WEBVTT - AWS CEO Matt Garman Talks AI Roadmap

0:00:02.520 --> 0:00:14.320
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news.

0:00:16.320 --> 0:00:19.680
<v Speaker 2>Welcome to our Bloomberg radio and television audiences worldwide. We

0:00:19.720 --> 0:00:25.239
<v Speaker 2>go right now to a conversation with Matt Garman, AWS CEO. Matt,

0:00:25.400 --> 0:00:28.280
<v Speaker 2>it's good to catch up. It has been basically one

0:00:28.360 --> 0:00:31.200
<v Speaker 2>year that you've been in the role as AWS CEO.

0:00:32.080 --> 0:00:34.840
<v Speaker 2>Is a place to start what has been the biggest

0:00:35.240 --> 0:00:37.960
<v Speaker 2>achievement in that time for AWS.

0:00:39.040 --> 0:00:40.839
<v Speaker 3>Yeah, thanks for having me on. It's nice to be

0:00:40.880 --> 0:00:44.640
<v Speaker 3>here again. Yeah, it's been a fantastic year of innovation.

0:00:45.000 --> 0:00:48.760
<v Speaker 3>It's really been incredible and as I look out there,

0:00:49.120 --> 0:00:50.920
<v Speaker 3>one of the things that I've been most excited about

0:00:51.040 --> 0:00:54.720
<v Speaker 3>is how fast our customers are innovating and ten adopting

0:00:54.840 --> 0:00:57.360
<v Speaker 3>many of the new technologies that we have. And as

0:00:57.400 --> 0:01:00.880
<v Speaker 3>you think about customers that are on this cloud mication journey,

0:01:01.120 --> 0:01:02.840
<v Speaker 3>many of them have been doing that for over the

0:01:02.960 --> 0:01:05.920
<v Speaker 3>last several years, but this year in particular, that we've

0:01:05.959 --> 0:01:09.800
<v Speaker 3>really seen an explosion of AI technologies, of agentic technologies,

0:01:10.160 --> 0:01:14.080
<v Speaker 3>and increasingly we're seeing more and more customers move their

0:01:14.240 --> 0:01:17.240
<v Speaker 3>entire estates into the cloud and AWS.

0:01:17.240 --> 0:01:18.839
<v Speaker 4>So it's been really fun to see.

0:01:18.880 --> 0:01:21.880
<v Speaker 3>It's been an incredible pace of technology and it's been

0:01:21.880 --> 0:01:22.920
<v Speaker 3>a really fun first year.

0:01:23.959 --> 0:01:26.640
<v Speaker 2>The moment that investors kind of sat up and paid

0:01:26.680 --> 0:01:32.000
<v Speaker 2>attention was when Amazon said that it's AI business was

0:01:32.040 --> 0:01:35.040
<v Speaker 2>at a multi billion dollar run rate in terms of sales.

0:01:35.440 --> 0:01:38.120
<v Speaker 2>What we don't understand as well is what proportion of

0:01:38.160 --> 0:01:40.040
<v Speaker 2>that is AWS infrastructure?

0:01:41.120 --> 0:01:43.760
<v Speaker 3>Yeah, that is AWS, right, And so the key is

0:01:43.800 --> 0:01:47.640
<v Speaker 3>that's a mix of customers running their own models. Some

0:01:47.680 --> 0:01:50.480
<v Speaker 3>of that is on Amazon Bedrock, which is our own

0:01:50.520 --> 0:01:53.800
<v Speaker 3>hosted models, where we have first party models like Amazon Nova,

0:01:53.840 --> 0:01:55.720
<v Speaker 3>as well as many of the third party models like

0:01:55.760 --> 0:01:59.160
<v Speaker 3>Anthropics models, and some of those are applications things like

0:02:00.120 --> 0:02:04.840
<v Speaker 3>on q which helps people do automated software development, as

0:02:04.880 --> 0:02:07.040
<v Speaker 3>well as a host of other capabilities, and so there's

0:02:07.080 --> 0:02:08.799
<v Speaker 3>a mix of that, and I think part of the

0:02:08.840 --> 0:02:11.440
<v Speaker 3>most interesting thing about being at a multi billion dollar

0:02:11.480 --> 0:02:14.400
<v Speaker 3>run rate is we're at the very earliest stages of

0:02:14.440 --> 0:02:17.919
<v Speaker 3>how AI is going to completely transform every single customer

0:02:17.960 --> 0:02:18.360
<v Speaker 3>out there.

