WEBVTT - Amazon Web Services CEO Matt Garman Talks Capex Investment

0:00:02.520 --> 0:00:07.080
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news.

0:00:07.520 --> 0:00:11.400
<v Speaker 2>AWS CEO Matt Garman joins us now on Bloomberg Television

0:00:11.680 --> 0:00:13.800
<v Speaker 2>and Bloomberg Radio. Matt, it's great to have you back

0:00:13.840 --> 0:00:16.200
<v Speaker 2>on the show. You know, this is the fastest growth

0:00:16.280 --> 0:00:19.160
<v Speaker 2>for AWS in nearly five years. But to kind of

0:00:19.440 --> 0:00:22.599
<v Speaker 2>capture a moment in time, could we start by talking

0:00:22.640 --> 0:00:25.320
<v Speaker 2>about how much of that was driven by the frontier labs,

0:00:25.520 --> 0:00:28.560
<v Speaker 2>the big ones Open AI and Thropic, or how much

0:00:28.560 --> 0:00:31.840
<v Speaker 2>of it was something broader across the enterprise, something bigger

0:00:31.840 --> 0:00:32.200
<v Speaker 2>in AI.

0:00:33.240 --> 0:00:36.040
<v Speaker 1>Yeah, it's actually across the board this growth that we're

0:00:36.040 --> 0:00:38.879
<v Speaker 1>seeing right now, and much of that is some of

0:00:38.880 --> 0:00:41.560
<v Speaker 1>the AI labs that are building their models on top

0:00:41.600 --> 0:00:44.400
<v Speaker 1>of AWS. But actually a lot of that growth is

0:00:44.440 --> 0:00:48.000
<v Speaker 1>spread across all of our startup and enterprise customers, where

0:00:48.040 --> 0:00:51.800
<v Speaker 1>AI is really impacting almost every single industry. And so

0:00:51.840 --> 0:00:55.240
<v Speaker 1>whether it's financial services companies or healthcare companies, or retail

0:00:55.280 --> 0:00:59.480
<v Speaker 1>companies or media companies. Really the growth is coming across

0:00:59.520 --> 0:01:03.600
<v Speaker 1>the board as companies of all different sizes in all

0:01:03.600 --> 0:01:08.040
<v Speaker 1>different industries are looking to use AI to grow their business.

0:01:08.080 --> 0:01:10.360
<v Speaker 1>And so we're seeing a lot of growth from the

0:01:10.400 --> 0:01:14.480
<v Speaker 1>top frontier labs, but for AWS, We're not like some others.

0:01:14.480 --> 0:01:17.319
<v Speaker 1>Maybe it's concentrated on just one or two large customers,

0:01:17.560 --> 0:01:20.319
<v Speaker 1>but it's actually growth from a really broad set of customers,

0:01:20.760 --> 0:01:21.679
<v Speaker 1>which is nice to see.

0:01:23.200 --> 0:01:23.400
<v Speaker 3>Matt.

0:01:23.440 --> 0:01:26.240
<v Speaker 2>When AWS says the AI business has a revenue run

0:01:26.280 --> 0:01:28.479
<v Speaker 2>rate of twenty five billion dollars, what does that mean?

0:01:28.560 --> 0:01:30.200
<v Speaker 3>What does the figuring compass?

0:01:31.560 --> 0:01:36.399
<v Speaker 1>Yeah, that includes both training from very large companies like

0:01:36.440 --> 0:01:40.680
<v Speaker 1>Anthropic and Open Ai and other large tech companies, as

0:01:40.680 --> 0:01:42.440
<v Speaker 1>well as many startups. But it also and then a

0:01:42.480 --> 0:01:45.559
<v Speaker 1>big chunk of that is really inferenced that that broad

0:01:45.560 --> 0:01:48.680
<v Speaker 1>swath of companies are doing. And so as companies think

0:01:48.680 --> 0:01:51.680
<v Speaker 1>about how they take models in something on Amazon Bedrock

0:01:51.880 --> 0:01:54.080
<v Speaker 1>and run those models to get value out of their business,

0:01:54.200 --> 0:01:56.600
<v Speaker 1>some of them are automating processes. Many of them are

0:01:56.680 --> 0:01:59.760
<v Speaker 1>running agentic workloads that they build on AWS to further

0:02:00.160 --> 0:02:04.280
<v Speaker 1>drive efficiencies in their business or deliver new customer experiences.

