsible AI 549 00:27:20,819 --> 00:27:25,329 Speaker 2: like ISO 42001, so we are one of the first 550 00:27:25,329 --> 00:27:28,979 Speaker 2: cloud providers to fully support that standard, and, and what 551 00:27:28,979 --> 00:27:32,179 Speaker 2: that means is organizations that are concerned about trustworthy and 552 00:27:32,180 --> 00:27:34,738 Speaker 2: responsible AI know that all of the services and how 553 00:27:34,739 --> 00:27:38,060 Speaker 2: we run our AI services meet that international standard. So, 554 00:27:38,380 --> 00:27:40,530 Speaker 2: so that's sort of a standards level. 555 00:27:41,030 --> 00:27:44,109 Speaker 2: I'd say secondly then from a, from a platform perspective 556 00:27:44,109 --> 00:27:47,229 Speaker 2: as well. So I explained a little bit about, about 557 00:27:47,229 --> 00:27:49,869 Speaker 2: Bedrock earlier and, and so some of the features that 558 00:27:49,869 --> 00:27:53,550 Speaker 2: we built into Bedrock. Um, so our guardrail's feature is 559 00:27:53,550 --> 00:27:56,670 Speaker 2: a way of just removing harmful content and allowing customers 560 00:27:56,670 --> 00:28:01,109 Speaker 2: to remove harmful content. Uh, we created a feature called 561 00:28:01,109 --> 00:28:02,869 Speaker 2: automated automated reasoning. 562 00:28:03,310 --> 00:28:05,760 Speaker 2: And actually this has been derived from, turns out we 563 00:28:05,760 --> 00:28:09,680 Speaker 2: had a, you know, a bunch of scientists in our 564 00:28:09,680 --> 00:28:11,910 Speaker 2: organization that had gotten very good at doing kind of 565 00:28:11,910 --> 00:28:15,160 Speaker 2: mathematical proofs of proving data if it's sent from one 566 00:28:15,160 --> 00:28:17,238 Speaker 2: side of a network to another and solving some of 567 00:28:17,239 --> 00:28:19,560 Speaker 2: those areas. So we were able to take some of 568 00:28:19,560 --> 00:28:22,119 Speaker 2: that expertise and think about, well, if we're trying to 569 00:28:22,119 --> 00:28:25,599 Speaker 2: solve the problem of hallucination in models and wrong data 570 00:28:25,599 --> 00:28:28,069 Speaker 2: coming back, can we use some of those theories? And 571 00:28:28,069 --> 00:28:30,399 Speaker 2: so that's what we did. We took that science and 572 00:28:30,400 --> 00:28:31,400 Speaker 2: built it into 573 00:28:31,760 --> 00:28:36,160 Speaker 2: Automated reasoning to stop hallucination in models as well. Uh, 574 00:28:36,199 --> 00:28:38,390 Speaker 2: in a Nova family of models, then we've built some 575 00:28:38,390 --> 00:28:41,719 Speaker 2: of these capabilities automatically in there to, you know, automatically 576 00:28:41,719 --> 00:28:45,119 Speaker 2: remove harmful content and so on, uh, for our customers. So, 577 00:28:45,199 --> 00:28:47,359 Speaker 2: so we're also, you know, investing a lot in the, 578 00:28:47,400 --> 00:28:50,560 Speaker 2: in the technology and the stack to to help from 579 00:28:50,560 --> 00:28:53,839 Speaker 2: a security perspective as well. Uh, again, um, you know, 580 00:28:53,920 --> 00:28:57,400 Speaker 2: very much if customers are running their models and putting data. 581 00:28:58,119 --> 00:29:00,400 Speaker 2: into their models on AWS, you know, that data is 582 00:29:00,400 --> 00:29:02,890 Speaker 2: never going to be going externally. It's always within their 583 00:29:02,890 --> 00:29:07,469 Speaker 2: own controlled and, you know, virtual environment that's secured. And 584 00:29:07,469 --> 00:29:10,000 Speaker 2: from a security perspective, so, you know, we support over 585 00:29:10,000 --> 00:29:16,439 Speaker 2: 143 different security standards and compliance certifications. So again, you know, 586 00:29:16,479 --> 00:29:18,719 Speaker 2: the bar is, is, is very high just in terms 587 00:29:18,719 --> 00:29:21,000 Speaker 2: of all of the standards that we support for the 588 00:29:21,000 --> 00:29:23,839 Speaker 2: sort of core security and infrastructure. So. 589 00:29:24,280 --> 00:29:25,739 Speaker 2: So all in all, you know, we, we, you know, 590 00:29:25,780 --> 00:29:27,739 Speaker 2: we're trying to kind of uh ensure that we balance 591 00:29:27,739 --> 00:29:31,339 Speaker 2: agility with, with uh providing environments that meet the highest 592 00:29:31,339 --> 00:29:34,060 Speaker 2: safety standards. If I think of, um, you know, examples 593 00:29:34,060 --> 00:29:37,770 Speaker 2: like like Bolt tech, the uh uh insure tech, um, 594 00:29:38,390 --> 00:29:40,510 Speaker 2: A customer of ours, so you know they 595 00:29:41,180 --> 00:29:45,510 Speaker 2: operating over 30 different countries internationally. They wanted to build 596 00:29:45,510 --> 00:29:49,689 Speaker 2: some Gen AI solutions around claims processing and and underwriting, 597 00:29:49,739 --> 00:29:52,550 Speaker 2: and so we built this with them using bedrock. Uh, 598 00:29:52,699 --> 00:29:54,819 Speaker 2: it uses again a number of models, so not just 599 00:29:54,819 --> 00:29:56,939 Speaker 2: one model, they're able to kind of change and use 600 00:29:56,939 --> 00:29:59,430 Speaker 2: different models, but they're able to meet all of the 601 00:29:59,430 --> 00:30:02,979 Speaker 2: regulatory regulatory requirements and all the 30 different markets that 602 00:30:02,979 --> 00:30:06,579 Speaker 2: they're operating in as well. So, you know, we're trying 603 00:30:06,579 --> 00:30:09,699 Speaker 2: to help customers go fast with the standards of compliance. 604 00:30:09,805 --> 00:30:12,354 Speaker 2: Support we're, you know, trying to build as much into the, 605 00:30:12,435 --> 00:30:16,035 Speaker 2: into the services that we offer as well, but um yeah, 606 00:30:16,114 --> 00:30:19,584 Speaker 2: we you know, ultimately we know that just ensuring that we're, 607 00:30:19,714 --> 00:30:23,074 Speaker 2: you know, just always focused on providing the highest standards 608 00:30:23,074 --> 00:30:27,915 Speaker 2: of safety, security, and responsible AI is is our focus, right? 