- 11 hours ago
Category
🗞
NewsTranscript
00:00You've built Quik, AI that works across all applications, but I think that the main issue
00:05or point is that it builds context over time. What were you solving for with Quik?
00:11Yeah, Quik is the way for companies to get AI and an AI assistant in the hands of every employee
00:18that they have across the company, and it allows customers to leverage data across all of their
00:24enterprise. But excitingly, today, we're launching a desktop application, and I personally have been
00:29using it for the last couple of weeks, and it's the single best AI productivity enhancement I've
00:34tried of anything. It's really incredible. It allows you to work across your email, across your
00:37Slack, across all the information on your desktop, quickly pull information across various different
00:42settings, and it learns with you as you go. So it learns who your co-workers are, learns who you
00:46operate with, learns who you write to. It's an incredible productivity enhancer, and it's something
00:52that we're launching today and really excited about. AWS is pushing deeper into applications.
00:59In the agentic AI context, that is everything from healthcare, hiring, supply chain. What was the
01:08need that you saw? Because a lot of the customer base of AWS will offer something quite similar.
01:14Yeah. Look, we think, and it's true, many of our customers are going to reinvent many of the
01:18applications that are out there, but we see that with agents out there today in AI, every application
01:24is being remade. And with Amazon Quick, we saw that opportunity to remake personal productivity
01:30inside of an enterprise setting. And with Connect, we're doing the same across many enterprise
01:36applications, whether it's with supply chain, whether it's with contact centers, whether it's with
01:40healthcare. We see the opportunity to apply AI and agents to a number of these applications. And these are
01:46areas where Amazon has a lot of expertise and where we can add that expertise into customers using AI to
01:53completely reimagine many of these traditional 30, 40, 50-year-old processes that companies use.
01:58It makes you think of names like Salesforce, ServiceNow, SAP, key customers. What kind of
02:04conversations did you have with them about why the broad suite through Amazon Connect is useful?
02:11Yeah. And many of these, the things that we're building are actually quite complimentary to many
02:15of those partner offerings. And so I think if you think about actually Connect, Connect and Salesforce
02:21have had a partnership for a really long time where many customers use those together in order to
02:26provide even better outcomes for their customers. And we think that's going to continue to be the case
02:31today where things like Connect decisions is really a helpful add-on to customers that are doing supply
02:39chain. And it's an agentic helper to help you make great supply chain decisions, but use it together with
02:45your existing supply chain solution. And so it's really complimentary to many of those products that
02:49our customers and our partners build on top of AWS today.
02:53What's the strategy and goal for AWS in pushing deeper into the application layer? What do you think
03:00will change for you rather than what you just outlined on the customer side?
03:04We've always thought that AWS, and we always have, produced offerings for customers at every layer
03:09of the stack. We obviously excel at the infrastructure layer where AWS Cloud is by far the largest and the
03:16most established. And we've also built services higher up the stack, things like databases,
03:21things like analytics services, things like AI services. And increasingly with Connect for the
03:27last 10 years, we've been offering applications in the top layer. And we've always said there's
03:32millions and millions and millions of customers who build applications in that top layer. We always
03:36think that we'll build a few. Some of those will show customers things that they're able to do with
03:42technology. Sometimes it's to expose areas where we're experts and we think that we can provide value
03:48to customers. But we always know that there's going to be lots of players up there at the
03:52application layer.
03:53What AWS has talked about is AI revenue run rate, which, getting into detail, seems to focus a lot on
03:59the frontier models, how they're accessed, how they're used. But if you think about Quick Connect applications,
04:06is that something, Matt, that you see making meaningful contribution in the AI revenue bucket?
04:10Absolutely. You're absolutely right. A lot of our AI revenue today is on Amazon bedrock. And it's
04:16through accessing frontier models. But we see that every application in the world is going to be
04:22remade with AI. We think agents are going to totally change how work is done, how jobs are efficiently
04:28done. And we think we have some opportunities to have some unique offerings in that space that really
04:34change the game. And so we think that Amazon Quick is one of those. We think Connect is one of
04:39those.