0:02:18.680 --> 0:02:20.760
<v Speaker 4>And we talk to customers and we look at where the.

0:02:20.760 --> 0:02:24.880
<v Speaker 3>Technology landscape is, and we firmly believe that every single business,

0:02:24.919 --> 0:02:27.840
<v Speaker 3>every single industry, and really every single job is going

0:02:27.919 --> 0:02:30.960
<v Speaker 3>to be fundamentally transformed by AI. And I think we're

0:02:31.000 --> 0:02:33.440
<v Speaker 3>starting to see the early start the stages of that.

0:02:33.639 --> 0:02:35.880
<v Speaker 3>But again we're just at the very earliest stages that

0:02:35.919 --> 0:02:37.800
<v Speaker 3>I think what's going to be possible, and so that

0:02:37.880 --> 0:02:40.200
<v Speaker 3>multi billion dollar business that we have today is really

0:02:40.320 --> 0:02:41.000
<v Speaker 3>just the start.

0:02:42.080 --> 0:02:46.119
<v Speaker 2>Can you give me a generative AI revenue number.

0:02:47.000 --> 0:02:49.239
<v Speaker 4>For the world or for awls?

0:02:49.280 --> 0:02:52.080
<v Speaker 2>Are you guys for AWS? Maybe Amazon as a whole.

0:02:52.240 --> 0:02:55.240
<v Speaker 3>Yeah, Like I said, we are in multiple billions of dollars,

0:02:55.600 --> 0:02:59.240
<v Speaker 3>and that's for customers using AWS. We also use lots

0:02:59.280 --> 0:03:02.160
<v Speaker 3>of generative AI inside of Amazon for a wide range

0:03:02.200 --> 0:03:05.040
<v Speaker 3>of things. We use it to optimize our fulfillment centers.

0:03:05.520 --> 0:03:07.240
<v Speaker 3>We use it when you go to the retail site

0:03:07.280 --> 0:03:11.680
<v Speaker 3>to summarize reviews, or to help customers find products in

0:03:11.680 --> 0:03:15.280
<v Speaker 3>a faster and more interesting way. We use AI in

0:03:15.480 --> 0:03:19.160
<v Speaker 3>Alexa in our new Alexa Plus offering, where we conversationally

0:03:19.200 --> 0:03:23.680
<v Speaker 3>talk to customers through the Alexa interface and help them

0:03:23.680 --> 0:03:26.640
<v Speaker 3>accomplish things through voice that they were never able to

0:03:26.680 --> 0:03:30.280
<v Speaker 3>do before. So every single aspect of what Amazon does

0:03:31.080 --> 0:03:35.280
<v Speaker 3>leverages AI, and our customers are exactly the same. Customers

0:03:35.280 --> 0:03:38.600
<v Speaker 3>are looking to AWS to completely change, whether it's their

0:03:38.600 --> 0:03:42.880
<v Speaker 3>contact centers through something like Amazon Connect where it shows

0:03:43.000 --> 0:03:45.320
<v Speaker 3>AI capabilities so that you don't have to go program

0:03:45.360 --> 0:03:48.000
<v Speaker 3>it all the way down to our custom chips or

0:03:48.080 --> 0:03:51.120
<v Speaker 3>Nvidia processors or anything where customers at the metal are

0:03:51.120 --> 0:03:53.520
<v Speaker 3>building their own models. We have the whole range of

0:03:53.520 --> 0:03:57.000
<v Speaker 3>people that are building AI on top of AWS as

0:03:57.040 --> 0:03:58.400
<v Speaker 3>well as Amazon themselves.

