1 00:00:00,120 --> 00:00:04,280 Speaker 1: Joining us now. Amazon Web Services CEO Adam Selipski and 2 00:00:04,320 --> 00:00:06,800 Speaker 1: Adam welcome to Bloomberg Technology. I want to start with 3 00:00:06,840 --> 00:00:10,760 Speaker 1: some of the mechanics of this deal. Does Anthropic still 4 00:00:10,880 --> 00:00:14,960 Speaker 1: pay Amazon to use AWS Cloud or is it structured 5 00:00:15,040 --> 00:00:18,680 Speaker 1: such that the investment that you make in Anthropic is 6 00:00:18,680 --> 00:00:22,200 Speaker 1: in the form of cash and credits for AWS Cloud. 7 00:00:24,360 --> 00:00:27,400 Speaker 2: Good morning, Thanks for having me. We're very excited for 8 00:00:27,520 --> 00:00:33,239 Speaker 2: this expanded relationship with Anthropic and the investment is a 9 00:00:33,280 --> 00:00:39,600 Speaker 2: financial investment, as you say. And in addition, Anthropic will 10 00:00:39,640 --> 00:00:43,400 Speaker 2: be training future versions of its models and running its 11 00:00:43,440 --> 00:00:49,920 Speaker 2: models on AWS using our training chips and Inferentia chips. 12 00:00:50,520 --> 00:00:53,400 Speaker 2: Those models will be guaranteed to be available for years 13 00:00:53,440 --> 00:00:57,920 Speaker 2: to come in our Amazon Bedrock Managed service for llms, 14 00:00:57,960 --> 00:01:02,320 Speaker 2: which provides the very wide choice of models, and AWS 15 00:01:02,320 --> 00:01:06,280 Speaker 2: customers will actually receive early access to key features in 16 00:01:06,360 --> 00:01:09,880 Speaker 2: Anthropics models in the future, such as fine tuning and 17 00:01:09,959 --> 00:01:15,000 Speaker 2: customization of models. In addition, Anthropics are very talented technical 18 00:01:15,040 --> 00:01:19,200 Speaker 2: teams and we anticipate working closely with them to actually 19 00:01:19,240 --> 00:01:23,200 Speaker 2: improve future versions of our training in Inferentia at chips. 20 00:01:23,400 --> 00:01:26,640 Speaker 2: So there are a lot of different benefits for our 21 00:01:26,880 --> 00:01:30,120 Speaker 2: joint end customers. From this relationship and we're very excited 22 00:01:30,120 --> 00:01:31,240 Speaker 2: to be leaders in this together. 23 00:01:32,440 --> 00:01:35,119 Speaker 1: Adam, there's a lot of emphasis on moving a maker 24 00:01:35,160 --> 00:01:38,360 Speaker 1: of foundation models at that scale once your proprietary silicon. 25 00:01:38,800 --> 00:01:43,960 Speaker 1: How quickly will Anthropics start running AI workloads on trainium 26 00:01:43,959 --> 00:01:44,720 Speaker 1: and Inferentia. 27 00:01:45,880 --> 00:01:49,200 Speaker 2: Well, we've been working with Anthropic, they've been a customers 28 00:01:49,480 --> 00:01:52,560 Speaker 2: of ours since I think they're founding over a couple 29 00:01:52,600 --> 00:01:55,400 Speaker 2: of years ago, and so they use a variety of 30 00:01:55,400 --> 00:01:59,520 Speaker 2: different technologies for a variety of different workloads on AWS 31 00:02:02,760 --> 00:02:06,880 Speaker 2: that they'll be using GPUs on AWS and will also 32 00:02:06,960 --> 00:02:10,799 Speaker 2: be using large quantities of Trainingum and Inferentia. So I 33 00:02:10,880 --> 00:02:14,000 Speaker 2: think everything's going to move very quickly and it'll all 34 00:02:14,040 --> 00:02:16,840 Speaker 2: be a mix of technologies depending on their needs at 35 00:02:16,840 --> 00:02:17,160 Speaker 2: the time. 36 00:02:18,440 --> 00:02:22,000 Speaker 1: Adam, what's the mood like within Amazon and AWS this morning? 