In this keynote analysis from AWS re:Invent 2025, theCUBE’s John Furrier joins analysts Paul Nashawaty, Zeus Kerravala and Sarbjeet Johal to unpack how Amazon is redefining cloud infrastructure through the lens of agentic AI. The panel breaks down Matt Garman’s declaration that "agents are the new cloud," exploring key announcements surrounding the Nova model family, AgentCore and Amazon Bedrock. The discussion highlights AWS’ strategic pivot from merely abstracting infrastructure complexity to abstracting work itself, effectively bridging the gap between professional coders and "citizen developers" while unifying the experience for builders at every level.
The conversation digs deeper into the practical realities of enterprise AI adoption, emphasizing the critical role of security, governance and compliance in moving from proof-of-concept to production. Kerravala, Johal and Nashawaty analyze AWS’ vertically integrated approach – spanning from custom silicon like Trainium and Inferentia to the application layer – and how this full-stack strategy allows customers to train models on proprietary data with improved price-performance. The group also debates the evolving competitive landscape, noting how AWS is equipping organizations to build autonomous, long-running agents that function as teammates rather than just tools.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS re:Invent 2025. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AWS re:Invent 2025
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AWS re:Invent 2025.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
AWS re:Invent 2025. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to AWS re:Invent 2025
Please sign in with LinkedIn to continue to AWS re:Invent 2025. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Julia White, AWS
In this keynote analysis from AWS re:Invent 2025, theCUBE’s John Furrier joins analysts Paul Nashawaty, Zeus Kerravala and Sarbjeet Johal to unpack how Amazon is redefining cloud infrastructure through the lens of agentic AI. The panel breaks down Matt Garman’s declaration that "agents are the new cloud," exploring key announcements surrounding the Nova model family, AgentCore and Amazon Bedrock. The discussion highlights AWS’ strategic pivot from merely abstracting infrastructure complexity to abstracting work itself, effectively bridging the gap between professional coders and "citizen developers" while unifying the experience for builders at every level.
The conversation digs deeper into the practical realities of enterprise AI adoption, emphasizing the critical role of security, governance and compliance in moving from proof-of-concept to production. Kerravala, Johal and Nashawaty analyze AWS’ vertically integrated approach – spanning from custom silicon like Trainium and Inferentia to the application layer – and how this full-stack strategy allows customers to train models on proprietary data with improved price-performance. The group also debates the evolving competitive landscape, noting how AWS is equipping organizations to build autonomous, long-running agents that function as teammates rather than just tools.
In this interview during theCUBE's coverage of AWS re:Invent, Julia White, chief marketing officer of AWS, joins theCUBE’s John Furrier to break down the major announcements reshaping the enterprise AI landscape. White details the launch of "AI Factories," an opinionated infrastructure approach designed to bring large-scale compute capabilities directly to customers' existing data centers, specifically for highly regulated and sovereign needs. She explains how the new Nova Forge empowers organizations to create custom frontier models by securely blending thei...Read more
exploreKeep Exploring
What is the concept being discussed regarding AI factories and how does it relate to customer needs?add
What is the connection between the Trainium news and its impact on customer optimization and performance in AWS?add
What are the key advantages of using AWS for building agents and how does it relate to customer needs and marketing strategies?add
What are the benefits for companies and partners utilizing AWS technology?add
What are the key considerations for scaling AI agents within an enterprise to ensure they provide real customer-facing value?add
>> Hi, I am John Furrier with theCUBE. We are here in Seattle, Washington to get a breakdown of all the news and announcements at re:Invent. A lot to unpack. Julia White's here, theCUBE alumni. Welcome back. I'm still on your home turf and it's away game for me. Great to see you.
Julia White
>> Great to be back. Love it.
John Furrier
>> Chief Marketing Officer at Amazon. You have the challenge and opportunity to present the great news that's out now in the wild. Pretty needle moving in Amazonian terms, game changing on a couple fronts. One, I love the AI factory kind of positioning that leans more towards cloud plus next generation. But the frontier agents really is I think a real level up, legit next level kind of capability. We've seen the agent hype, people are playing with agents, we're seeing some use cases. And then the Nova News I think will democratize model building and bring what is very expensive proposition to a data estate. So anyone who has data, it brings... So huge news. Where do you want to start? I mean, I love AI factories since I've been talking a lot about AI factories first.