0:03:59.280 --> 0:04:04.520
<v Speaker 2>We always credit AWS is being number one hyperscala. But

0:04:04.960 --> 0:04:07.440
<v Speaker 2>just what you said there about what the client's using

0:04:08.400 --> 0:04:12.480
<v Speaker 2>the silicon level through to capacity, it would really help

0:04:12.560 --> 0:04:16.560
<v Speaker 2>if you could proportionately tell me what percentage of workloads

0:04:16.560 --> 0:04:20.360
<v Speaker 2>are being run for training and which proportion of workloads

0:04:20.480 --> 0:04:21.560
<v Speaker 2>being run for inference.

0:04:21.960 --> 0:04:25.039
<v Speaker 4>Sure, yeah, and that changes over time. I think.

0:04:25.040 --> 0:04:28.599
<v Speaker 3>Look as we progress over time, more and more of

0:04:28.680 --> 0:04:31.240
<v Speaker 3>the AI workloads are being inference. I'd say in the

0:04:31.320 --> 0:04:34.600
<v Speaker 3>early stages of AI and general of AI, a lot

0:04:34.600 --> 0:04:36.880
<v Speaker 3>of that usage was dominated by training as people were

0:04:36.880 --> 0:04:40.120
<v Speaker 3>building these very large models with small amounts of usage.

0:04:40.160 --> 0:04:42.599
<v Speaker 3>Now the models are getting bigger and bigger, but the

0:04:42.720 --> 0:04:45.320
<v Speaker 3>usage is exploding at a rapid rate, and so I

0:04:45.480 --> 0:04:48.880
<v Speaker 3>expect that over the fullness of time, eighty percent, ninety percent,

0:04:48.960 --> 0:04:51.080
<v Speaker 3>the vast majority of usage is going to be in

0:04:51.240 --> 0:04:54.160
<v Speaker 3>inference out there, and really and just for all those

0:04:54.200 --> 0:04:58.159
<v Speaker 3>out there, inference It really is how AI is embedded

0:04:58.240 --> 0:05:01.080
<v Speaker 3>in the applications that everybody uses. And so as we

0:05:01.120 --> 0:05:03.360
<v Speaker 3>think about our customers building, you know, there's a small

0:05:03.440 --> 0:05:05.560
<v Speaker 3>number of people who are going to be building these models,

0:05:05.880 --> 0:05:08.520
<v Speaker 3>but everyone out there is going to use inference as

0:05:08.560 --> 0:05:11.599
<v Speaker 3>a core building block in everything they do. And every

0:05:11.640 --> 0:05:14.280
<v Speaker 3>application is going to have inference, and already is starting

0:05:14.320 --> 0:05:17.520
<v Speaker 3>to see inference built in to every application. And we

0:05:17.600 --> 0:05:20.320
<v Speaker 3>think about it as just the new building block. It's

0:05:20.400 --> 0:05:22.480
<v Speaker 3>just like compute, it's just like storage, it's just like

0:05:22.520 --> 0:05:23.200
<v Speaker 3>a database.

0:05:23.520 --> 0:05:25.039
<v Speaker 4>Inference is a core building block.

0:05:25.080 --> 0:05:26.880
<v Speaker 3>And so as you talk to people who are building

0:05:26.920 --> 0:05:30.039
<v Speaker 3>new applications, they don't think about it as AI is

0:05:30.040 --> 0:05:32.440
<v Speaker 3>over here and my application is over here. They really

0:05:32.440 --> 0:05:35.520
<v Speaker 3>think about AI is embedded in the experience. And so

0:05:35.680 --> 0:05:38.120
<v Speaker 3>it's increasingly I think it's going to be difficult for

0:05:38.200 --> 0:05:40.080
<v Speaker 3>people to say what part of your revenue is going

0:05:40.080 --> 0:05:42.360
<v Speaker 3>to be driven by AI. It's just part of the

0:05:42.400 --> 0:05:44.479
<v Speaker 3>application that you're building, and it's going to be a

0:05:44.520 --> 0:05:46.839
<v Speaker 3>core part of that experience, and it's going to deliver

0:05:47.200 --> 0:05:51.160
<v Speaker 3>lots of benefits from efficiency, from capabilities, and from user

0:05:51.200 --> 0:05:53.480
<v Speaker 3>experience for all sorts of applications and industries.