37 00:02:22,000 --> 00:02:25,440 Speaker 1: There are lots of talented engineers that have been working 38 00:02:25,480 --> 00:02:29,720 Speaker 1: on large language models generative AI tools internally, and now 39 00:02:29,760 --> 00:02:33,040 Speaker 1: you're turning to a third party who's highly regarded as 40 00:02:33,040 --> 00:02:35,400 Speaker 1: a leader in building foundation models. 41 00:02:37,400 --> 00:02:39,880 Speaker 2: The mood here is great. We are a company of 42 00:02:39,919 --> 00:02:42,880 Speaker 2: inventors who we love to build, and there's never been 43 00:02:42,919 --> 00:02:46,040 Speaker 2: a better time to be a builder at AWS. Than 44 00:02:46,160 --> 00:02:50,839 Speaker 2: right now. And as I mentioned before, a big part 45 00:02:50,880 --> 00:02:53,760 Speaker 2: of our strategy in AI and generative AI specifically, it 46 00:02:53,880 --> 00:02:56,960 Speaker 2: is all about customer choice and there's not going to 47 00:02:57,000 --> 00:03:01,080 Speaker 2: be any one solution that were for all customers for 48 00:03:01,120 --> 00:03:04,320 Speaker 2: all use cases. And Thropic has done an amazing job. 49 00:03:04,360 --> 00:03:08,040 Speaker 2: They're clearly a leader in this space, and it's really 50 00:03:08,080 --> 00:03:11,720 Speaker 2: important for customers that we continue to generate new capabilities 51 00:03:11,960 --> 00:03:15,800 Speaker 2: together at the same time. Really one of the hallmarks 52 00:03:15,800 --> 00:03:20,760 Speaker 2: of our Amazon Bedrock managed service for generative AI is choice, 53 00:03:20,800 --> 00:03:23,240 Speaker 2: and so Amazon is going to continue to build its 54 00:03:23,240 --> 00:03:26,919 Speaker 2: own Titan models which are going to be available later 55 00:03:27,000 --> 00:03:31,840 Speaker 2: this year. Obviously Onthropics models are prominent in Bedrock, and 56 00:03:31,880 --> 00:03:34,839 Speaker 2: we will have models from other leading providers as well 57 00:03:35,160 --> 00:03:37,960 Speaker 2: as we have today. So it's still an amazing time 58 00:03:38,320 --> 00:03:40,960 Speaker 2: to build here at Amazon. We think our models are 59 00:03:41,000 --> 00:03:42,840 Speaker 2: going to be great as well, and it's about customers 60 00:03:42,880 --> 00:03:44,760 Speaker 2: choosing the right tool for the job. 61 00:03:45,520 --> 00:03:48,200 Speaker 3: Talking about choice, and I just want to re welcome 62 00:03:48,240 --> 00:03:52,080 Speaker 3: our TV and radio audiances with Adam Sleipski. What's so 63 00:03:52,240 --> 00:03:55,680 Speaker 3: notable is that well Anthropic took a chunk of change 64 00:03:55,680 --> 00:04:00,440 Speaker 3: one hundred million dollars worth from Google already. What interested 65 00:04:00,480 --> 00:04:03,040 Speaker 3: as to how you feel that is perhaps a concern 66 00:04:03,080 --> 00:04:06,320 Speaker 3: for you or not the relationship that anthropag already has 67 00:04:06,920 --> 00:04:08,680 Speaker 3: with a previous cloud provider. 68 00:04:09,960 --> 00:04:13,440 Speaker 2: Now, we feel great about the relationship with Anthropic. It's 69 00:04:13,480 --> 00:04:16,359 Speaker 2: been a good relationship, and I think today's announcement just 70 00:04:16,440 --> 00:04:22,880 Speaker 2: makes it a deeper and longer term Anthropic will use 71 00:04:22,920 --> 00:04:26,760 Speaker 2: AWS as its primary cloud provider for mission critical workloads, 72 00:04:26,800 --> 00:04:33,200 Speaker 2: including building foundational foundation models and doing AI safety research, 73 00:04:33,480 --> 00:04:36,920 Speaker 2: and will run the majority of its workloads on AWS. 74 00:04:36,960 --> 00:04:39,800 Speaker 2: So we feel great about being able to provide the 75 00:04:39,839 --> 00:04:43,279 Speaker 2: capacity and the expertise, and of course the security, the 76 00:04:43,400 --> 00:04:47,120 Speaker 2: enterprise grade security that is so important to AWS customers. 