0:02:04.480 --> 0:02:07.080
<v Speaker 1>We consider all of that work, whether it's agent growth,

0:02:07.120 --> 0:02:10.760
<v Speaker 1>whether it's inference, and some of that is training new models,

0:02:10.800 --> 0:02:13.600
<v Speaker 1>all of that encompasses the AI business for us.

0:02:14.560 --> 0:02:16.960
<v Speaker 2>I think I've asked you this question at various points

0:02:17.000 --> 0:02:20.360
<v Speaker 2>in time since you became AWS CEO. But is there

0:02:20.400 --> 0:02:24.320
<v Speaker 2>a percentage split right now training versus inference that you

0:02:24.320 --> 0:02:24.800
<v Speaker 2>can give me.

0:02:26.000 --> 0:02:28.720
<v Speaker 1>Yeah, you do, and it actually keeps shifting, I would

0:02:28.760 --> 0:02:31.520
<v Speaker 1>say more and more. It keeps shifting more towards inference.

0:02:31.600 --> 0:02:34.720
<v Speaker 1>I think we still see large training clusters being used

0:02:34.720 --> 0:02:37.880
<v Speaker 1>by a number of companies, but as these as these

0:02:37.880 --> 0:02:40.320
<v Speaker 1>models get really popular and really powerful, more and more

0:02:40.360 --> 0:02:44.040
<v Speaker 1>companies are integrating that inference into their workloads. So, you know,

0:02:44.080 --> 0:02:46.079
<v Speaker 1>I don't know the exact percentage today, but it keeps

0:02:46.080 --> 0:02:48.440
<v Speaker 1>shifting more and more towards inference, and we expect that

0:02:48.520 --> 0:02:51.840
<v Speaker 1>to continue as the economics makes sense where you really

0:02:51.840 --> 0:02:55.079
<v Speaker 1>want most of that cost and spend being where value

0:02:55.120 --> 0:02:56.560
<v Speaker 1>is being created for end customers.

0:02:58.200 --> 0:03:01.560
<v Speaker 2>Amazon's overall CAPEX number for this year is very big,

0:03:01.800 --> 0:03:04.360
<v Speaker 2>two hundred and twenty billion dollars, So it's up twenty billion,

0:03:04.400 --> 0:03:09.080
<v Speaker 2>and Andy Jesse explained that's mostly AI, but it also

0:03:09.080 --> 0:03:12.280
<v Speaker 2>accounts for higher memory pricing. Right, capex higher because of

0:03:12.400 --> 0:03:16.680
<v Speaker 2>the cost environment's higher. But from aws's perspective, what's the

0:03:16.720 --> 0:03:18.080
<v Speaker 2>trajectory for next year?

0:03:18.560 --> 0:03:20.320
<v Speaker 3>You expect that capex.

0:03:19.840 --> 0:03:21.880
<v Speaker 2>Will be biggest still because I think one of the

0:03:21.919 --> 0:03:23.840
<v Speaker 2>things that Amazon is quite clear about is that even

0:03:24.000 --> 0:03:26.760
<v Speaker 2>at two hundred and twenty billion dollars, it might not

0:03:26.840 --> 0:03:28.320
<v Speaker 2>be enough to meet current demand.