609 00:30:28,045 --> 00:30:30,834 Speaker 1: So in statistics we always talked about type 1 and 610 00:30:30,834 --> 00:30:33,994 Speaker 1: type 2 error, and if you want to avoid any 611 00:30:33,994 --> 00:30:36,594 Speaker 1: sort of false hypothesis being put out, you can just 612 00:30:36,594 --> 00:30:38,525 Speaker 1: choose not to accept any hypothesis at all. 613 00:30:38,969 --> 00:30:41,790 Speaker 1: So same with the hallucination, is it a feature or 614 00:30:41,790 --> 00:30:44,859 Speaker 1: a bug of large language models? Because if it's a feature, 615 00:30:45,189 --> 00:30:46,869 Speaker 1: then I suppose we can what we can do is 616 00:30:46,869 --> 00:30:49,089 Speaker 1: best is to come up with clever solutions to minimize it, 617 00:30:49,150 --> 00:30:50,949 Speaker 1: but we will never be able to eliminate it. What 618 00:30:50,949 --> 00:30:52,109 Speaker 1: are your thoughts on that? 619 00:30:52,270 --> 00:30:55,069 Speaker 2: Yeah, I mean, I think, I mean we're very much 620 00:30:55,069 --> 00:30:57,170 Speaker 2: trying to, you know, think of it as, you know, 621 00:30:57,229 --> 00:31:00,709 Speaker 2: eliminating it or providing customers choice to be able to 622 00:31:00,709 --> 00:31:02,670 Speaker 2: decide and still have, you know, human in the loop 623 00:31:02,670 --> 00:31:06,040 Speaker 2: on a lot of these aspects as well and so. 624 00:31:06,329 --> 00:31:10,079 Speaker 2: I think we're trying to again provide combinations of both guardrails, 625 00:31:10,239 --> 00:31:12,819 Speaker 2: both tools that can automatically begin to 626 00:31:13,170 --> 00:31:16,130 Speaker 2: Filter out noise, but again, just make it very transparent 627 00:31:16,130 --> 00:31:18,930 Speaker 2: for for customers to always be, you know, very clear 628 00:31:18,930 --> 00:31:21,140 Speaker 2: on and auditable on what's happened. 629 00:31:21,449 --> 00:31:23,650 Speaker 1: Now earlier I think at the very beginning of this 630 00:31:23,650 --> 00:31:26,449 Speaker 1: conversation you talked about, you know, 3 servers per service 631 00:31:26,449 --> 00:31:29,890 Speaker 1: is sort of helpful, yeah. So that brings me to 632 00:31:29,890 --> 00:31:32,890 Speaker 1: the question of reliability. Uh, you, you want these things 633 00:31:32,890 --> 00:31:37,910 Speaker 1: to be always on. You don't want any sort of stoppages. Uh, 634 00:31:38,329 --> 00:31:41,329 Speaker 1: is it becoming more challenging given the computational needs and 635 00:31:41,329 --> 00:31:42,569 Speaker 1: the high intensity? 636 00:31:43,199 --> 00:31:46,949 Speaker 1: Operations required around LLMs or is it something that the industry, 637 00:31:47,119 --> 00:31:48,910 Speaker 1: including yours, is trying to be on top of, 638 00:31:49,640 --> 00:31:52,800 Speaker 2: yeah, I would say it's something that again rather like security. 639 00:31:52,880 --> 00:31:56,959 Speaker 2: I think we've always really focused on resilience and reliability 640 00:31:56,959 --> 00:31:59,689 Speaker 2: just as a kind of a fundamental immutable tenant that 641 00:31:59,689 --> 00:32:02,640 Speaker 2: we have in terms of how we think about data 642 00:32:02,640 --> 00:32:04,520 Speaker 2: center design and so. 643 00:32:05,119 --> 00:32:07,640 Speaker 2: the kind of unique way that we build our regions, 644 00:32:07,760 --> 00:32:10,449 Speaker 2: so we call a region like Singapore or like Thailand, 645 00:32:11,119 --> 00:32:12,959 Speaker 2: a region, and we call it a region because it's 646 00:32:12,959 --> 00:32:16,680 Speaker 2: got this pattern of three different or configuration of three 647 00:32:16,680 --> 00:32:21,750 Speaker 2: different data centers essentially that are physically isolated and separated. 648 00:32:21,839 --> 00:32:24,619 Speaker 2: And so, you know, if there are issues with, you know, 649 00:32:24,719 --> 00:32:26,959 Speaker 2: natural disasters in one area and so on, there's, you know, 650 00:32:27,280 --> 00:32:29,800 Speaker 2: there's protection and resilience there. 651 00:32:30,459 --> 00:32:33,540 Speaker 2: And what we found, uh, you know, just over the, 652 00:32:33,550 --> 00:32:36,089 Speaker 2: again the, uh, you know, I suppose also, you know, we've, 653 00:32:36,229 --> 00:32:39,229 Speaker 2: we've built a lot of experience in terms of running these, uh, 654 00:32:39,310 --> 00:32:41,550 Speaker 2: you know, cloud data centers at scale now over the 655 00:32:41,550 --> 00:32:43,589 Speaker 2: last nearly sort of two decades and so on that 656 00:32:43,589 --> 00:32:44,430 Speaker 2: AWS has. 657 00:32:44,920 --> 00:32:47,199 Speaker 2: has been in operation and we like to say there's 658 00:32:47,199 --> 00:32:51,400 Speaker 2: no compression algorithm for experience that we've learned in terms 659 00:32:51,400 --> 00:32:55,079 Speaker 2: of of running the, you know, the largest internet scale 660 00:32:55,079 --> 00:32:58,869 Speaker 2: websites for for Netflix or Zoom or or customers like 661 00:32:58,869 --> 00:33:01,439 Speaker 2: that is is, you know, there's a lot of uh. 662 00:33:02,420 --> 00:33:03,930 Speaker 2: Uh, things that you need to get right in terms 663 00:33:03,930 --> 00:33:07,219 Speaker 2: of the infrastructure and reliability and so on and so many, 664 00:33:07,260 --> 00:33:10,420 Speaker 2: many mission critical customers and, and so, um, yeah, so 665 00:33:10,420 --> 00:33:12,849 Speaker 2: I mean, it's interesting actually, just, uh, some of the 666 00:33:12,849 --> 00:33:15,920 Speaker 2: uh APJ research that's been published so I think just 667 00:33:15,920 --> 00:33:19,170 Speaker 2: last year, a report by Frost and Sullivan that just 668 00:33:19,170 --> 00:33:22,790 Speaker 2: looked at the number of incident hours of of cloud providers, 669 00:33:23,140 --> 00:33:25,380 Speaker 2: and we are by far the lowest, probably about 2.5 670 00:33:25,380 --> 00:33:28,589 Speaker 2: times fewer incident hours than the next, you know, nearest 671 00:33:28,589 --> 00:33:30,099 Speaker 2: cloud provider, and so. 672 00:33:30,550 --> 00:33:32,400 Speaker 2: Um, so we, uh, you know, we don't take that 673 00:33:32,400 --> 00:33:35,410 Speaker 2: for granted. We continue to, to work out how to, uh, 674 00:33:35,420 --> 00:33:39,780 Speaker 2: you know, continue to focus on, on resilience and reliability, but, uh, yeah, 675 00:33:40,140 --> 00:33:42,250 Speaker 2: it's a, it's a big part of how we design 676 00:33:42,459 --> 00:33:45,369 Speaker 2: data centers. It's built into the, the kind of pattern of, of, 677 00:33:45,380 --> 00:33:46,410 Speaker 2: of how we do that. 678 00:33:47,010 --> 00:33:51,180 Speaker 1: So earlier you're talking about solutions to make the processes 679 00:33:51,180 --> 00:33:53,739 Speaker 1: more efficient, which I guess another way of saying it 680 00:33:53,739 --> 00:33:56,140 Speaker 1: is making more energy efficient. 681 00:33:57,280 --> 00:34:00,130 Speaker 1: Now, these sort of heavy computational solutions that we are 682 00:34:00,130 --> 00:34:04,089 Speaker 1: sort of embracing with LLMs are energy hungry. I'm seeing 683 00:34:04,089 --> 00:34:07,130 Speaker 1: stories of tariff rates and electricity going up sharply in 684 00:34:07,130 --> 00:34:09,709 Speaker 1: some southern southwestern part of the US because all of 685 00:34:09,709 --> 00:34:11,689 Speaker 1: a sudden these data centers are very hungry and they're 686 00:34:11,689 --> 00:34:14,610 Speaker 1: demanding a lot of electricity. Now, the typical answer to 687 00:34:14,610 --> 00:34:16,649 Speaker 1: that is, well, we can always build solar farms to 688 00:34:16,649 --> 00:34:18,770 Speaker 1: provide electricity to run the data centers, so therefore, the 689 00:34:18,770 --> 00:34:19,569 Speaker 1: solution is green. 