Julia White
>> All right. Let's go there.
John Furrier
>> What does an AI factory service mean for AWS?
Julia White
>> Right. Well, as you know more than anyone that a concept of AI factories is a big conversation. So what we're bringing forward is a very opinionated AWS AI factory that gives customers that ability to build out the AI infrastructure they need, particularly for highly regulated sovereign needs on a very large scale, for that class of customers. And we're bringing our expertise to their infrastructure. So with their data center capabilities, with their GPUs, we can bring ours, but our services and running that. We've had almost 20 years of know-how of how to do this at scale better than anybody. So fusing that with this kind of unique customer optimization and how we do it is the genesis of that idea.
John Furrier
>> I mean it's interesting because AI factories is to me just a bunch of, the way people are building out in the market, you see other people doing it, it's a lot of servers. And the cloud, you guys have done that. Amazon's core competency is building large scale systems. I mean the networking is hot now on some of these large scale systems. You've been doing networking in your systems. But the other core competency I find interesting is that you're building data centers too. You're actually deploying billions of dollars. I think we've covered that news. Is that tied in? Are you going to bring data center, like large scale factory data center, Amazonian service level to people to build data centers or to use Amazon services in your data centers?
Julia White
>> What are we doing here?
John Furrier
>> Yeah.
Julia White
>> So let's step back. What's the need for the customer? We have these really big customers who need large scale AI capabilities, but they need it customized to them. They need it in their geography for their organization, those boundaries that are really important to hold. But they don't want to create this whole other thing that doesn't work with the rest of their systems. They don't want... A lot of them are obviously AWS customers. They'd like it to all work seamlessly together. And so that's what the AWS AI factory is, bringing and providing that large scale compute, but within the customer's existing data center footprint so it can work as an extension. And we're in the same AWS services like Bedrock and model choice there, SageMaker, those type of things, AgentCore, so they can build in a consistent way but have that dedicated infrastructure set for them.
John Furrier
>> So building systems to the on-premise with the customer or building data centers for them, or both?
Julia White
>> Really working with their existing footprint, because that's where customers are starting. And these are very much customer-specific implementations. So there'll be a variance and I'm sure over time we'll evolve how we do it.
John Furrier
>> It's an extension to the cloud. That's awesome. I love the AI factory thing. It's going to enable a lot more tokens. And that's going to bring in the agent wave, so all this stuff. Trainium has also got performance news. That was hot news. That's just more speeds and feeds. But I want to talk about the Nova model first and then that ties to Trainium. I think this will go there.
Julia White
>> Okay. I'll go that way.
John Furrier
>> Nova has got a new thing called Forge. This is unique. Explain the thinking behind it and why this way.
Julia White
>> So again, start with what's the customer need. Customers of course have been working to fine-tune models and get them to be more and more specific to their domain and their specific business. That's where you get the best value out of it. But there's limitations to how much you can do with fine-tuning. And if you do too much, the model actually kind of loses what it was originally trained on. So you have this kind of tension in the system. So we stepped back and said, how could we fundamentally change that and not have to compromise as it's been. So that was the invention, is the invention of Nova Forge, which is the first ever ability for customers to take foundational models, Nova, and bring their own data and mix it with Amazon-provided data and actually start training the model at whatever checkpoint within Nova they want based on what they're trying to do. And it gives the customers that first ever ability to say, hey, here's my data that I want to train on, here's the Amazon data, and get that exact frontier model that's just for them. So really customers now for the first time ever can very effectively build their own unique frontier model, which is amazing.
John Furrier
>> And frontier is defined as large-scale, reasoning, multi step, thinking.
Julia White
>> Right. Super powerful.
John Furrier
>> And expensive to build.
Julia White
>> Right. They couldn't build them on their own.
John Furrier
>> I mean, I think this is a democratizing opportunity. But I want to just go back a little quick because what you're doing is the customer's needs flipped. Now, the biggest complaint of the large-language models, frontier models was I don't want to put my queries in there because-
Julia White
>> Right. I'm sharing the data.
John Furrier
>> And every company's like don't use the models. So they're afraid to cross-pollinate. We've seen use cases there. You change that with this. You bring, hey, just use it.
Julia White
>> This is all isolated within-
John Furrier
>> You flipped the script.