0:05:54.279 --> 0:05:57.640
<v Speaker 2>But present day, it's fair to say majority is still training.

0:05:58.080 --> 0:06:00.480
<v Speaker 3>No, I think that at this point more definitely more

0:06:00.560 --> 0:06:02.000
<v Speaker 3>usage as inference than training.

0:06:02.960 --> 0:06:06.080
<v Speaker 2>We want to welcome our radio and television audiences around

0:06:06.080 --> 0:06:09.560
<v Speaker 2>the world. We're speaking to AWS CEO Matt Garman, who

0:06:09.600 --> 0:06:14.279
<v Speaker 2>officially next week celebrates one year in that role leading AWS.

0:06:14.680 --> 0:06:19.280
<v Speaker 2>A new metric that has been discussed, particularly this earning season.

0:06:19.320 --> 0:06:22.600
<v Speaker 2>We discussed it with Nvidia CEO Jensen one this week

0:06:22.960 --> 0:06:27.640
<v Speaker 2>is token growth and tokenization. Has AWS got a metric

0:06:27.680 --> 0:06:28.559
<v Speaker 2>to share on that front?

0:06:29.080 --> 0:06:30.760
<v Speaker 3>I don't have any metrics to share on that front,

0:06:30.760 --> 0:06:32.640
<v Speaker 3>but I think it's one of the measures that we

0:06:32.680 --> 0:06:35.120
<v Speaker 3>can look at as the numbers of tokens that are

0:06:35.120 --> 0:06:37.279
<v Speaker 3>being served out there, but it's not the only one,

0:06:37.520 --> 0:06:40.120
<v Speaker 3>and I increasingly think that people are going to be

0:06:40.160 --> 0:06:43.920
<v Speaker 3>thinking about these things differently. Tokens are a particularly interesting

0:06:44.720 --> 0:06:47.080
<v Speaker 3>thing to look at when you're thinking about text generation,

0:06:47.560 --> 0:06:49.640
<v Speaker 3>but not all things are created equal.

0:06:49.680 --> 0:06:51.200
<v Speaker 4>I think, particularly as you think about.

0:06:51.040 --> 0:06:54.800
<v Speaker 3>AI reasoning models, the input and output tokens don't necessarily

0:06:56.240 --> 0:06:59.160
<v Speaker 3>talk about the work that's being done, and increasingly you're

0:06:59.200 --> 0:07:01.799
<v Speaker 3>seeing models can do work for a really long period

0:07:01.839 --> 0:07:05.080
<v Speaker 3>of time before they output tokens, and so you're having

0:07:05.120 --> 0:07:09.240
<v Speaker 3>these models that can sometimes think for hours at a time. Right,

0:07:09.279 --> 0:07:12.000
<v Speaker 3>you ask these things to go and actually do research

0:07:12.000 --> 0:07:13.760
<v Speaker 3>on your behalf. They can go out to the internet,

0:07:13.920 --> 0:07:16.920
<v Speaker 3>they can pull information back, they can synthesize, they can

0:07:17.000 --> 0:07:19.920
<v Speaker 3>redo things. If you think about coding and que developer,

0:07:20.560 --> 0:07:23.160
<v Speaker 3>we're seeing lots of coding where it goes and actually

0:07:23.480 --> 0:07:27.120
<v Speaker 3>reasons and does iterations and iterations and improves on itself,

0:07:27.320 --> 0:07:30.400
<v Speaker 3>looks at what it's done, and then eventually outputs the

0:07:30.520 --> 0:07:32.800
<v Speaker 3>end result. And so at some point kind of the

0:07:33.320 --> 0:07:35.920
<v Speaker 3>final output token is not really the best measure of

0:07:35.960 --> 0:07:38.920
<v Speaker 3>how much work is being done. If you think about images,

0:07:38.960 --> 0:07:41.280
<v Speaker 3>if you think about videos, there's a lot of content

0:07:41.360 --> 0:07:43.920
<v Speaker 3>that's being created and a lot of thought that's being done.

0:07:43.920 --> 0:07:46.000
<v Speaker 4>And so tokens are one aspect of it.