690 00:34:19,969 --> 00:34:22,159 Speaker 1: So I want you to sort of address the sustainability 691 00:34:22,159 --> 00:34:25,600 Speaker 1: question that in the last 5 years with this LLM 692 00:34:25,600 --> 00:34:28,520 Speaker 1: GAI revolution, we have decided to go on to this 693 00:34:28,520 --> 00:34:32,919 Speaker 1: big next level of energy consumption. How are we sort 694 00:34:32,919 --> 00:34:35,959 Speaker 1: of managing our sustainability aspirations against that? 695 00:34:36,040 --> 00:34:38,040 Speaker 2: Yeah, yeah, yeah, sure. I know there's there's, you know, 696 00:34:38,120 --> 00:34:39,520 Speaker 2: there's a lot to cover and a lot we're sort 697 00:34:39,520 --> 00:34:41,569 Speaker 2: of focused on in this space, as you say, it's 698 00:34:42,029 --> 00:34:45,417 Speaker 2: Yeah, look, look sustainability and transition to green energy and 699 00:34:45,418 --> 00:34:47,729 Speaker 2: so on is something that we are again both at 700 00:34:47,729 --> 00:34:50,579 Speaker 2: an Amazon and AWS level, you know, super focused on 701 00:34:50,579 --> 00:34:53,618 Speaker 2: with climate pledge to, you know, to get to net 702 00:34:53,618 --> 00:34:56,658 Speaker 2: zero by 2040, but also we set a goal to 703 00:34:56,658 --> 00:35:01,089 Speaker 2: power all of Amazon's internal all of Amazon's operations rather, uh, 704 00:35:01,099 --> 00:35:04,448 Speaker 2: with renewable energy by 2030 and actually we've achieved that 705 00:35:04,498 --> 00:35:07,549 Speaker 2: that goal already in terms of meeting that which we're we're, 706 00:35:07,739 --> 00:35:09,099 Speaker 2: you know, we're super proud of, but 707 00:35:09,500 --> 00:35:11,100 Speaker 2: Some of the ways that we're working on it the 708 00:35:11,100 --> 00:35:15,139 Speaker 2: kind of multilayers to this. So firstly, um, again, external 709 00:35:15,139 --> 00:35:18,850 Speaker 2: research by folks like Accenture have shown that actually moving 710 00:35:19,419 --> 00:35:23,379 Speaker 2: applications from a traditional on-premise data center to AWS cloud 711 00:35:23,379 --> 00:35:26,819 Speaker 2: is about 4.1 times more energy efficient. So just by 712 00:35:26,820 --> 00:35:30,780 Speaker 2: moving to cloud to AWS Cloud, you can reduce energy consumption. 713 00:35:31,409 --> 00:35:34,969 Speaker 2: And actually, um, moving an application across can reduce its 714 00:35:34,969 --> 00:35:38,760 Speaker 2: carbon footprint by about 99% as well. So just actually 715 00:35:39,090 --> 00:35:41,570 Speaker 2: in terms of organizations looking uh to, you know, to 716 00:35:41,570 --> 00:35:44,169 Speaker 2: meet some of their green and sustainability goals, there's there's 717 00:35:44,169 --> 00:35:46,090 Speaker 2: benefits just in in moving to cloud. 718 00:35:46,570 --> 00:35:48,928 Speaker 2: We've been investing a lot in terms of how we 719 00:35:48,929 --> 00:35:51,570 Speaker 2: think about and evolve our data center designs. So our 720 00:35:51,929 --> 00:35:56,089 Speaker 2: latest data center designs are about provide about 12% more 721 00:35:56,090 --> 00:35:58,530 Speaker 2: compute power, which means we need less data centers. So 722 00:35:58,530 --> 00:36:01,449 Speaker 2: that's that's sort of helpful in terms of that. We 723 00:36:01,449 --> 00:36:04,080 Speaker 2: focus a lot on thinking about heat energy as well, 724 00:36:04,169 --> 00:36:09,169 Speaker 2: so particularly in the subtropical climate here in Southeast Asia. So. 725 00:36:09,689 --> 00:36:12,888 Speaker 2: We've been able to work on innovation to reduce mechanical 726 00:36:12,889 --> 00:36:17,759 Speaker 2: heat energy by 46% as well. Uh, so that's, that's important. 727 00:36:17,810 --> 00:36:21,679 Speaker 2: And then also have liquid cooling technologies that help across 728 00:36:21,679 --> 00:36:24,649 Speaker 2: our AI training workloads, and so on. So within the 729 00:36:24,649 --> 00:36:27,070 Speaker 2: data centers, there's, you know, a lot of focus to, to, 730 00:36:27,139 --> 00:36:30,010 Speaker 2: to help, um, you know, optimize and make them more efficient. 731 00:36:30,770 --> 00:36:34,169 Speaker 2: But the silicon layer then as well, so, so, uh, 732 00:36:34,179 --> 00:36:38,370 Speaker 2: we have built custom chips essentially that that provide a 733 00:36:38,370 --> 00:36:41,689 Speaker 2: far better power utilization. So we have a family of 734 00:36:41,689 --> 00:36:46,040 Speaker 2: processors called Graviton. They're in the actually 4th generation now, 735 00:36:46,610 --> 00:36:51,360 Speaker 2: but these uh use about 60% less energy than traditional 736 00:36:51,560 --> 00:36:55,799 Speaker 2: X86 kind of instance types that might be used. So again, very, 737 00:36:55,889 --> 00:36:58,020 Speaker 2: very energy efficient and 738 00:36:58,260 --> 00:37:01,679 Speaker 2: I briefly mentioned earlier Tranium, which is again a custom 739 00:37:01,679 --> 00:37:04,638 Speaker 2: silicon for training large language models, and this is about 740 00:37:04,639 --> 00:37:07,770 Speaker 2: 3 times more energy efficient than than previous models and 741 00:37:08,129 --> 00:37:12,610 Speaker 2: or previous uh chipsets. So we find that customers like Anthropic, 742 00:37:13,010 --> 00:37:17,169 Speaker 2: Apple reinvent show last year are usingranium for for training 743 00:37:17,169 --> 00:37:20,009 Speaker 2: models as well to to get some of these uh 744 00:37:20,010 --> 00:37:20,969 Speaker 2: energy efficiencies. 745 00:37:21,530 --> 00:37:23,929 Speaker 2: And then in terms of the kind of broader strategy, 746 00:37:24,540 --> 00:37:28,939 Speaker 2: we're absolutely investing in renewable energy, so we have probably 747 00:37:28,939 --> 00:37:34,699 Speaker 2: more than 600 projects now globally around renewable energy, right 748 00:37:34,699 --> 00:37:39,860 Speaker 2: here in Southeast Asia we have partnerships with EDP renewables 749 00:37:39,860 --> 00:37:43,659 Speaker 2: here in Singapore and Semcorp. We've got about 4 solar 750 00:37:43,659 --> 00:37:44,209 Speaker 2: projects in. 751 00:37:44,610 --> 00:37:48,570 Speaker 2: Indonesia with, with, with uh PLN. So again, we, you know, 752 00:37:48,810 --> 00:37:52,770 Speaker 2: continue to invest in uh renewable projects and finding ways 753 00:37:52,770 --> 00:37:56,449 Speaker 2: to scale renewable energy. So, so really we're, we're really 754 00:37:56,449 --> 00:37:58,919 Speaker 2: trying to tackle it from a point of view that we, 755 00:37:59,010 --> 00:37:59,669 Speaker 2: we know. 756 00:38:00,379 --> 00:38:04,850 Speaker 2: In order to really fulfill digital inclusion areas that that 757 00:38:04,850 --> 00:38:08,479 Speaker 2: can drive like financial inclusion and impacts the society, we 758 00:38:08,479 --> 00:38:11,439 Speaker 2: need to provide our services broadly and in order to 759 00:38:11,439 --> 00:38:13,000 Speaker 2: do that, we need to make sure that we're doing 760 00:38:13,000 --> 00:38:16,379 Speaker 2: it in a way that's really sustainable and and and 761 00:38:16,379 --> 00:38:21,000 Speaker 2: you know, just really provides a long term future, a 762 00:38:21,000 --> 00:38:23,489 Speaker 2: long term sustainable future. So very much, you know, at 763 00:38:23,489 --> 00:38:25,370 Speaker 2: the strategy just to continue to 764 00:38:26,080 --> 00:38:30,299 Speaker 2: Lower the power requirements of training models of the chip 765 00:38:30,300 --> 00:38:32,779 Speaker 2: sets that we run and then also find ways to 766 00:38:33,179 --> 00:38:35,750 Speaker 2: create renewable energy as well. 767 00:38:36,340 --> 00:38:39,169 Speaker 1: So earlier when we were talking about the multi-billion dollar 768 00:38:39,169 --> 00:38:41,899 Speaker 1: investment plans in the region, a subset of that is 769 00:38:41,899 --> 00:38:43,770 Speaker 1: into renewable energy investment as well. 770 00:38:44,320 --> 00:38:46,810 Speaker 2: Uh, so yeah, we don't break out just the subset 771 00:38:46,810 --> 00:38:48,879 Speaker 2: of that. But yeah, we, we, we, you know, we 772 00:38:48,879 --> 00:38:52,479 Speaker 2: have teams that are working on, you know, uh, renewable 773 00:38:52,479 --> 00:38:55,219 Speaker 2: energy across, you know, Amazon as well as across AWS 774 00:38:55,219 --> 00:38:55,589 Speaker 2: as well. 775 00:38:55,639 --> 00:38:56,080 Speaker 1: So you're 776 00:38:56,080 --> 00:38:57,989 Speaker 1: traveling around the region a lot. You mentioned trips to 777 00:38:57,989 --> 00:39:00,840 Speaker 1: Indonesia and Vietnam recently, you've been to Malaysia to to. 778 00:39:00,899 --> 00:39:04,209 Speaker 1: What's your sense in the region? Are we connecting the 779 00:39:04,209 --> 00:39:08,689 Speaker 1: renewables need with the big likely pick up in data 780 00:39:08,689 --> 00:39:09,770 Speaker 1: centers and all these things? Yeah, 781 00:39:09,889 --> 00:39:11,290 Speaker 2: I think I see it more and more. I kind 782 00:39:11,290 --> 00:39:14,250 Speaker 2: of mentioned just the example of PLN in Indonesia where 783 00:39:14,250 --> 00:39:19,000 Speaker 2: we've got 4 renewable projects running there. So, so in fact, 784 00:39:19,489 --> 00:39:22,389 Speaker 2: you know, just recently as well, we, we announced that 785 00:39:22,389 --> 00:39:26,610 Speaker 2: it's a public announcement with Petronas around so the Gentari 786 00:39:26,610 --> 00:39:27,449 Speaker 2: renewable energy. 787 00:39:27,889 --> 00:39:30,340 Speaker 2: Uh, business unit, how again we were procuring, you know, 788 00:39:30,419 --> 00:39:34,780 Speaker 2: green energy from Genttare as well. So, uh, so yeah, there's, there's, there's, 789 00:39:34,840 --> 00:39:37,290 Speaker 2: I'm definitely seeing, you know, the flywheel, you know, within 790 00:39:37,290 --> 00:39:40,540 Speaker 2: ASEAN and so on, uh, really help, you know, fuel 791 00:39:40,540 --> 00:39:43,889 Speaker 2: and feed the renewable energy into into data centers. 792 00:39:45,500 --> 00:39:48,419 Speaker 1: Personally sort of riveted by this, this dynamic because when 793 00:39:48,419 --> 00:39:51,219 Speaker 1: I see particularly in southern part of Malaysia, lots of 794 00:39:51,219 --> 00:39:54,179 Speaker 1: data centers coming up and I think the one refrain 795 00:39:54,179 --> 00:39:55,590 Speaker 1: I hear from Malaysia is that the 796 00:39:56,360 --> 00:39:58,860 Speaker 1: AWS of the world would not invest in them unless 797 00:39:58,860 --> 00:40:02,379 Speaker 1: they have a renewable uh sort of backbone to that. 798 00:40:02,780 --> 00:40:05,830 Speaker 1: And so, so no, keep leading by example. Thank you, 799 00:40:05,860 --> 00:40:09,569 Speaker 1: thank you. Um, you mentioned the word financial inclusion earlier, 800 00:40:09,659 --> 00:40:12,339 Speaker 1: and I think once I spoke with you before, you 801 00:40:12,340 --> 00:40:17,830 Speaker 1: had also used synonymously democratization of AI. So what 802 00:40:18,219 --> 00:40:20,979 Speaker 1: Doesn't mean to you both the inclusion and the democratization 803 00:40:20,979 --> 00:40:21,929 Speaker 1: aspect. Yeah, 804 00:40:22,100 --> 00:40:24,330 Speaker 2: no thanks, yeah, I touched on it just a little 805 00:40:24,330 --> 00:40:27,600 Speaker 2: bit there, I guess we're thinking about how we, you know, provide, 806 00:40:27,739 --> 00:40:31,330 Speaker 2: you know, the broadest access to to to our services and, 807 00:40:31,379 --> 00:40:33,699 Speaker 2: you know, probably firstly kind of going back to the 808 00:40:33,699 --> 00:40:35,879 Speaker 2: regional investments that we're making clearly that, you know, that 809 00:40:35,879 --> 00:40:39,820 Speaker 2: provides the foundations and the services that that our customers 810 00:40:39,820 --> 00:40:43,010 Speaker 2: and that nations can can help access and build with. 811 00:40:43,459 --> 00:40:45,830 Speaker 2: And so, um, you know, on top of that, of course, 812 00:40:45,889 --> 00:40:49,229 Speaker 2: we really want to provide uh knowledge and skills. So 813 00:40:49,229 --> 00:40:53,429 Speaker 2: we have over 500 different courses now available through something 814 00:40:53,429 --> 00:40:56,908 Speaker 2: called SkillsBuilder. These are free online digital courses. These are 815 00:40:56,909 --> 00:41:01,870 Speaker 2: available multiple languages, Bahasa Indonesian, Vietnamese, Thai, etc. So again, 816 00:41:01,949 --> 00:41:05,429 Speaker 2: we're really trying to, you know, up level up level skills. 817 00:41:05,820 --> 00:41:08,110 Speaker 2: Um, I touched on a couple of kind of startup 818 00:41:08,110 --> 00:41:10,830 Speaker 2: examples earlier, but we're really focused on helping, you know, 819 00:41:10,949 --> 00:41:13,109 Speaker 2: startups build and grow in the region. We have a 820 00:41:13,110 --> 00:41:18,270 Speaker 2: program specifically dedicated to invest in startups called Activate. Activate 821 00:41:18,270 --> 00:41:23,540 Speaker 2: provides skills, learning, investments to help startups bootstrap their, their, uh, 822 00:41:23,550 --> 00:41:26,469 Speaker 2: early journey to the cloud. So, uh, so we're really 823 00:41:26,469 --> 00:41:28,790 Speaker 2: just trying to find ways to, uh, to really sort 824 00:41:28,790 --> 00:41:31,790 Speaker 2: of fuel skills and, you know, provide broad access. 