Julia White
>> Right. It all stays within the, it was a VPN. It isn't blending data that goes out into the world. It stays just to that customized frontier model for that customer. Again, we really work back from what are the limitations and what are customers trying to get out of models and how do we bring the assets that we have with Nova, with the training data, with ability from AWS, massive customer data to bring that together into a .
John Furrier
>> I was talking to Matt about it. I'm like, "It sounds like it's like a software development kit for models." It's like here's the Nova model, put your data estate in there, and play with it, figure it out.
Julia White
>> But even us bringing Amazon data for the customer to use as well.
John Furrier
>> Explain that piece.
Julia White
>> So basically we have the SageMaker HyperPod loaded with the Amazon data that we use to train Nova with. So the customer can actually, again, really find that right blend of their own data and then the Nova training data, so it doesn't forget everything it was originally trained on and it can blend in the right way.
John Furrier
>> That's amazing because I think one of the things that's come up from our current reporting on the large scale factories and environments now is the number one concern besides some of the stuff like liquid cool and make sure things don't melt is GPU utilization or XPU utilization. And a lot of it comes down to flawed model architecture. So the question is, will customers see a performance increase or is Trainium tuned to this? How does the Trainium news fit into this? Is there a connection?
Julia White
>> Well, I do think when a customer is able to get exactly the model they want, it will be optimized in that way too. So you're not spending compute on things that are not useful or you're overfunctioning in areas as well. So that will certainly help. But if we flip to Trainium news, I mean I think just the momentum around Trainium has been amazing. And seeing our third generation show up and just continuing, just absolutely crushing that price performance promise for our customers. And then even Trainium 4, announcement of what we're already underway with and the performance we're seeing there. And this is again, just a unique differentiator for AWS.
John Furrier
>> Share some of the numbers because a lot of people, they might not understand the level, and you have a lot of data. The Anthropic relationship, has been working with Trainium, all the internal Amazon and AWS work. What's, scope the performance and cost savings gains?
Julia White
>> If you look down from a cost perspective or from a memory perspective, from a performance perspective, we're literally getting, we're having it and then having it again as we go forward into each of those, in some cases even more, right up to 80% from a price performance perspective. So huge. Each one of these generations is big, it's not incremental, big step function changes. And obviously that's a huge, huge driver of what people can do and how -
John Furrier
>> And managing that with the infrastructure tied to the software allows for the AI to help.
Julia White
>> Well, think about, because these are our chips, we can just absolutely optimize every last aspect of what we do across the infrastructure from top to bottom, which again is very unique for any provider.
John Furrier
>> Okay. The final news piece, the big pillar is the frontier agents.
Julia White
>> Yes.
John Furrier
>> Again, back to the word frontier. Explain why the word frontier model to their data with Nova, you did that already, but now frontier agents, explain what that means because this is a little nuanced.
Julia White
>> Yeah, it's a new concept. But it really stemmed from as we were building out these agentic capabilities, seeing that if you could do at least three characteristics, it creates an enormous capability jump and performance jump and outcomes jump. And the three capabilities that we really learned by doing and seeing and experimenting was one, completely autonomous. You can set it at a very ridiculously ambiguous task and it can just run and decide and figure out. So that was the first characteristic. And then it was about just massively scalable and then scale up and scale out and that kind of lingo where the agent can add multiple of a certain type or it can actually become more diffuse in both ways, and so massively scalable in that way. And then the third one was long-running, hours, weeks that it can just go and work without having to be constantly redirected or having an intervention. And so when you put those three together, both the inputs and the outputs are just game changing. And so we felt like this is a new class of agents, hence the frontier agent terms.
John Furrier
>> It's my favorite announcement because I can see the value. We were talking to Deepak and some other senior engineers at AWS, and the capabilities that Amazon is bringing to ages has been great. And the adoption by developers on Kiro, for instance, has been phenomenal. And some of the conversations were, what if you could have an agent that could know the entire AWS estate? And so you got Kiro, which could just build anything. So it's easy to build agents. It's kind of like the console for agents. Then you got the DevOps piece, which is a lot of the mechanics and the heavy lifting and running the cloud. And then you got security. So you have now three major agents. Was that intentional? And why that focus?