04:39We think Kiro on the coding side can really help people. And when you have some of these
04:44applications, it often inspires others to build around them. And we think it's a great opportunity
04:48for us where we think we have differentiated ideas to really change how things have been done for a long
04:53time.
04:53Open AI.
04:54Yeah.
04:55What's new? What's changed in your relationship?
04:58Yeah. Well, we're quite excited about the partnership that we're announcing with them today.
05:02So, yeah. So, for a long time, when we built Amazon bedrock as a way for our customers to access
05:09frontier models and access AI models, we've always started with the position that we wanted to offer
05:15choice. And we wanted to offer all of the best models available out there.
05:19And today, we're excited to be bringing OpenAI into bedrock. I think it's something that our
05:25customers have asked for for a really long time.
05:27And we're talking specifically about frontier models, not just open-weighted models, which
05:32was the extent of the relationship.
05:32That's right. We've had open-weight models for a while, but yes, this is the frontier model. So,
05:37starting in preview today, we have OpenAI's model 5.4 and 5.5 is coming in the next couple of
05:44weeks.
05:45We're also collaborating on a new offering, which we called Managed Agents, featuring OpenAI,
05:51which is a complete managed agent capability so customers can really easily make stateful agents
05:56and build agentic applications together with Bedrock and OpenAI. And it's really just the start
06:02of a long-term partnership that we've established together. As the teams have gotten together, we see
06:07opportunity for us to really invent new capabilities for our customers to go and build interesting
06:12applications together and really excited about working together with the OpenAI team
06:18and unlocking more things that customers can build on AWS.
06:22Think back to 2023. OpenAI's models were only available on Azure, right? Since then until now,
06:32what kind of behaviors did you recognize in the AWS customer base? Like, did they literally open
06:39Azure accounts so that they could have access to something that wasn't available in AWS? And that
06:45was kind of the indicator to you that you needed to make all models available? Our customers have long
06:50told us they want to be able to use the absolute best cloud with the best set of models available.
06:55And
06:55they wanted to run all their applications in AWS. And they weren't able to run, and they did every,
07:00and they did. Oftentimes, our customers run every part of their application in AWS. And then they'd access,
07:05if they wanted to use the OpenAI models, they'd either use them direct from them in their first
07:08party or through Azure or somewhere else. Now they don't have to make that trade-off. And
07:13with the OpenAI team and the AWS team, we're really excited to not have to force customers into that
07:19choice. And if they want to build all in the AWS set of tools and capabilities, it's where their
07:25applications run. It's where most customers store their data. It's where they feel most secure. And they can now
07:31run that together with OpenAI models, together with the whole suite of models that we have in AWS,
07:37including Anthropic models, including Open models, including Nova models, a whole set of models. And
07:43they can combine the best of what they want to do in order to build their applications.
07:46It's early days. I appreciate that. But what is the sort of one metric you could point me to that
07:53gives you the conviction there is absolutely demand there at the enterprise level or otherwise for OpenAI's
08:00models? Yeah. Well, look, I think the team announced a statistic recently that Codex, which is also
08:08available in AWS in preview starting today, in the last couple of weeks has grown from 2 million users
08:14to 4 million users in just a really short amount of time. That is incredible. I mean, that is really
08:19unprecedented growth. And it just shows that they're building really, really capable opportunities out
08:25there. But our customers still say, I would love to access and power Codex through Bedrock. So I have
08:31the security of Bedrock and AWS. Now with the launches today, all of those millions and millions
08:36of customers that are loving Codex can choose Bedrock as the provider and get the AWS security that
08:41comes behind that. I think about Claude as a comparison or case study, right? Yeah. What was so
08:46interesting about Claude through Bedrock is Anthropic got the benefit of a platform. That's right. You
08:52channeled business to them. That's right. But equally, people want to use Claude. So that channeled
08:57business to Bedrock. That's right. Do you see that playing out similarly with OpenAI? That's exactly
09:02what one of the things the OpenAI team is so excited about is partnering with AWS is that exact
09:06partnership. And I think it's one of the things that we do well at AWS is partner with folks that
09:12we're building together with. And just like with Anthropic, which we have a fantastic partnership with,
09:17and we've really built this business together with Anthropic where we lean into them. We help their
09:21customers build applications on top of AWS and find out the best ways to use their Claude models.