77 00:04:47,279 --> 00:04:49,680 Speaker 2: And we also feel great about working with Anthropic to 78 00:04:49,720 --> 00:04:54,039 Speaker 2: make sure that our trainum and inferential technology, our chips 79 00:04:54,120 --> 00:04:57,840 Speaker 2: are as cutting edge as possible going forward for years 80 00:04:57,839 --> 00:04:58,200 Speaker 2: to come. 81 00:04:58,880 --> 00:05:01,280 Speaker 3: I'm interested in drilling down sort of what ED was 82 00:05:01,320 --> 00:05:04,200 Speaker 3: going about the feeling internally right now, because I look 83 00:05:04,200 --> 00:05:06,880 Speaker 3: at some of the analyst reaction to this. Adam and Webbush, 84 00:05:06,880 --> 00:05:10,520 Speaker 3: for example, they say this signals a newfound urgency in 85 00:05:10,600 --> 00:05:13,840 Speaker 3: Amazon's strategy to further integrate generative AI among your AW 86 00:05:14,080 --> 00:05:17,760 Speaker 3: suite of services. That urgency was there a lack of 87 00:05:17,839 --> 00:05:21,919 Speaker 3: understanding or indeed a reality that Amazon was behind the 88 00:05:22,000 --> 00:05:24,839 Speaker 3: curve here a little bit when it came to the 89 00:05:24,880 --> 00:05:27,479 Speaker 3: integration of generative AI. Because we've been looking at open 90 00:05:27,520 --> 00:05:29,160 Speaker 3: Ai and Microsoft for a while now. 91 00:05:30,240 --> 00:05:33,359 Speaker 2: We've been saying for many, many months, Carolyn, that we 92 00:05:33,440 --> 00:05:36,159 Speaker 2: are fully urgent. We have a strategy that we really love. 93 00:05:36,520 --> 00:05:39,240 Speaker 2: It is different than some other cloud provider strategies. It's 94 00:05:39,279 --> 00:05:45,680 Speaker 2: true we have a strategy of providing absolutely uncompromising security, 95 00:05:46,040 --> 00:05:48,279 Speaker 2: which I don't think is true for our cloud providers. 96 00:05:48,480 --> 00:05:51,599 Speaker 2: We have a strategy of providing customers the choices to 97 00:05:51,720 --> 00:05:55,480 Speaker 2: use whatever's best for their job at hand. So and 98 00:05:55,600 --> 00:05:58,400 Speaker 2: Tropic is going to be an amazing set of models 99 00:05:58,400 --> 00:06:02,960 Speaker 2: for many, many use cases. And Amazon is fully invested 100 00:06:03,000 --> 00:06:05,800 Speaker 2: in building its own Titan models, which I think will 101 00:06:05,839 --> 00:06:09,040 Speaker 2: be really useful for other customers and other circumstances. And 102 00:06:09,440 --> 00:06:13,839 Speaker 2: of course our other model provider partners through Bedrock. So 103 00:06:14,640 --> 00:06:17,360 Speaker 2: I really think it's an ill founded premise that there's 104 00:06:17,400 --> 00:06:20,000 Speaker 2: been some change in urgency. We're fully urgent here on 105 00:06:20,080 --> 00:06:23,080 Speaker 2: generative AI for one reason and one reason alone. It's 106 00:06:23,120 --> 00:06:27,400 Speaker 2: because our customers need us to have great generative AI capabilities. 107 00:06:27,440 --> 00:06:30,280 Speaker 2: So many of them have their data platforms on AWS, 108 00:06:30,320 --> 00:06:33,720 Speaker 2: and if you've got your data here, you really want 109 00:06:33,760 --> 00:06:36,880 Speaker 2: to have your generative AI and all the powerful capabilities 110 00:06:36,880 --> 00:06:39,719 Speaker 2: that you need from those capabilities in the same place. 111 00:06:39,760 --> 00:06:42,800 Speaker 2: And so we have been are and will continue to 112 00:06:42,839 --> 00:06:44,919 Speaker 2: be very motivated to deliver for customers. 