0:03:29.400 --> 0:03:31.800
<v Speaker 1>Yeah, one of the things that we're quite excited about

0:03:31.919 --> 0:03:35.360
<v Speaker 1>is just the potential business for us is just massive,

0:03:35.400 --> 0:03:38.360
<v Speaker 1>and we see this as a huge opportunity for us

0:03:38.440 --> 0:03:41.360
<v Speaker 1>to really invest and help customers take advantage of the

0:03:41.400 --> 0:03:44.240
<v Speaker 1>AI opportunity, and so we will keep investing. We think

0:03:44.280 --> 0:03:47.680
<v Speaker 1>that there's a big opportunity for us and for customers,

0:03:48.400 --> 0:03:51.080
<v Speaker 1>And as Andy kind of called out last week, we

0:03:51.160 --> 0:03:53.880
<v Speaker 1>have really great insight into what that demand is going

0:03:53.960 --> 0:03:56.120
<v Speaker 1>to be from customers, and so when we invest and

0:03:56.160 --> 0:03:59.360
<v Speaker 1>we will keep investing in CAPEX next year as well,

0:04:00.280 --> 0:04:03.280
<v Speaker 1>we have great insight into when that revenue is going

0:04:03.280 --> 0:04:06.040
<v Speaker 1>to land. And so it's pretty well known for us,

0:04:06.040 --> 0:04:09.520
<v Speaker 1>And as Andy mentioned, much of our capacity has already

0:04:09.560 --> 0:04:11.800
<v Speaker 1>spoken through through the end of twenty seven and even

0:04:11.840 --> 0:04:14.280
<v Speaker 1>through much of twenty eight, and so as we're investing,

0:04:14.320 --> 0:04:17.560
<v Speaker 1>we're getting five year commitments from customers. We're getting these

0:04:17.560 --> 0:04:20.839
<v Speaker 1>long term commitments, and so we're out there making investments

0:04:20.880 --> 0:04:22.839
<v Speaker 1>to keep being able to grow the business and try

0:04:22.839 --> 0:04:25.960
<v Speaker 1>to meet customer demand. But as you say, today, demands

0:04:26.000 --> 0:04:30.960
<v Speaker 1>still significantly outstripped supply. And we're trying to build and

0:04:31.040 --> 0:04:33.640
<v Speaker 1>invest to keep up with what customers are asking for.

0:04:34.240 --> 0:04:36.640
<v Speaker 2>And so we expect the CAPEX number will be bigger

0:04:36.680 --> 0:04:38.040
<v Speaker 2>next year than it is this year.

0:04:39.600 --> 0:04:42.160
<v Speaker 1>I expect us to keep investing as we see opportunities,

0:04:42.160 --> 0:04:44.600
<v Speaker 1>and some of that we'll see how the market continues

0:04:44.600 --> 0:04:48.279
<v Speaker 1>to grow. But right now we're excited about our investments

0:04:48.279 --> 0:04:49.720
<v Speaker 1>and we'll continue to invest.

0:04:51.440 --> 0:04:53.720
<v Speaker 2>The chip business has also got a lot of momentum.

0:04:53.839 --> 0:04:55.880
<v Speaker 2>You know, Trainium for some time has been a big

0:04:55.920 --> 0:04:57.840
<v Speaker 2>part of the strategy, and we've discussed that. I've been

0:04:57.839 --> 0:05:00.000
<v Speaker 2>able to go to Austin spend a lot of time

0:05:00.040 --> 0:05:05.719
<v Speaker 2>in Anapurna Labs to look at the server design. Let's

0:05:05.760 --> 0:05:07.920
<v Speaker 2>start by asking what when you say the chip business

0:05:07.920 --> 0:05:09.919
<v Speaker 2>as a revenue run rate twenty five billion dollars, what

0:05:09.960 --> 0:05:12.280
<v Speaker 2>does that mean? That is the business of selling the

0:05:12.360 --> 0:05:16.080
<v Speaker 2>chips to third parties or renting out capacity based on

0:05:16.120 --> 0:05:16.599
<v Speaker 2>the chips.