825 00:41:32,330 --> 00:41:35,250 Speaker 2: Uh, and then I, I touched briefly, um, earlier just 826 00:41:35,250 --> 00:41:37,928 Speaker 2: on a mentioned Kuiper, our low earth orbit, um. 827 00:41:38,800 --> 00:41:42,530 Speaker 2: Satellite network and again, really this is part of the 828 00:41:42,530 --> 00:41:45,929 Speaker 2: story as well as we think particularly about Southeast Asia 829 00:41:45,929 --> 00:41:51,129 Speaker 2: and some of the broad archipelago geography and providing digital 830 00:41:51,129 --> 00:41:54,370 Speaker 2: connectivity as well across that and Kuiper's going to do that, 831 00:41:54,479 --> 00:41:57,439 Speaker 2: you know, and provide a way to help people access 832 00:41:57,439 --> 00:42:01,250 Speaker 2: digital services wherever they are. Uh, and so I think 833 00:42:01,250 --> 00:42:04,790 Speaker 2: that's going to also, you know, really just provide, uh, 834 00:42:05,010 --> 00:42:07,399 Speaker 2: you know, a different level of opportunity for 835 00:42:07,989 --> 00:42:11,330 Speaker 2: Uh, individuals to participate in the digital economy to to 836 00:42:11,330 --> 00:42:15,408 Speaker 2: really be able to, you know, access intelligence as well 837 00:42:15,409 --> 00:42:18,610 Speaker 2: through generative AI and the genetic AI and models that 838 00:42:18,929 --> 00:42:23,489 Speaker 2: that probably, you know, just would have been prohibitive before. So, so, uh, 839 00:42:23,610 --> 00:42:26,260 Speaker 2: so Kuiper's kind of part of the story, uh, from, 840 00:42:26,330 --> 00:42:28,419 Speaker 2: from a connectivity perspective and then 841 00:42:28,770 --> 00:42:32,689 Speaker 2: When we think about also thinking about generative AI within 842 00:42:32,689 --> 00:42:35,639 Speaker 2: that context as well, you know, we, we're, as I 843 00:42:35,639 --> 00:42:37,959 Speaker 2: was describing with Bedrock, we really want to make sure 844 00:42:37,959 --> 00:42:40,899 Speaker 2: that then when customers are accessing those services, they've got 845 00:42:41,330 --> 00:42:44,120 Speaker 2: a broad set of models to to access and use. They've, 846 00:42:44,169 --> 00:42:48,449 Speaker 2: we've just actually uh I think you mentioned uh Deepeek earlier, 847 00:42:48,530 --> 00:42:52,049 Speaker 2: but we've just also just uh very recently added Deepeek's 848 00:42:52,050 --> 00:42:56,070 Speaker 2: 67 billion parameter chat model into Bedrock as well. We've 849 00:42:56,070 --> 00:42:56,570 Speaker 2: just added. 850 00:42:56,949 --> 00:43:00,580 Speaker 2: Open AIs, open weight models into bedrock as well. So, 851 00:43:00,909 --> 00:43:04,909 Speaker 2: so again, providing you know the broadest and deepest set 852 00:43:04,909 --> 00:43:09,350 Speaker 2: of capabilities for for customers will also help with that democratization. 853 00:43:09,389 --> 00:43:12,110 Speaker 2: So yeah, we really want to, um, you know, ensure 854 00:43:12,110 --> 00:43:15,790 Speaker 2: that customers can get easily connected digitally, can have access 855 00:43:15,790 --> 00:43:19,110 Speaker 2: to training and skills, can have access to funding programs 856 00:43:19,110 --> 00:43:21,590 Speaker 2: if they're startups, and then, you know, once they're using 857 00:43:21,590 --> 00:43:22,469 Speaker 2: the technology. 858 00:43:22,949 --> 00:43:26,399 Speaker 2: Um, really have the best of open source and first 859 00:43:26,399 --> 00:43:29,560 Speaker 2: party technologies. And so, you know, there's a lot happening 860 00:43:29,560 --> 00:43:32,600 Speaker 2: in the open source space around Agente AI and we 861 00:43:32,600 --> 00:43:35,219 Speaker 2: support all of the key frameworks like Strands, which is 862 00:43:35,219 --> 00:43:39,750 Speaker 2: a framework for building agente AI and model context protocol, 863 00:43:39,760 --> 00:43:43,159 Speaker 2: all of these, all of these, uh, types of open 864 00:43:43,159 --> 00:43:47,399 Speaker 2: source frameworks. AWS has historically always embraced open source and 865 00:43:47,399 --> 00:43:50,560 Speaker 2: that kind of continues in this Gen AIgente AI era 866 00:43:50,560 --> 00:43:51,319 Speaker 2: as well. So. 867 00:43:51,620 --> 00:43:55,399 Speaker 2: So that's kind of how we're thinking about democratization and access. 868 00:43:55,810 --> 00:43:59,600 Speaker 1: When you were earlier talking about Ke showcasing their KAI, 869 00:43:59,649 --> 00:44:02,159 Speaker 1: I was thinking that if indeed they managed to cut 870 00:44:02,159 --> 00:44:05,569 Speaker 1: weeks if not months, from preparatory work when they're looking 871 00:44:05,570 --> 00:44:09,540 Speaker 1: at projects to invest in, wouldn't that also mean cutting 872 00:44:09,540 --> 00:44:12,389 Speaker 1: a bunch of young trainees who normally do that grunt 873 00:44:12,389 --> 00:44:14,090 Speaker 1: work and their jobs would disappear? 874 00:44:14,929 --> 00:44:18,229 Speaker 2: I think, look, it's, it's a really key question and 875 00:44:18,229 --> 00:44:20,790 Speaker 2: you know, I think an important topic to discuss, and 876 00:44:20,790 --> 00:44:23,739 Speaker 2: I think um firstly, I would say I do think, 877 00:44:23,750 --> 00:44:25,669 Speaker 2: and you're probably seeing it as well, you know, here 878 00:44:25,669 --> 00:44:28,750 Speaker 2: in DBS, you know, some of the nature of work is, is, 879 00:44:28,820 --> 00:44:30,919 Speaker 2: is training and is changing, sorry, and 880 00:44:31,620 --> 00:44:33,929 Speaker 2: You know, without, without, without actually being flippant, but there 881 00:44:33,929 --> 00:44:37,340 Speaker 2: was one anecdote that I just wanted to share. I 882 00:44:37,340 --> 00:44:41,250 Speaker 2: saw K Supachai, the chair of CP Group, just recently 883 00:44:41,250 --> 00:44:44,689 Speaker 2: in a, in a podcast type type video recording talking 884 00:44:44,689 --> 00:44:47,679 Speaker 2: about generative AI, and he was asked a similar question 885 00:44:47,679 --> 00:44:50,729 Speaker 2: and he said, I think the first person who will 886 00:44:50,729 --> 00:44:52,850 Speaker 2: lose their job will be me. It will be the CEO. 887 00:44:53,270 --> 00:44:56,820 Speaker 2: And he said, because we're, you know, we're democratizing access 888 00:44:56,820 --> 00:45:00,489 Speaker 2: to intelligence, everyone's got the same information that CEOs previously had. 889 00:45:00,570 --> 00:45:03,489 Speaker 2: Everyone's able to pull together agents to do things more 890 00:45:03,489 --> 00:45:04,529 Speaker 2: effectively and so. 891 00:45:05,260 --> 00:45:07,300 Speaker 2: So I think, you know, there's a grain of reality 892 00:45:07,300 --> 00:45:09,699 Speaker 2: and what he's saying as well just around, I think, 893 00:45:09,709 --> 00:45:13,138 Speaker 2: as we're thinking about, you know, democratizing technologies, but