113 00:06:45,760 --> 00:06:47,760 Speaker 1: Adam, what does this mean for the kind of ramp 114 00:06:47,839 --> 00:06:51,159 Speaker 1: up or path forward for trainingum and inferentia. You've put 115 00:06:51,160 --> 00:06:53,760 Speaker 1: a lot of emphasis that anthrop it brings you a 116 00:06:54,440 --> 00:06:58,640 Speaker 1: maker or creator foundation models at scale, we now need 117 00:06:58,680 --> 00:07:03,040 Speaker 1: to ramp up I guess your third party manufacturing relationships 118 00:07:03,080 --> 00:07:05,800 Speaker 1: to say, okay, let's get more trainium on more inferential 119 00:07:05,880 --> 00:07:07,960 Speaker 1: online to support the workloads. 120 00:07:09,080 --> 00:07:12,360 Speaker 2: Well, it's absolutely true that there is a huge demand 121 00:07:12,960 --> 00:07:17,240 Speaker 2: for all of the different ships with which people do 122 00:07:17,760 --> 00:07:21,800 Speaker 2: a generative AI workloads. And so we absolutely have already 123 00:07:21,840 --> 00:07:26,360 Speaker 2: been ramping up our training and inferential supply chain and 124 00:07:26,640 --> 00:07:29,600 Speaker 2: ramping up the supply that we can create as quickly 125 00:07:29,600 --> 00:07:34,040 Speaker 2: as possible. And yes, Andthropic will have access to very 126 00:07:34,040 --> 00:07:38,360 Speaker 2: significant quantities of compute which will have trainum and inferentia 127 00:07:38,360 --> 00:07:42,120 Speaker 2: in them. So yes, that's one of many reasons why 128 00:07:42,160 --> 00:07:45,080 Speaker 2: we continue to ramp up and to provide a very 129 00:07:45,160 --> 00:07:49,680 Speaker 2: robust AWS controlled supply chain for AI chips. 130 00:07:49,720 --> 00:07:53,160 Speaker 3: And is that where the revenue boost comes adam, because 131 00:07:53,200 --> 00:07:55,360 Speaker 3: we're looking at the share price reaction is higher on 132 00:07:55,400 --> 00:07:58,400 Speaker 3: the day. When does this all start to really drive 133 00:07:58,680 --> 00:08:01,520 Speaker 3: adoption money on the bottom line for Amazon? 134 00:08:02,800 --> 00:08:06,680 Speaker 2: Well, I think that AI in general. Look, AWS has 135 00:08:06,680 --> 00:08:09,840 Speaker 2: had machine learning services since at least twenty seventeen when 136 00:08:09,880 --> 00:08:14,040 Speaker 2: we released our sage Maker machine learning service, which has 137 00:08:14,080 --> 00:08:16,720 Speaker 2: over one hundred thousand AWS customers on it. So we've 138 00:08:16,720 --> 00:08:20,120 Speaker 2: been doing machine learning for a long time inside of AWS, 139 00:08:20,680 --> 00:08:24,280 Speaker 2: and obviously more recently have had a significant number of 140 00:08:24,320 --> 00:08:28,760 Speaker 2: generative AI customers, and we will certainly continue to ramp 141 00:08:28,840 --> 00:08:32,000 Speaker 2: up anticipate quite steeply. We have many sources of growth 142 00:08:32,040 --> 00:08:37,200 Speaker 2: inside of AWS where a scaled and relatively sizable business 143 00:08:37,559 --> 00:08:41,839 Speaker 2: at this point, and customers are running their data platforms 144 00:08:41,840 --> 00:08:46,200 Speaker 2: on AWS. They are building out more and more applications 145 00:08:46,240 --> 00:08:50,840 Speaker 2: for things like supply chain and contact center management on AWS. 146 00:08:51,120 --> 00:08:53,400 Speaker 2: I've still a whole lot of storage and compute and 147 00:08:53,520 --> 00:08:56,760 Speaker 2: database workloads ramping on our AWS, so we have many 148 00:08:56,800 --> 00:09:00,559 Speaker 2: sources of growth I anticipate, but there's absolutely no doubt 149 00:09:00,600 --> 00:09:02,880 Speaker 2: the generator of AI looks like it's going to be 150 00:09:03,480 --> 00:09:06,560 Speaker 2: an explosive additional source of growth in the years ahead. 