Julia White
>> Yeah. And the agents that... I think most of the world's agents are going to be built by companies around the world. And Amazon's going to build a handful of these agents. And we really are sticking to what do we really know best? Where do we have unique vantage point and insight? And obviously software development is one of those places. It's also a place where we're seeing just a lot of adoption and interest in the market, so it's a good mix. It really started from a project by one of our distinguished engineers. It was interesting, and Matt has a great story around this of the initial proposal was it's going to take 30 engineers 18 months to do this project. And then they started and they started using these agent capabilities in a different way, and they actually got the project done with six people in 76 days. Just totally different. And so we're like, wow. But then once you start writing more code and being more productive from a code development perspective, then you're like, well, we have to actually be better and faster at securing that code. And then of course we need the DevOps capabilities to manage and run that code. And so it was a little bit of a step-by-step. We felt like let's not just start create one opportunity and then it creates a new bottleneck. Let's solve the bottleneck of the whole software development lifecycle.
John Furrier
>> You guys can build, deploy, and manage with the agents. So you're basically taking your AWS core competency and kind of priming the pump in the market, but also giving value to the customers who are going to want to try to figure out how do I increase my staff.
Julia White
>> Right. And the capabilities and the outputs they can generate.
John Furrier
>> I mean, everyone has an Amazon team. Our team is like three people, and they'd love 10. Who wouldn't want 10 more engineers.
Julia White
>> Or a hundred. Right?
John Furrier
>> Yeah. Like, okay, what's that new service? Go test it out. All right. So how are you going to market this? Because I think a couple things I want to ask is one, I love the AI factories. So that's going to be a big pillar. You've got the frontier agents, which is going to create I think a surge in value in the enterprise.
Julia White
>> Yeah. I think they're going to see results so much faster.
John Furrier
>> What's the strategy? How are you thinking about the positioning? And to the target customer, who's the target for you? How are you thinking about the go-to-market here?
Julia White
>> Right. Well, the wonderful thing about AWS is across all the different parts of the stack, we have incredible innovation. We didn't even talk about AgentCore and the ability to build agents at scale. And so it's an opportunity that no matter which customer I meet with, you can kind of walk into like, oh, this is where you have the most need, you have the most need. So it's great that way.
John Furrier
>> It's hard to have one strategy when you have a horizontal customer base.
Julia White
>> But I think it starts with really the promise of agentic AI has been out there as a top conversation. But what I think about from a marketing perspective, this is the reality of it. We've built the capabilities and the complete ones as well. So you can see what the high watermark of what a frontier agent looks like. And then we have all the capabilities for you to build exactly what you need. So we're kind of giving you the tools as well as example outcomes and then let them go.
John Furrier
>> It's interesting, I've been seeing a lot of conversations and stories around how people are being successful in AI and how they're marketing that. And it seems to be let the work do the talking kind of approach. And I saw a story where it was about these domain specific analysts are using AI who were taking business from Accenture, not Accenture, McKinsey and BCG, which are the classic big consulting firms, with the junior leaders and then the seniors all been months and months. So what's happening is that the speed of value was what they were talking about. How could they lose business to this small little upstart boutique? And they were specialized. And there seems to be a lot of that going on where you can come in and innovate as a builder.
Julia White
>> Right. Well, think about, go back to the very beginning of AWS and cloud. When AWS was invented, a small business could have the same technology capabilities as a big enterprise. That totally democratized what people could do from any size of company. So I think the same point of this agentic approach that we have is that we're showing people what amazing looks like from an agent outcome, but we're giving every single person, big, small, otherwise, the tools to build whatever they might need, again, from the price performance at the infrastructure level to their own custom frontier model to the building blocks to create these frontier agents of their own.
John Furrier
>> Well, the keynote certainly hits a really cool future acceleration. How's the impact on the ecosystem, customer value clear? What's your view on the ecosystem partners? Are they jumping into it? What's the vibe there?
Julia White
>> Yeah, gosh, huge opportunity both to help customers build on and do all the customization work they want with it. But I think about the other great partners that are benefiting are like ISV partners or an Adobe or Rider AI who are able to take advantage of all the technology that AWS offers and then fundamentally change their experiences that they offer. You're seeing Adobe with Firefly and their creative marketing suite doing incredible things because of the capabilities AWS AI is providing them. Same with Rider AI. They're like a leader in these agentic experiences and all that built and running with AWS.
John Furrier
>> I think every company that's building on AWS just gets a step function gain with agents.