within 894 00:45:13,139 --> 00:45:16,540 Speaker 2: an organization, you know, having a broader set of access 895 00:45:16,540 --> 00:45:20,459 Speaker 2: to information, enabling people to maybe use their creativity and 896 00:45:20,459 --> 00:45:23,138 Speaker 2: their critical thinking in ways that they might have not 897 00:45:23,139 --> 00:45:25,780 Speaker 2: been able to before. So I think there's, you know, 898 00:45:25,870 --> 00:45:29,459 Speaker 2: there's definitely an era where we'll see sort of roles, 899 00:45:29,520 --> 00:45:31,219 Speaker 2: you know, evolve and change and 900 00:45:31,389 --> 00:45:33,949 Speaker 2: I think, you know, historically, as we've seen with any 901 00:45:33,949 --> 00:45:37,479 Speaker 2: technology era, uh, what we've seen as the outcome has been, 902 00:45:37,580 --> 00:45:40,219 Speaker 2: you know, a far greater sort of set of opportunities 903 00:45:40,219 --> 00:45:42,899 Speaker 2: for individuals. And if I think about, you know, my 904 00:45:42,899 --> 00:45:46,020 Speaker 2: own career and I think about probably my, my, my parents, um, 905 00:45:46,139 --> 00:45:48,260 Speaker 2: who were whatever came before Gen X, but they were, 906 00:45:48,340 --> 00:45:50,149 Speaker 2: you know, Gen W if that exists, but. 907 00:45:50,510 --> 00:45:53,939 Speaker 2: But they, uh, they could have never imagined probably my 908 00:45:53,939 --> 00:45:57,120 Speaker 2: career and the jobs that exist around the the tech 909 00:45:57,120 --> 00:45:59,120 Speaker 2: industry and so on. So I feel that, you know, 910 00:45:59,239 --> 00:46:02,199 Speaker 2: we're probably, uh, you know, the beginning of a of 911 00:46:02,199 --> 00:46:04,320 Speaker 2: an era like that. If I, if I maybe just 912 00:46:04,320 --> 00:46:06,639 Speaker 2: reflect on within Amazon what we're kind of seeing, you know, 913 00:46:06,679 --> 00:46:08,360 Speaker 2: we see probably over 1000 different. 914 00:46:08,949 --> 00:46:11,830 Speaker 2: Use cases of Gen AI and some of these things emerging, 915 00:46:11,899 --> 00:46:15,669 Speaker 2: be it how Alexa is responding to customers and answering 916 00:46:15,669 --> 00:46:19,429 Speaker 2: questions or be it how Amazon fulfillment centers are using 917 00:46:19,429 --> 00:46:21,669 Speaker 2: some of these technologies to, to, you know, to manage 918 00:46:21,669 --> 00:46:27,300 Speaker 2: inventory or how on our Amazon.com retail platforms we're helping 919 00:46:27,590 --> 00:46:31,000 Speaker 2: third party sellers build advertising and and uh you know, 920 00:46:31,310 --> 00:46:34,110 Speaker 2: content more quickly using tool sets or. 921 00:46:34,590 --> 00:46:37,830 Speaker 2: You know, in more customer facing teams in AWS, whether it's, 922 00:46:37,919 --> 00:46:41,750 Speaker 2: you know, uh, you know, helping forecasting the business and 923 00:46:41,760 --> 00:46:45,919 Speaker 2: and summarizing information, documentation. So there's, there's, I think, and 924 00:46:45,919 --> 00:46:48,600 Speaker 2: I'm sure as you know, within your business and other 925 00:46:48,600 --> 00:46:50,840 Speaker 2: customers you meet with sort of seeing all of these 926 00:46:50,840 --> 00:46:54,070 Speaker 2: kind of areas of, of, of, you know, productivity, uh, 927 00:46:54,120 --> 00:46:55,850 Speaker 2: begin to emerge, but, but, um. 928 00:46:56,250 --> 00:46:58,370 Speaker 2: I think that, you know, the, as, as we kind 929 00:46:58,370 --> 00:47:00,090 Speaker 2: of move forward, I think some of the, you know, 930 00:47:00,169 --> 00:47:02,969 Speaker 2: the facts remain around. I, I think of um also 931 00:47:02,969 --> 00:47:06,888 Speaker 2: analogies like when the calculator was invented, you still need 932 00:47:06,889 --> 00:47:08,889 Speaker 2: to understand how to do maths to make use of 933 00:47:08,889 --> 00:47:10,889 Speaker 2: the calculator and I think some of these things really 934 00:47:10,889 --> 00:47:14,729 Speaker 2: apply to generative AI and agentic AI that uh that 935 00:47:14,729 --> 00:47:18,850 Speaker 2: ability to, yeah, to think critically, to understand how to 936 00:47:18,850 --> 00:47:20,770 Speaker 2: how to make use of these technologies, but 937 00:47:21,209 --> 00:47:23,360 Speaker 2: But you know, I, I, you know, I kind of, um, 938 00:47:23,770 --> 00:47:26,290 Speaker 2: you know, remain sort of excited about just I think 939 00:47:26,290 --> 00:47:30,040 Speaker 2: the opportunity for reaching growth that it provides organizations where, 940 00:47:30,780 --> 00:47:33,189 Speaker 2: You know, most people are, and I, I love the 941 00:47:33,189 --> 00:47:35,620 Speaker 2: example from, from your son. I'll use an example of 942 00:47:35,620 --> 00:47:38,060 Speaker 2: actually related to my son, but I was, we were, 943 00:47:38,179 --> 00:47:40,580 Speaker 2: we were talking about some sporting things earlier and I 944 00:47:40,580 --> 00:47:43,540 Speaker 2: was actually just looking at arranging his flights, uh, to, 945 00:47:43,610 --> 00:47:47,020 Speaker 2: to come out to, to Singapore and um I thought 946 00:47:47,020 --> 00:47:49,899 Speaker 2: I'd use agentic AI tool just to go off and 947 00:47:49,899 --> 00:47:52,939 Speaker 2: do all the website comparisons and I was watching the 948 00:47:52,939 --> 00:47:54,979 Speaker 2: Agentic tool just go off, fill in all the forms, 949 00:47:55,060 --> 00:47:57,729 Speaker 2: do all the searching. I was meanwhile doing some work, 950 00:47:57,820 --> 00:48:00,500 Speaker 2: you know, 30 minutes later it had completed its analysis. 951 00:48:01,260 --> 00:48:03,699 Speaker 2: I've been quite productive doing some work and then it 952 00:48:03,699 --> 00:48:05,500 Speaker 2: was like, OK, it's that flight on that date, it's 953 00:48:05,500 --> 00:48:07,780 Speaker 2: the best option. So, so I think there's some of 954 00:48:07,780 --> 00:48:12,379 Speaker 2: this kind of undifferentiated uh administrative type work that that 955 00:48:12,379 --> 00:48:15,020 Speaker 2: I think will change through and you know, we're seeing it, right? 956 00:48:15,060 --> 00:48:17,699 Speaker 2: I think often we see these things first in the 957 00:48:17,699 --> 00:48:20,790 Speaker 2: consumer environment as as well uh start to ripple through 958 00:48:20,790 --> 00:48:23,370 Speaker 2: into into and across enterprises, but. 959 00:48:23,639 --> 00:48:25,830 Speaker 2: I think ultimately, you know, we, we want to be 960 00:48:25,830 --> 00:48:29,029 Speaker 2: able to, you know, empower teams to be agile, you know, 961 00:48:29,350 --> 00:48:32,549 Speaker 2: quickly hustle around a problem, you know, build solutions, have 962 00:48:32,550 --> 00:48:35,388 Speaker 2: the skills to do that. So I think, I think 963 00:48:35,389 --> 00:48:38,510 Speaker 2: some of, um, a bit back to can superchat's example, right? 