151 00:09:08,040 --> 00:09:10,320 Speaker 1: Adam, we put a lot of emphasis on the up 152 00:09:10,360 --> 00:09:13,880 Speaker 1: to four billion dollars, and you know, I understand and 153 00:09:13,960 --> 00:09:17,679 Speaker 1: thank you for explaining how the relationship will work in practice. 154 00:09:17,960 --> 00:09:20,040 Speaker 1: If I put to you this is an example of 155 00:09:20,080 --> 00:09:23,839 Speaker 1: Amazon or AWS basically paying a leader in the field 156 00:09:23,840 --> 00:09:27,760 Speaker 1: of AI, handing over cash to allow to make them 157 00:09:27,920 --> 00:09:31,120 Speaker 1: use TRAININGUM and inferentia, how would you respond to that 158 00:09:31,160 --> 00:09:33,400 Speaker 1: and explain to me how you bring new customers on 159 00:09:33,480 --> 00:09:36,839 Speaker 1: board who are really interested in the AI accelerators that 160 00:09:36,880 --> 00:09:41,120 Speaker 1: you have built without having to invest in them as 161 00:09:41,160 --> 00:09:42,000 Speaker 1: a sort of backup. 162 00:09:43,160 --> 00:09:47,560 Speaker 2: Sure. Well, the I think the really big news today 163 00:09:47,840 --> 00:09:53,720 Speaker 2: is the new expanded relationship between and Propic and Amazon, 164 00:09:54,520 --> 00:09:59,000 Speaker 2: in which they will have access to really large quantities 165 00:09:59,080 --> 00:10:03,800 Speaker 2: of trainingmen chips. Customers will have access to those models, 166 00:10:03,840 --> 00:10:08,440 Speaker 2: including early access to critical features through Amazon Bedrock, and 167 00:10:08,880 --> 00:10:11,480 Speaker 2: Amazon will get to AWS. We'll get to work with 168 00:10:11,520 --> 00:10:15,640 Speaker 2: Anthropic to ensure that, you know, we optimize our TRAININGM 169 00:10:15,640 --> 00:10:19,000 Speaker 2: and inferential technology going forward. That's the benefit for customers. 170 00:10:19,440 --> 00:10:21,800 Speaker 2: And yes, as part of this, we're pleased to be 171 00:10:21,880 --> 00:10:25,360 Speaker 2: making an additional an initial investment of one point twenty 172 00:10:25,400 --> 00:10:29,440 Speaker 2: five billion dollars into Anthropic. It's a financial investment and 173 00:10:29,880 --> 00:10:32,120 Speaker 2: that could go up as high as four billion as 174 00:10:32,200 --> 00:10:35,319 Speaker 2: you as you said over time, But it's really driven 175 00:10:35,320 --> 00:10:37,360 Speaker 2: around customer value and what this is going to mean 176 00:10:37,400 --> 00:10:41,760 Speaker 2: to customers who are very, very determined as they should be, 177 00:10:42,360 --> 00:10:46,040 Speaker 2: to figure out generative AI strategies. We already are working 178 00:10:46,120 --> 00:10:50,199 Speaker 2: in depth with customers, as is Anthropic on forming those 179 00:10:50,200 --> 00:10:53,040 Speaker 2: strategies and actually moving to execution. We have a lot 180 00:10:53,080 --> 00:10:57,480 Speaker 2: of great customers from Lonely Planet to Nexus, Lexus and 181 00:10:57,559 --> 00:11:01,000 Speaker 2: a number of others who are actually moving production with 182 00:11:01,120 --> 00:11:05,280 Speaker 2: generative AI on AWS and Anthropic, And in addition, as 183 00:11:05,320 --> 00:11:08,800 Speaker 2: you alluded to, we'll be working with all of the 184 00:11:09,080 --> 00:11:12,240 Speaker 2: partners that our customers want to do business with. If 185 00:11:12,240 --> 00:11:14,800 Speaker 2: it's an important partner to our customers, it's going to 186 00:11:14,800 --> 00:11:16,440 Speaker 2: be an important partner to us as well. 187 00:11:17,960 --> 00:11:21,920 Speaker 1: Amazon Web Services CEO Adam Sleipski, thank you.