0:05:17.400 --> 0:05:20.200
<v Speaker 1>Yeah, renting out capacity based on those chips. So today

0:05:20.200 --> 0:05:22.760
<v Speaker 1>we run all of our own chips inside of the

0:05:22.800 --> 0:05:26.280
<v Speaker 1>AWS cloud, and that includes both Trainingum chips as well

0:05:26.320 --> 0:05:30.320
<v Speaker 1>as Graviton chips or which our general purpose processors. Graviton

0:05:30.560 --> 0:05:34.280
<v Speaker 1>is incredibly popular We've been building Graviton for many years now,

0:05:34.320 --> 0:05:37.240
<v Speaker 1>and in fact, almost all of our large customers use

0:05:37.279 --> 0:05:40.039
<v Speaker 1>Graviton as some part of their deployment, and it's a

0:05:40.120 --> 0:05:43.640
<v Speaker 1>key part of how help customers deliver extra value by

0:05:43.640 --> 0:05:48.160
<v Speaker 1>lowering their costs. Trainium is wildly popular, and as we've

0:05:48.160 --> 0:05:50.880
<v Speaker 1>mentioned a couple of times now, we're largely sold out

0:05:50.920 --> 0:05:54.200
<v Speaker 1>through the end of next year for Trainium capacity and

0:05:54.240 --> 0:05:57.479
<v Speaker 1>it's really really popular. We're starting to see both large

0:05:57.480 --> 0:06:00.760
<v Speaker 1>customers as well as startups really love the benefits of

0:06:00.839 --> 0:06:03.920
<v Speaker 1>running on Trainium and having that choice in a cloud

0:06:03.920 --> 0:06:07.520
<v Speaker 1>where you can run on purpose built processors like Trainium

0:06:07.600 --> 0:06:11.240
<v Speaker 1>that give you really great AI performance, together with in

0:06:11.320 --> 0:06:14.120
<v Speaker 1>Nvidia chips for when you need some of that processing

0:06:14.480 --> 0:06:18.120
<v Speaker 1>for GPUs as well, gives customers the right mix that

0:06:18.120 --> 0:06:21.480
<v Speaker 1>they're oftentimes looking for, and so we find that combination

0:06:21.600 --> 0:06:24.920
<v Speaker 1>to be quite powerful and our Trainium three chips are

0:06:24.960 --> 0:06:26.360
<v Speaker 1>really really popular with customers.

0:06:26.440 --> 0:06:28.479
<v Speaker 3>Right now, we're live.

0:06:28.320 --> 0:06:31.640
<v Speaker 2>On Bloomberg Television and Bloomberg Radio. This is Bloomberg Tech.

0:06:31.680 --> 0:06:34.320
<v Speaker 2>In conversation with Matt garm and the COO of AWS

0:06:35.080 --> 0:06:39.240
<v Speaker 2>explain the difference in economics between capacity based on Trainium

0:06:39.320 --> 0:06:43.520
<v Speaker 2>and capacity based on Blackwell, for example, what is the

0:06:43.600 --> 0:06:46.440
<v Speaker 2>selling point of going with Amazon Silicon.

0:06:48.320 --> 0:06:49.120
<v Speaker 3>Yeah, we think that.

0:06:49.560 --> 0:06:52.040
<v Speaker 1>Look, we think customers want choice, and so we offer

0:06:52.800 --> 0:06:56.120
<v Speaker 1>great offerings for both products. Some customers really like to

0:06:56.160 --> 0:06:59.560
<v Speaker 1>build on in Nvidia GPUs and those are fantastic, and

0:06:59.560 --> 0:07:02.880
<v Speaker 1>we're one of Nvidia's absolute largest customers in the world.

0:07:03.120 --> 0:07:05.200
<v Speaker 1>We've been offering in Vidia GPUs in the cloud for

0:07:05.520 --> 0:07:10.520
<v Speaker 1>over a decade now, and AWS is the most scalable

0:07:10.880 --> 0:07:14.680
<v Speaker 1>and secure and stable place to run in Vidia servers anywhere,

0:07:15.280 --> 0:07:17.640
<v Speaker 1>and so we're very excited about that business. But when

0:07:17.640 --> 0:07:20.160
<v Speaker 1>you add trainingum, we're able to lower the costs for

0:07:20.240 --> 0:07:23.440
<v Speaker 1>many workloads, and so customers like to have that opportunity