0:07:46.040 --> 0:07:48.240
<v Speaker 3>And it's an interesting measure, but I don't think it's

0:07:48.280 --> 0:07:52.360
<v Speaker 3>the only measure to look at. Although they are rapidly increasing.

0:07:53.840 --> 0:07:58.440
<v Speaker 2>Project RAY near Massive Custom Server Design project. Yeah, what

0:07:58.560 --> 0:08:02.040
<v Speaker 2>is the operational statu and latest on project right now?

0:08:02.200 --> 0:08:05.760
<v Speaker 3>Yeah, So we're incredibly excited about so project right here

0:08:06.880 --> 0:08:09.840
<v Speaker 3>is a collaboration that we have with our partners at

0:08:09.840 --> 0:08:14.400
<v Speaker 3>Anthropic to build the largest compute cluster that they'll use

0:08:14.440 --> 0:08:17.400
<v Speaker 3>to train their next generation of their claud models, and

0:08:18.080 --> 0:08:20.680
<v Speaker 3>Anthropic has the very best models out there today. Claude

0:08:20.680 --> 0:08:23.320
<v Speaker 3>four just launched, I think it was last week, and

0:08:23.800 --> 0:08:28.400
<v Speaker 3>it's been getting incredible adoption out there from our customer base.

0:08:29.720 --> 0:08:31.880
<v Speaker 3>Anthropic is going to be training their next version of

0:08:31.920 --> 0:08:35.280
<v Speaker 3>their model on top of Trainium two, which is Amazon's

0:08:35.280 --> 0:08:40.440
<v Speaker 3>custom built accelerator processors purpose built for AI workloads, and

0:08:40.480 --> 0:08:44.120
<v Speaker 3>we're building one of the largest clusters ever released. It's

0:08:44.120 --> 0:08:47.160
<v Speaker 3>an enormous cluster, more than five times the size of

0:08:47.200 --> 0:08:50.200
<v Speaker 3>the cluster compared to the last one that they trained on,

0:08:50.200 --> 0:08:52.160
<v Speaker 3>which again is the world's leading model.

0:08:52.320 --> 0:08:53.800
<v Speaker 4>So we're super excited about that.

0:08:54.440 --> 0:08:57.120
<v Speaker 3>We're landing Trainium to servers now and they're already in

0:08:57.160 --> 0:09:00.679
<v Speaker 3>operation and Nthropic has already is our using parts of

0:09:00.679 --> 0:09:01.360
<v Speaker 3>that cluster, and.

0:09:01.320 --> 0:09:02.720
<v Speaker 4>So super excited about that.

0:09:02.760 --> 0:09:04.800
<v Speaker 3>And the performance that we're seeing out of Trainium too

0:09:05.520 --> 0:09:08.760
<v Speaker 3>continues to be very impressive and really pushes the envelope

0:09:08.800 --> 0:09:12.120
<v Speaker 3>I think on what's possible both from an absolute performance

0:09:12.120 --> 0:09:14.880
<v Speaker 3>basis as well as a cost, performance and scale basis.

0:09:15.040 --> 0:09:16.880
<v Speaker 3>I think some of those are equally going to be

0:09:16.960 --> 0:09:19.439
<v Speaker 3>really important as we move forward in this world, because

0:09:19.480 --> 0:09:22.240
<v Speaker 3>today much of the feedback you get is that AI

0:09:22.360 --> 0:09:25.679
<v Speaker 3>is still too expensive. But costs are coming down pretty aggressively,

0:09:25.880 --> 0:09:28.280
<v Speaker 3>and it's still too expensive, and so we think there's

0:09:28.280 --> 0:09:30.680
<v Speaker 3>a number of things that need to happen there. Innovation

0:09:30.760 --> 0:09:32.720
<v Speaker 3>on the silicon level is one of those things that

0:09:32.800 --> 0:09:35.240
<v Speaker 3>needs to help bring the cost down, as well as

0:09:35.240 --> 0:09:38.079
<v Speaker 3>innovation on the software side and algorithmic side so that

0:09:38.120 --> 0:09:41.120
<v Speaker 3>you have to use less compute per unit of inference

0:09:41.440 --> 0:09:44.000
<v Speaker 3>or training. So all of those are important to bring

0:09:44.000 --> 0:09:46.160
<v Speaker 3>that cost down to make it more and more possible

0:09:46.840 --> 0:09:49.120
<v Speaker 3>for ADI to be used in all of the places

0:09:49.320 --> 0:09:52.360
<v Speaker 3>that we think that it will be over time Matt.