964 00:48:38,590 --> 00:48:40,989 Speaker 2: I think we'll see organization structures start to reflect that 965 00:48:40,989 --> 00:48:44,468 Speaker 2: a bit more as, as, um, as, as teams can, 966 00:48:44,510 --> 00:48:46,689 Speaker 2: you know, maybe uh make use of, make use of 967 00:48:46,689 --> 00:48:49,750 Speaker 2: agents to, to execute some tasks and take them into 968 00:48:49,750 --> 00:48:50,750 Speaker 2: a new market and. 969 00:48:51,350 --> 00:48:54,020 Speaker 2: I think the uh the Techcom bank example that I 970 00:48:54,020 --> 00:48:56,320 Speaker 2: mentioned just from from Hanoi last week, I think was, 971 00:48:56,379 --> 00:48:58,010 Speaker 2: you know, which is very interesting how 972 00:48:58,850 --> 00:49:00,770 Speaker 2: You know, they're thinking of how, you know, there's an 973 00:49:00,770 --> 00:49:03,800 Speaker 2: envelope around the the resource of their organization, but an 974 00:49:03,800 --> 00:49:05,879 Speaker 2: ambition around their growth and how do they get to 975 00:49:05,879 --> 00:49:11,620 Speaker 2: that ambition by, by, uh, you know, looking at scaling mechanisms. So, uh, um, 976 00:49:12,080 --> 00:49:13,969 Speaker 2: so yeah, I, you know, I, I, I, I think, 977 00:49:14,189 --> 00:49:16,359 Speaker 2: as I say, I think it's the the the the dawn, 978 00:49:16,399 --> 00:49:19,239 Speaker 2: the beginning of the beginning of a of an era 979 00:49:19,239 --> 00:49:21,800 Speaker 2: that we're in, which is, is gonna be super exciting 980 00:49:21,800 --> 00:49:22,750 Speaker 2: to see play out. 981 00:49:22,959 --> 00:49:25,550 Speaker 1: So, beginning of the beginning, indeed. So now, I normally 982 00:49:25,800 --> 00:49:28,199 Speaker 1: don't leave the hardest question for the last, but I will. 983 00:49:29,239 --> 00:49:33,830 Speaker 1: Give you that uh challenge. Um, so MIT NANDA project 984 00:49:33,830 --> 00:49:36,479 Speaker 1: a couple of weeks ago released this report where they 985 00:49:36,479 --> 00:49:39,399 Speaker 1: interviewed lots and lots of enterprises in the US and 986 00:49:39,399 --> 00:49:42,719 Speaker 1: the some some basic takeaways are what we're discussing that, 987 00:49:42,760 --> 00:49:45,870 Speaker 1: you know, there's a lot of enthusiasm when you do 988 00:49:46,110 --> 00:49:50,520 Speaker 1: corporate um one on one interviews on efficiency enhancement, you 989 00:49:50,520 --> 00:49:53,199 Speaker 1: get a lot of evidence of workflows being improved and 990 00:49:53,199 --> 00:49:53,520 Speaker 1: so on. 991 00:49:53,889 --> 00:49:57,379 Speaker 1: But when this report tried to establish direct, you know, 992 00:49:57,610 --> 00:50:02,610 Speaker 1: attribution to P&L, most companies said not yet. We're going 993 00:50:02,610 --> 00:50:05,569 Speaker 1: through the process, it's not infused enough in the organization 994 00:50:05,570 --> 00:50:07,770 Speaker 1: to be able to actually see the dollars and cents evidence. 995 00:50:08,409 --> 00:50:12,879 Speaker 1: Do you find that too conservative an approach, and are you, 996 00:50:12,889 --> 00:50:15,530 Speaker 1: in your experience through your travel and through your client engagement, 997 00:50:15,689 --> 00:50:18,000 Speaker 1: seeing something more encouraging than that? Yeah, 998 00:50:18,290 --> 00:50:20,448 Speaker 2: yeah, I would say yes, I would say I am 999 00:50:20,449 --> 00:50:24,010 Speaker 2: seeing two parts to how I'd answer that. I'd say, firstly, 1000 00:50:24,070 --> 00:50:27,100 Speaker 2: I am seeing some, you know, more positive examples of 1001 00:50:27,100 --> 00:50:28,959 Speaker 2: that for sure. I think probably just the 1002 00:50:29,419 --> 00:50:32,669 Speaker 2: Example, I referenced with Telecome and their customer service and 1003 00:50:32,679 --> 00:50:35,759 Speaker 2: and being able to completely compress the, you know, the 1004 00:50:35,760 --> 00:50:38,600 Speaker 2: time to to service and resolve issues and so on 1005 00:50:38,600 --> 00:50:43,080 Speaker 2: that's resulting in, you know, improved customer satisfaction or or 1006 00:50:43,080 --> 00:50:46,919 Speaker 2: how Commonwealth Bank Australia are making 55 million AI driven 1007 00:50:46,919 --> 00:50:48,639 Speaker 2: decisions on a daily basis. So, you know, some of 1008 00:50:48,639 --> 00:50:52,280 Speaker 2: these things are, you know, operating and executing at scale 1009 00:50:52,280 --> 00:50:54,280 Speaker 2: that is, is, is having an impact in a, in 1010 00:50:54,280 --> 00:50:56,010 Speaker 2: a business for sure already, I think. 1011 00:50:56,489 --> 00:50:59,449 Speaker 2: I think, um, you know, or even if I look at, um, 1012 00:50:59,629 --> 00:51:02,449 Speaker 2: you know, perhaps the area around software development, um, you know, 1013 00:51:02,489 --> 00:51:06,000 Speaker 2: it's super interesting just what's been happening in the ability 1014 00:51:06,000 --> 00:51:09,500 Speaker 2: to produce source code more quickly and uh I think 1015 00:51:09,500 --> 00:51:11,649 Speaker 2: uh just some work that we did within Amazon where 1016 00:51:11,649 --> 00:51:15,649 Speaker 2: we were looking to upgrade some legacy Java based code 1017 00:51:15,649 --> 00:51:18,629 Speaker 2: environments and and so uh we were able to use 1018 00:51:18,629 --> 00:51:20,449 Speaker 2: our agentic AI tools to 1019 00:51:21,199 --> 00:51:26,359 Speaker 2: 450 developer years of of upgrade time in doing that. 1020 00:51:26,389 --> 00:51:30,310 Speaker 2: And so, and that was undifferentiated work moving it from 1021 00:51:30,310 --> 00:51:33,080 Speaker 2: Java version X to Java version Y and testing it 1022 00:51:33,080 --> 00:51:35,759 Speaker 2: and doing all of those sort of things. So, so 1023 00:51:35,760 --> 00:51:38,198 Speaker 2: I think what's kind of interesting, I think in this 1024 00:51:38,199 --> 00:51:41,350 Speaker 2: space as well is, is a lot of organizations are 1025 00:51:41,689 --> 00:51:44,820 Speaker 2: Um, actually weighed down by technology in the sense of 1026 00:51:44,820 --> 00:51:49,030 Speaker 2: there's probably 2 or 3 decades of technology debt systems 1027 00:51:49,030 --> 00:51:51,290 Speaker 2: that have been built and are hard to change, expensive 1028 00:51:51,290 --> 00:51:53,429 Speaker 2: to change, and so on, and you know, on premise. 