0:07:23.480 --> 0:07:26.560
<v Speaker 1>where they can take these really massively scaled AI workloads

0:07:26.800 --> 0:07:29.360
<v Speaker 1>or inference art workloads where we can tune them because

0:07:29.360 --> 0:07:33.320
<v Speaker 1>we kind of control that whole stack, control the training chips,

0:07:33.560 --> 0:07:36.200
<v Speaker 1>the data centers, and the inference stack, and can really

0:07:36.240 --> 0:07:39.080
<v Speaker 1>tune that whole stack. And that helps customers lower their

0:07:39.120 --> 0:07:42.400
<v Speaker 1>costs and improve performance. And we're really starting to see

0:07:42.440 --> 0:07:46.920
<v Speaker 1>that flywheel go where model providers of all types and

0:07:47.280 --> 0:07:50.920
<v Speaker 1>customers love that they get that value and great performance.

0:07:50.960 --> 0:07:53.640
<v Speaker 1>And we think that that's a real flywheel that we

0:07:53.640 --> 0:07:56.240
<v Speaker 1>can keep turning and are quite excited about continuing to

0:07:56.240 --> 0:07:56.960
<v Speaker 1>grow that business.

0:07:57.880 --> 0:07:59.760
<v Speaker 3>Could you make it tangible for me? What is the cost?

0:07:59.800 --> 0:08:03.680
<v Speaker 1>Say, it's workload by workloads, so there's not like a

0:08:03.720 --> 0:08:07.480
<v Speaker 1>specific cost savings, but for workloads where we see this

0:08:07.520 --> 0:08:12.200
<v Speaker 1>optimization happen, customers can oftentimes save twenty thirty percent off

0:08:12.240 --> 0:08:15.000
<v Speaker 1>of their inference costs when they run that on trainium interesting.

0:08:16.040 --> 0:08:19.520
<v Speaker 2>Would you consider selling the chips outright as opposed to

0:08:19.600 --> 0:08:22.040
<v Speaker 2>renting capacity based on them.

0:08:22.160 --> 0:08:24.400
<v Speaker 1>Yeah, we've mentioned that we might consider that in the future.

0:08:24.440 --> 0:08:26.280
<v Speaker 1>Right now, we have so much demand inside of the

0:08:26.320 --> 0:08:29.760
<v Speaker 1>AWS cloud that we really love that business and so

0:08:29.840 --> 0:08:32.520
<v Speaker 1>that's the place we're focusing right now. But it's an

0:08:32.520 --> 0:08:35.520
<v Speaker 1>interesting opportunity and it's something that we would definitely consider

0:08:35.520 --> 0:08:36.000
<v Speaker 1>in the future.

0:08:37.040 --> 0:08:40.559
<v Speaker 2>The other big development was AWS signing the Open Weights letter.

0:08:41.440 --> 0:08:44.720
<v Speaker 2>You yourself, you know, communicate up that. Why was it

0:08:44.760 --> 0:08:47.760
<v Speaker 2>important to you, Matt to sign that and participate in that?

0:08:48.960 --> 0:08:49.240
<v Speaker 3>Yeah?

0:08:49.600 --> 0:08:51.760
<v Speaker 1>I think that Again, a lot of this boils down

0:08:51.800 --> 0:08:55.760
<v Speaker 1>to customers wanting choice, and the big frontier labs have

0:08:55.840 --> 0:08:58.760
<v Speaker 1>awesome models today, whether it's open AI running on bedrock,

0:08:59.040 --> 0:09:01.880
<v Speaker 1>whether it's anthropy running on bed Rock. Customers really love

0:09:02.000 --> 0:09:04.640
<v Speaker 1>using those models, but they also like being able to

0:09:04.679 --> 0:09:08.559
<v Speaker 1>customize models, and so having a broad ecosystem of open

0:09:08.600 --> 0:09:13.040
<v Speaker 1>weights models, I think is incredibly important for innovation. It's

0:09:13.040 --> 0:09:15.680
<v Speaker 1>important for our customers, and so you think about things

0:09:15.720 --> 0:09:18.520
<v Speaker 1>like the Neematron models from Nvideo, or you think about

0:09:18.880 --> 0:09:21.400
<v Speaker 1>Kimmi three or some of the Chinese open weights models.