0:09:52.440 --> 0:09:56.400
<v Speaker 2>On Wednesday, Nvidia CEO Jensen Wang summarized inference demand for me.

0:09:56.440 --> 0:09:58.160
<v Speaker 2>I just wanted to play you that SoundBite.

0:09:58.320 --> 0:09:59.319
<v Speaker 4>Sure, well, we.

0:09:59.320 --> 0:10:02.440
<v Speaker 5>Got a whole bunch of engines firing right now. The

0:10:02.480 --> 0:10:06.719
<v Speaker 5>biggest one, of course, is the reasoning AI inference. The

0:10:06.760 --> 0:10:12.200
<v Speaker 5>demand is just off the charts. You see the popularity

0:10:12.280 --> 0:10:13.600
<v Speaker 5>of all these AI services.

0:10:13.640 --> 0:10:17.520
<v Speaker 2>Now your pitch for trainium too. And as you know,

0:10:17.559 --> 0:10:19.680
<v Speaker 2>I've kind of taken a part the serve a design

0:10:19.720 --> 0:10:23.360
<v Speaker 2>and looked at it is the efficiency and cost efficiency

0:10:23.440 --> 0:10:27.160
<v Speaker 2>relative to Nvidia tech. Are you seeing that same demand

0:10:27.240 --> 0:10:31.840
<v Speaker 2>Jensen outlined for Trainium two outside of the relationship with Amthropic.

0:10:33.160 --> 0:10:35.680
<v Speaker 3>Yeah, Look, we're seeing it across a number of different places,

0:10:35.679 --> 0:10:38.680
<v Speaker 3>but it's not really Trainingum two versus in Nvidia, and

0:10:38.720 --> 0:10:40.240
<v Speaker 3>I think that's not really the right way to think

0:10:40.240 --> 0:10:42.640
<v Speaker 3>about it. I think there's plenty of room. The opportunity

0:10:42.679 --> 0:10:45.280
<v Speaker 3>in this space is massive. It's not one versus the other.

0:10:45.320 --> 0:10:47.439
<v Speaker 3>We think that there's plenty of room for both these

0:10:47.480 --> 0:10:49.760
<v Speaker 3>and Jensen and I speak about this all the time

0:10:49.960 --> 0:10:53.520
<v Speaker 3>that in Vidia is an incredibly fantastic platform. They've built

0:10:53.520 --> 0:10:57.160
<v Speaker 3>a really strong platform that's useful and is the leading

0:10:57.200 --> 0:11:00.320
<v Speaker 3>platform for many many applications out there, and so we

0:11:00.480 --> 0:11:03.640
<v Speaker 3>are incredible design partners with them. We make sure that

0:11:03.679 --> 0:11:06.240
<v Speaker 3>we have the latest in video technology for everyone, and

0:11:06.320 --> 0:11:08.920
<v Speaker 3>we continue to push the envelope on what's possible with

0:11:09.040 --> 0:11:11.880
<v Speaker 3>all of the latest in Vidia capabilities. And we think

0:11:11.880 --> 0:11:14.880
<v Speaker 3>there's room for Trainium and other technologies as well, and

0:11:14.880 --> 0:11:18.200
<v Speaker 3>we're really excited about that, and so we have many

0:11:18.280 --> 0:11:21.720
<v Speaker 3>of the leading AI labs are incredibly excited about using

0:11:22.000 --> 0:11:24.440
<v Speaker 3>Trainium too, and really leaning into the benefits that you

0:11:24.480 --> 0:11:27.360
<v Speaker 3>get there, But for the law for a long time,

0:11:27.400 --> 0:11:29.920
<v Speaker 3>these things are going to be living in concert together,

0:11:30.120 --> 0:11:32.679
<v Speaker 3>and I think there's plenty of room, and customers want choice.