Julia White
>> Absolutely.
John Furrier
>> All right. What's your hope for next year, as re:Invent is always a great end of year event? What's your view for '26?
Julia White
>> Well, the thing I love about re:Invent but also coming out of it is just the excitement you see of customers of what they're going to go back and do. And so as we started in January, I love just sitting with the customer be like, all right, what are we doing, and see those results. So that's for me the fun part.
John Furrier
>> Well, the hardest question to ask is which one's your favorite?
Julia White
>> Oh, don't make me choose. That's the worst.
John Furrier
>> I think I love the AI factory vision. I think that's just categorically a generational term, concept. But I think as it gets reality, I think frontier is, the frontier agents I think is the one that feels like it's going to be fast. And I think the Nova one feels like rocket ship.
Julia White
>> Just a whole different way to unlock model value. But I agree. I think the Frontier agents, because they just work, it's going to be easy for everyone to see value out of them so, so quickly. And certainly some customers are going to build their own and do all that and we have all the capabilities. But I do think those Frontier agents are going to just unlock what people hadn't even thought was possible before.
John Furrier
>> Yeah, it's hard to pick a favorite. They're all good. Okay, final question on the customers. What is the biggest customer need right now? Because in the enterprise specifically, there's a huge enthusiastic alignment with the AI with the data. A lot of these news hits that. But there's been kind of a, I won't say pause, but a lot of workloads aren't getting into production. There's a lot of learning and a lot new startups coming in with kind of an AI stack. So you're seeing a lot of innovation coming in from the ecosystem, but then the enterprises have their incumbent estate. And we've seen a stall of AI adopts in the enterprise. Most of it's been, at least on the revenue side, cloud's been doing great, you guys have been doing great.
Julia White
>> Yeah, but the scaling those. Right?
John Furrier
>> Enterprise scale for net new, okay, here's five new major things, not the low-hanging fruit, RAG search and things that they've been doing, real customer-facing value.
Julia White
>> Yes. And I think that obviously talking a lot about re:Invent, but I think the key unlock, and if I just take it from where am I hearing from an enterprise sector the most excitement back in and where they're like, this is amazing and game-changing is actually Amazon Bedrock AgentCore, and these are those agentic building blocks. So that, because a lot of the early from AI and prompting now into agents like that, agents are where we're going to see the real value. I think everyone understands that. Then how do I do agents and actually scale them in the core of my enterprise, not on the fringes and these kind of nice-to-have things. And when you start doing that, you have to care about the same building blocks you need for enterprise applications, identity management, security, governance, the runtime capability, policy, all of those things. And if you don't have them, you're building them yourselves. And that's what we heard loud and clear from customers, like, "I'm spending hours and days and months of engineering time doing this grunt work. We're not getting the value." And they're like, "We can do that. That's what our sweet spot is. That's our DNA." And so the set of AgentCore fundamental agentic building blocks is what is now really unlocking, and that's what I'm hearing from our enterprise customers of like, oh my gosh, this suddenly takes us from fringe agent experiences into the fundamental core of our business.
John Furrier
>> So tinkering was happening. AgentCore gives an interface and then-
Julia White
>> Building blocks. Yeah....
John Furrier
>> the building blocks all can be done there.
Julia White
>> Yes. Yes. That gives you that enterprise platform.
John Furrier
>> And they're enthused by that and they're customer .
Julia White
>> That is where I'm seeing just incredible deep appreciation and just they're like, this is the unlock. We were doing stuff, we had ideas, but this is the unlock, this agentic platform that I can build on.
John Furrier
>> Well, a lot of people that know theCUBE, we cover re:Invent all the time. Everyone wants to know how the re:Play party is going to go. What's your...
Julia White
>> It's going to be amazing. It's always a good time. What I like about it is there's literally something for everybody. You can roller skate, you can rock out to EDM, you can smash cars. It's all there.
John Furrier
>> It's always a favorite. It's always a great way to cap the event. Thanks for coming on.
Julia White
>> You bet.
John Furrier
>> Appreciate sharing all the insights for re:Invent. All right. I'm John Furrier with theCUBE here in Seattle, getting the scoop and breaking down the news and really what it means and a lot more learning happening at re:Invent. There'll be a ton of sessions and of course we'll get theCUBE coverage. Thanks for watching.
John Furrier