0:09:21.640 --> 0:09:24.480
<v Speaker 1>Those really help customers be able to customize to their

0:09:24.520 --> 0:09:27.720
<v Speaker 1>own data and really be able to innovate. And so

0:09:28.080 --> 0:09:31.640
<v Speaker 1>we think that it's important to not overlegislate there and

0:09:31.800 --> 0:09:34.640
<v Speaker 1>give that flexibility for customers to use whichever models they

0:09:34.640 --> 0:09:37.040
<v Speaker 1>find to be the best fit. And that's actually why

0:09:37.280 --> 0:09:39.559
<v Speaker 1>we really focus on having all of those available inside

0:09:39.559 --> 0:09:41.520
<v Speaker 1>of Bedrock, so the customers can choose which ones they

0:09:41.520 --> 0:09:42.120
<v Speaker 1>want to use.

0:09:42.679 --> 0:09:45.480
<v Speaker 2>Not to over legislate. I mean, I had the conversation

0:09:45.720 --> 0:09:49.000
<v Speaker 2>with Jensen Wong the day the letter was posted about

0:09:49.480 --> 0:09:51.640
<v Speaker 2>why then, what was the rationale? It does seem like

0:09:51.640 --> 0:09:55.520
<v Speaker 2>the concern is that the US government overly regulates open models.

0:09:55.640 --> 0:09:57.439
<v Speaker 2>Was that sort of a motivation for you?

0:09:59.120 --> 0:10:01.080
<v Speaker 1>Yeah, I think that's the concern, and I think you

0:10:01.160 --> 0:10:02.960
<v Speaker 1>just want to make sure that we kind of have

0:10:03.000 --> 0:10:05.720
<v Speaker 1>an even playing field, and it doesn't mean that there's none.

0:10:05.720 --> 0:10:07.840
<v Speaker 1>By the way, I think there should probably be the

0:10:07.880 --> 0:10:10.520
<v Speaker 1>same level of oversight of both frontier models as well

0:10:10.559 --> 0:10:12.120
<v Speaker 1>as open weights models, and I think we should have

0:10:12.120 --> 0:10:16.680
<v Speaker 1>a consistent framework, but kind of making sure that folks

0:10:16.720 --> 0:10:19.560
<v Speaker 1>realize how important that is to what customers in our

0:10:19.559 --> 0:10:21.960
<v Speaker 1>industry is out there building on and that those open

0:10:22.000 --> 0:10:23.560
<v Speaker 1>weights are a key building block of that, and we

0:10:23.600 --> 0:10:26.880
<v Speaker 1>want to make sure that, however legislation lands, it lands

0:10:26.920 --> 0:10:28.080
<v Speaker 1>evenly across all of those.

0:10:29.440 --> 0:10:31.960
<v Speaker 2>Matt a question from our Bloomback Tech audience for you

0:10:32.080 --> 0:10:36.120
<v Speaker 2>is aws's attitude towards Kimmi K three. The open weight

0:10:36.240 --> 0:10:40.040
<v Speaker 2>was released July twenty seventh. You know, given the platform's

0:10:40.040 --> 0:10:42.080
<v Speaker 2>AWS offers, what is the plan there.

0:10:43.600 --> 0:10:47.360
<v Speaker 1>Yeah, it's a great model. The team there has done

0:10:47.360 --> 0:10:49.640
<v Speaker 1>a really nice job and we see really great performance

0:10:49.679 --> 0:10:54.640
<v Speaker 1>there and we'll continue to support in Bedrock offering the

0:10:54.720 --> 0:10:57.800
<v Speaker 1>Kimmi models. I think the interesting thing is also as

0:10:57.840 --> 0:11:00.760
<v Speaker 1>you see many of these open weight companies, they're starting

0:11:00.760 --> 0:11:02.560
<v Speaker 1>to think about how they make money in these models