0:11:32.720 --> 0:11:34.599
<v Speaker 3>At the end of the day, Customers don't want to

0:11:34.640 --> 0:11:37.040
<v Speaker 3>be forced into using one platform or the other. They'd

0:11:37.040 --> 0:11:39.000
<v Speaker 3>love to have choice in Our job at AWS is

0:11:39.040 --> 0:11:41.160
<v Speaker 3>to give customers as much choice as possible.

0:11:42.120 --> 0:11:47.040
<v Speaker 2>What is general availability of Nvidia GB two hundred for AWS?

0:11:47.040 --> 0:11:52.040
<v Speaker 2>And have you, I guess, launched Grace Blackwell backed instances yet?

0:11:52.280 --> 0:11:56.040
<v Speaker 3>Yes, yep, so we've launched our they would call them

0:11:56.040 --> 0:11:59.240
<v Speaker 3>P six instances, And so those are available in AWS

0:11:59.280 --> 0:12:02.000
<v Speaker 3>today and customers are using them and liking them and

0:12:02.040 --> 0:12:04.800
<v Speaker 3>the performance is fantastic. So those are available today. We're

0:12:04.800 --> 0:12:08.640
<v Speaker 3>continuing to ramp capacity. We work very closely with the

0:12:08.720 --> 0:12:12.320
<v Speaker 3>Nvidia team to aggressively ramp capacity and demand as strong

0:12:12.600 --> 0:12:15.720
<v Speaker 3>for those P six instances. But customers are able to

0:12:15.720 --> 0:12:19.440
<v Speaker 3>go and test those out today, and like I said,

0:12:19.720 --> 0:12:22.280
<v Speaker 3>we're ramping capacity incredibly fast all around the world and

0:12:22.400 --> 0:12:24.440
<v Speaker 3>in our various different regions.

0:12:25.800 --> 0:12:30.400
<v Speaker 2>Now, what is your attitude to Claude Anthropics model being

0:12:30.440 --> 0:12:33.480
<v Speaker 2>available elsewhere on Azure Foundry for example?

0:12:35.000 --> 0:12:36.880
<v Speaker 4>Great I mean that's okay too.

0:12:36.920 --> 0:12:40.720
<v Speaker 3>I think many of our customers make their applications available

0:12:40.840 --> 0:12:44.800
<v Speaker 3>in different places, and we understand that various different customers

0:12:44.840 --> 0:12:48.480
<v Speaker 3>want to use capabilities in different areas and different clouds.

0:12:48.920 --> 0:12:51.280
<v Speaker 3>Our job is to make AWS and this is what

0:12:51.320 --> 0:12:54.640
<v Speaker 3>we do, is to make AWS the best place to

0:12:54.760 --> 0:12:58.640
<v Speaker 3>run every type of workload, and that includes anthropic claud models, but.

0:12:58.600 --> 0:13:00.000
<v Speaker 4>It includes a wide range of things.

0:13:00.040 --> 0:13:04.240
<v Speaker 3>And frankly, that's why we see big customers migrating over

0:13:04.280 --> 0:13:08.120
<v Speaker 3>to AWS. Take somebody like a Mondali's who's really gone

0:13:08.160 --> 0:13:11.280
<v Speaker 3>all in with AWS and moved some of their workloads

0:13:11.280 --> 0:13:13.360
<v Speaker 3>to there. One of the reasons is that they see

0:13:13.400 --> 0:13:16.360
<v Speaker 3>that we have capabilities sometimes using AI by the way,

0:13:16.600 --> 0:13:20.120
<v Speaker 3>in order to really help them optimize their costs and

0:13:20.600 --> 0:13:24.160
<v Speaker 3>have the most available, most secure platform in monthlies. This case,

0:13:24.400 --> 0:13:28.120
<v Speaker 3>they're taking many of their legacy Windows platforms and transforming

0:13:28.200 --> 0:13:30.800
<v Speaker 3>them into Linux applications and saving.

0:13:30.480 --> 0:13:32.240
<v Speaker 4>All of that licensing costs.