1029 00:51:54,229 --> 00:51:58,139 Speaker 2: But actually, what, what, uh, generative AI and agentic AI is, 1030 00:51:58,189 --> 00:52:01,189 Speaker 2: is enabling organizations to do is to begin to point 1031 00:52:01,189 --> 00:52:03,270 Speaker 2: agents to look at, well, can I migrate that and 1032 00:52:03,270 --> 00:52:06,560 Speaker 2: build new software and change that mainframes, these sorts of 1033 00:52:07,040 --> 00:52:10,279 Speaker 2: Legacy systems. So again, we're we're seeing actually we've launched 1034 00:52:10,280 --> 00:52:13,678 Speaker 2: a product called Transform that's exactly aimed at that, so 1035 00:52:13,679 --> 00:52:17,439 Speaker 2: taking sort of legacy environments that maybe, you know, the 1036 00:52:17,439 --> 00:52:20,429 Speaker 2: people that wrote the code left the organization 20 years ago, 1037 00:52:21,040 --> 00:52:23,439 Speaker 2: but the agents can go in and analyze it and 1038 00:52:23,439 --> 00:52:25,439 Speaker 2: kind of understand the plan and then build, you know, 1039 00:52:25,520 --> 00:52:27,669 Speaker 2: a new version of that in an open source stack 1040 00:52:27,669 --> 00:52:30,959 Speaker 2: and and modernize it. So, so I think, I think 1041 00:52:30,959 --> 00:52:32,159 Speaker 2: just that, um. 1042 00:52:32,889 --> 00:52:35,439 Speaker 2: That weight if you like, of technology burden as well. 1043 00:52:35,469 --> 00:52:38,129 Speaker 2: We're seeing organizations begin to lift and then probably the 1044 00:52:38,129 --> 00:52:41,330 Speaker 2: third point I'd say is there's just this point of 1045 00:52:41,330 --> 00:52:44,050 Speaker 2: scalability that I think we're just beginning to see and 1046 00:52:44,050 --> 00:52:46,689 Speaker 2: just beginning to see organizations start to think, how do 1047 00:52:46,689 --> 00:52:47,129 Speaker 2: I 1048 00:52:47,790 --> 00:52:50,229 Speaker 2: How they are building these, these, these things with us, 1049 00:52:50,239 --> 00:52:51,909 Speaker 2: you know, how can I create, you know, hundreds of 1050 00:52:51,909 --> 00:52:53,989 Speaker 2: use cases that are going to have a kind of 1051 00:52:53,989 --> 00:52:57,549 Speaker 2: a scaled impact across across the business and then on 1052 00:52:57,550 --> 00:52:59,549 Speaker 2: top of that, how can I begin to think about 1053 00:52:59,550 --> 00:53:03,149 Speaker 2: maybe agents as an extensibility point. But uh I was 1054 00:53:03,149 --> 00:53:06,250 Speaker 2: with a banking customer in Thailand a couple of weeks ago. 1055 00:53:06,550 --> 00:53:08,949 Speaker 2: He said they write 100 million lines of code each year. 1056 00:53:09,030 --> 00:53:12,070 Speaker 2: They're writing 10 million now using generative AI so. 1057 00:53:12,560 --> 00:53:15,138 Speaker 2: But you know, they, they, their development team are now 1058 00:53:15,139 --> 00:53:18,138 Speaker 2: just creating all sorts of new products and he was 1059 00:53:18,139 --> 00:53:20,379 Speaker 2: talking about a, you know, a new product for helping 1060 00:53:20,699 --> 00:53:23,179 Speaker 2: customers choose credit cards and so on based on an 1061 00:53:23,179 --> 00:53:25,939 Speaker 2: agentic approach and so on. So I think again, that 1062 00:53:25,939 --> 00:53:28,339 Speaker 2: ability of, you know, maybe some of the legacy type, 1063 00:53:28,459 --> 00:53:31,020 Speaker 2: you know, technology work that can be modernized and have 1064 00:53:31,020 --> 00:53:35,649 Speaker 2: the technology teams focused on some really kind of business impacting, uh, 1065 00:53:35,780 --> 00:53:37,260 Speaker 2: you know, capability areas. 1066 00:53:37,729 --> 00:53:40,370 Speaker 2: Uh, and then seeing scale. So, so I read that 1067 00:53:40,370 --> 00:53:42,770 Speaker 2: report with great interest as well, and, and, you know, 1068 00:53:42,889 --> 00:53:45,250 Speaker 2: there's there's definitely elements of it, of course, I think 1069 00:53:45,250 --> 00:53:48,489 Speaker 2: we've seen a lot of productivity enhancements, you know, within 1070 00:53:48,489 --> 00:53:52,129 Speaker 2: an organization or people building, you know, maybe point solutions 1071 00:53:52,129 --> 00:53:55,649 Speaker 2: for generative AI to solve a task to help, um, no, 1072 00:53:55,739 --> 00:53:58,560 Speaker 2: I don't know, maybe improve fraud detection in an area 1073 00:53:58,560 --> 00:54:00,609 Speaker 2: rather than these kind of broader end to end. 1074 00:54:01,229 --> 00:54:03,790 Speaker 2: Um, more what I would describe as AI native use 1075 00:54:03,790 --> 00:54:06,989 Speaker 2: cases that are, you know, kind of reimagining how to build, 1076 00:54:07,030 --> 00:54:09,590 Speaker 2: and so that's, uh, I think some of what we're 1077 00:54:09,590 --> 00:54:10,989 Speaker 2: seeing more and more of and 1078 00:54:11,669 --> 00:54:12,658 Speaker 1: so sooner than later. 1079 00:54:13,229 --> 00:54:15,229 Speaker 2: Yeah, yeah, no, I think, I think, you know, now 1080 00:54:15,229 --> 00:54:17,790 Speaker 2: in terms of just some of the, you know, the 1081 00:54:17,790 --> 00:54:20,949 Speaker 2: customers and the impacts that we're seeing for this, uh, so. 1082 00:54:21,469 --> 00:54:26,040 Speaker 2: Uh, yeah, and a very, uh, very, um, yeah, I 1083 00:54:26,040 --> 00:54:28,149 Speaker 2: think I think what's happened probably from the era of 1084 00:54:28,149 --> 00:54:31,429 Speaker 2: generative AI last year and the year before to agentic 1085 00:54:31,429 --> 00:54:33,790 Speaker 2: AI now is just completely again. 1086 00:54:34,760 --> 00:54:37,169 Speaker 2: Put a different sort of curve in terms of what's 1087 00:54:37,169 --> 00:54:41,229 Speaker 2: possible with with this technology, so. 1088 00:54:42,520 --> 00:54:45,149 Speaker 2: Yeah, I couldn't be, you know, more excited to to 1089 00:54:45,149 --> 00:54:46,300 Speaker 2: spend some time talking to you about it. 1090 00:54:46,600 --> 00:54:50,629 Speaker 1: It's infectious. You have certainly, you know, added to my enthusiasm. Jeff, 1091 00:54:50,909 --> 00:54:53,589 Speaker 1: thank you so much for your time and insights. Great. Thanks, 1092 00:54:53,709 --> 00:54:55,429 Speaker 2: Tam. Thanks so much for having me. It's been great 1093 00:54:55,429 --> 00:54:55,510 Speaker 2: to 1094 00:54:55,510 --> 00:54:57,550 Speaker 1: be here. Great to have you and thanks to our 1095 00:54:57,550 --> 00:55:00,550 Speaker 1: listeners as well. Kobe Time was produced by Ken Delbridge, 1096 00:55:00,709 --> 00:55:04,780 Speaker 1: Violet Lee and Daisy Sharma provided additional assistance. All 161 1097 00:55:04,780 --> 00:55:07,469 Speaker 1: episodes of this podcast are available on YouTube as well 1098 00:55:07,469 --> 00:55:09,189 Speaker 1: as on Apple, Google, and Spotify. 1099 00:55:09,550 --> 00:55:13,449 Speaker 1: This is for information only, does not constitute any investment 1100 00:55:13,449 --> 00:55:16,370 Speaker 1: advice and as far as DBS research and live streams 1101 00:55:16,370 --> 00:55:19,459 Speaker 1: are concerned, you can find them all by Googling DBS 1102 00:55:19,459 --> 00:55:21,320 Speaker 1: Research Library. Have a great day.