09:27And we've seen their business really take off. And that's what we anticipate doing with the OpenAI
09:31team as well. Have you, Matt, had to make an assessment of how much of the spending that Amazon's
09:36committed to is literally dedicated to OpenAI infrastructure-wise to support the ramp up of
09:43of availability? Yes. Well, we absolutely are expecting a lot of growth there, which we're
09:48quite excited about. And OpenAI has made a large commitment to AWS to use Tranium in particular,
09:54but across our AWS services to grow their business. And so, yes, they're strongly committed to growing
10:01their footprint on top of AWS. And we expect Bedrock to grow really rapidly as well. Does the demand meet
10:07the infrastructure that's been put in place? So where are we in the balance between future supply
10:13and current demand? There is still, I think, more demand out there than there is supply. Right. For
10:18sure. And is that true of OpenAI? I appreciate we're in preview as you speak. Well, yes. It's the
10:24first day, so I can't tell you that, yes. But yes, the OpenAI team, I think, would gladly take more
10:30capacity from us this year and next year and the year after that as we add it. And so we're
10:35still working
10:35really hard to continue to add more capacity, whether it's power, whether it's chips, whether
10:40it's memory, all around the world. And I think that we're planning for success across many dimensions
10:47there. But yes, today, the world is still largely supply constrained when it comes to AI capacity.
10:54We are in a strange situation where OpenAI models can now run on AWS, but Microsoft will benefit
11:01financially because of the latest terms of their agreement. They've ended exclusivity, but OpenAI
11:08continues to make payments to Microsoft. Yeah. How do you think about that? Oh, that's okay.
11:14Look, Microsoft has benefited from the growth of AWS since the very early days. In fact,
11:20we've supported Windows licenses as an example, I think, since 2007, 2008. I can't remember the exact
11:26year. But so Microsoft has benefited. They build great software. They have good partnerships and
11:31they should benefit from those. But customers really want to use those technologies. SQL Server
11:36is a good example that runs great on AWS. In fact, many of our customers tell us that SQL Server
11:40runs
11:41better, way better on AWS than it does on Azure. And so for those customers, running in AWS is where
11:47they
11:47get the best reliability, the most security and the broadest set of features. And so that's great. And if
11:52Microsoft benefits because their software or their partnerships or their licenses are being used
11:56inside of AWS, that's great. And in this case, we partner with Microsoft, just like we partner with
12:02Oracle and other providers that build software. And we want to make all capabilities available on AWS
12:08for people to build. You built bedrock and stateful runtime at a time where access to models was
12:19constrained. Now, if open AI is the case, it's available broadly everywhere. You know, what's
12:26the benefit of accessing it through AWS? What's the pitch? Yeah, you mean for the and we're calling
12:32this managed agents is managed agents, we called it stateful runtime before, but the launch name of
12:37it is called managed agents. And it's just we're just as excited about that as we've ever been, by the
12:41way. So that that that product is something that we why does a customer need to use AWS? That's the
12:46simple question. Why do you want to use AWS? Well, this is the only place that we'll have this
12:49managed agents products together with with open AI. And, and we think it's going to be the single
12:55best and easiest way for you to go and build agentic applications that learn with you over time that
13:00get better over time. It's a really simple way to it includes kind of this agentic harness and a bunch
13:06of these pieces together that makes it really easy to build agents that leverage the latest frontier
13:11models from open AI. And so we're very excited about that product. We think that that is going to take
13:16off like gangbusters, it's launching and preview today as well. And it'll be GA here in the next
13:20couple of weeks. And we think that that customers are going to really love how easy it is to build
13:25agentic applications with that product.
Comments