0:13:32.440 --> 0:13:35.000
<v Speaker 3>But we have many customers who are doing that, and

0:13:35.440 --> 0:13:38.280
<v Speaker 3>so our job is to make AWS by far the

0:13:38.280 --> 0:13:42.400
<v Speaker 3>most technically capable platform that has the most and widest

0:13:42.440 --> 0:13:44.679
<v Speaker 3>set of services, and that's.

0:13:44.480 --> 0:13:44.920
<v Speaker 4>What we do.

0:13:45.400 --> 0:13:47.920
<v Speaker 3>But I'm perfectly happy for other people to use, Like,

0:13:48.120 --> 0:13:51.679
<v Speaker 3>it's great that Claud's making their services available elsewhere and

0:13:52.280 --> 0:13:54.680
<v Speaker 3>we see the vast majority of that usage happening in AWS.

0:13:54.679 --> 0:13:54.880
<v Speaker 4>Though.

0:13:55.679 --> 0:13:58.680
<v Speaker 2>Will we see open AI models on AWS this year?

0:13:59.400 --> 0:14:02.360
<v Speaker 3>Well, just like you know, we encourage all of our

0:14:02.400 --> 0:14:05.400
<v Speaker 3>partners to be able to be available elsewhere. I'd love

0:14:05.440 --> 0:14:06.960
<v Speaker 3>for others to take that same tack.

0:14:08.760 --> 0:14:11.959
<v Speaker 2>Let's end it with this a question from the audience actually,

0:14:12.000 --> 0:14:14.199
<v Speaker 2>which is where you're going to grow data center capacity

0:14:14.280 --> 0:14:16.319
<v Speaker 2>around the world. I got a lot of questions from

0:14:16.440 --> 0:14:19.960
<v Speaker 2>Latin America and Europe in particular where Jensen flies to

0:14:20.080 --> 0:14:20.560
<v Speaker 2>next week?

0:14:20.640 --> 0:14:22.120
<v Speaker 4>Yeah. Great.

0:14:22.880 --> 0:14:26.040
<v Speaker 3>So in Latin America we're continuing to span expand our

0:14:26.040 --> 0:14:29.840
<v Speaker 3>capacity pretty aggressively. Actually, earlier this year we launched our

0:14:29.880 --> 0:14:32.880
<v Speaker 3>Mexico region, which has been really well received by customers,

0:14:33.080 --> 0:14:35.680
<v Speaker 3>and we've announced a new region in Chile. And we

0:14:35.720 --> 0:14:38.080
<v Speaker 3>already have and for many years have had a region

0:14:38.080 --> 0:14:41.160
<v Speaker 3>in Brazil which is quite popular and has many of

0:14:41.360 --> 0:14:44.840
<v Speaker 3>the largest financial institutions in South America running there. So

0:14:45.960 --> 0:14:49.560
<v Speaker 3>across Central and South America, we are continuing to rapidly expand.

0:14:50.200 --> 0:14:52.280
<v Speaker 3>In Europe we're expanding as well. We have many regions

0:14:52.320 --> 0:14:54.520
<v Speaker 3>already in Europe. One of the things I'm most excited

0:14:54.520 --> 0:14:56.680
<v Speaker 3>about actually is at the end of this year we're

0:14:56.680 --> 0:14:59.280
<v Speaker 3>going to be launching the European Sovereign Cloud, which is

0:14:59.320 --> 0:15:02.920
<v Speaker 3>a unique capability that no one has, which is completely

0:15:02.960 --> 0:15:07.840
<v Speaker 3>designed for critical EU focused sovereign workloads, and we think

0:15:08.440 --> 0:15:11.680
<v Speaker 3>given some of the concerns that folks have around data sovereignty,

0:15:11.920 --> 0:15:16.040
<v Speaker 3>particularly for government workloads as well as regulated workloads, we

0:15:16.040 --> 0:15:19.360
<v Speaker 3>think that's going to be an incredibly op popular opportunity

0:15:19.360 --> 0:15:19.880
<v Speaker 3>for everybody.

0:15:20.920 --> 0:15:24.120
<v Speaker 2>Matt Garman AWSCO, thank you very much.

0:15:24.280 --> 0:15:25.120
<v Speaker 4>Thank you for having me