0:11:02.600 --> 0:11:05.840
<v Speaker 1>as well, and so they're starting to introduce licensing around

0:11:05.880 --> 0:11:08.600
<v Speaker 1>some of these open weights models as when customers want

0:11:08.640 --> 0:11:11.200
<v Speaker 1>to run them in a cloud environment, and so I

0:11:11.280 --> 0:11:13.240
<v Speaker 1>anticipate that there's going to be a blending of some

0:11:13.280 --> 0:11:16.360
<v Speaker 1>of these models too, where these companies won't continue to

0:11:16.440 --> 0:11:18.920
<v Speaker 1>spend lots of money and then just offer their IP

0:11:19.160 --> 0:11:21.600
<v Speaker 1>up to the world. They're going to be licensing models.

0:11:21.640 --> 0:11:23.319
<v Speaker 1>They're going to have ways of making money on those

0:11:23.320 --> 0:11:25.520
<v Speaker 1>as well. And you're starting to see that with some

0:11:25.640 --> 0:11:28.680
<v Speaker 1>of the open weights models actually having some licensing around them.

0:11:29.360 --> 0:11:32.760
<v Speaker 1>And so we'll keep supporting these in bedrock and working

0:11:32.800 --> 0:11:35.720
<v Speaker 1>with those companies in order to offer these in the

0:11:35.760 --> 0:11:36.559
<v Speaker 1>best possible way.

0:11:38.559 --> 0:11:40.920
<v Speaker 2>That's why I'd like to end the conversation the business

0:11:40.960 --> 0:11:45.000
<v Speaker 2>case and economic opportunity for open models from aws's perspective.

0:11:45.440 --> 0:11:47.960
<v Speaker 2>So if the model makers themselves would like to make money,

0:11:48.120 --> 0:11:50.840
<v Speaker 2>you know, how does AWS see that going in your

0:11:50.920 --> 0:11:54.800
<v Speaker 2>favor to make money from wide use of open models?

0:11:54.840 --> 0:11:58.320
<v Speaker 1>Yeah, yeah, I mean we basically that is exactly what AWS.

0:11:58.400 --> 0:12:00.960
<v Speaker 1>It's a great platform for company needs to come and

0:12:01.080 --> 0:12:05.280
<v Speaker 1>offer their IP and their capabilities to the world. And

0:12:05.320 --> 0:12:08.280
<v Speaker 1>so everybody, whether they're startups, whether they're governments, whether they're

0:12:08.520 --> 0:12:12.120
<v Speaker 1>large enterprises, everybody can build on top of AWS. And

0:12:12.200 --> 0:12:14.839
<v Speaker 1>if they have choice and they have access to these

0:12:14.880 --> 0:12:18.160
<v Speaker 1>different models. It allows companies to be able to come

0:12:18.160 --> 0:12:20.760
<v Speaker 1>and monetize them and sell when they have value to customers,

0:12:21.280 --> 0:12:23.199
<v Speaker 1>to be able to offer value when they think that

0:12:23.240 --> 0:12:25.680
<v Speaker 1>they can go at lower prices or better performance. And

0:12:25.720 --> 0:12:28.440
<v Speaker 1>so having that open platform allows everybody to have that

0:12:28.520 --> 0:12:32.400
<v Speaker 1>competitive chance. And AWS is a fantastic partner for all

0:12:32.480 --> 0:12:34.720
<v Speaker 1>these model providers to get access to the broad set

0:12:34.720 --> 0:12:37.959
<v Speaker 1>of customers. And so it's a great ecosystem where model

0:12:37.960 --> 0:12:40.760
<v Speaker 1>providers benefit and customers benefit by having all of these

0:12:40.760 --> 0:12:41.440
<v Speaker 1>in the same place.

0:12:42.520 --> 0:12:46.720
<v Speaker 2>Matt Gorman, COOFAWS Amazon Web Services back on blom bug Tech.

0:12:46.800 --> 0:12:47.640
<v Speaker 3>Thank you very much.