Hello, I'm John Furrier with theCUBE. We are here on location in Seattle at AWS's headquarters with Matt Garman, the CEO, breaking down the future of cloud, but also looking back at this half year, kind of a halftime report for two days. It's been quite a year. AI speeds feels like a full year. Matt, great to see you. Thanks for taking the time and taking time out of your busy day. Thanks for coming on.
Yeah, happy to be here. Thanks.
I got to ask you, in the 20 years that I've worked with and covered AWS, I've never seen the speed and velocity of this year. It's kind of like a halftime report, mainly because it feels like you could probably do re:Invent right now and have all the news. It's been quite a year. Yeah, how are you feeling?
Great. And like you said, yeah, the innovation pace is at just a breathtaking rate right now, but that's what customers are excited about and I think it's not that often that a technology revolution like AI comes along that really just turbocharges what's possible. And across the stack, up and down for AWS, we're innovating both to support that from the infrastructure layer and the services as well as building brand new AI and agentic services. And customers are excited about it, we're excited about it. And like you said, it's moving like crazy.
Last time we talked before re:Invent, we talked about startups and I could see there's some progress in there with startups. Enterprise AI is hot. This year and kind of going into 2026 seems to be the year of cloud, next level cloud with collaboration with customers, partners, and ultimately their end users. So you have this kind of mix of what you guys always talk about working back, but the equation is so fast, but the collaboration with the customers is a key theme. They're all looking at reinventing their infrastructure because they have to usher in the pressure for the agentic wave, which clearly is kind of like this next SaaS kind of cloud play. This is a real factor in the customers. What are some of the things you're seeing in the business right now with the customers and in the collaborations? What are some of the top line things you talk about and what do you see?
Yeah, it's a great point. I think when we talk to customers, they're looking for us to help understand really how can they prepare their enterprise for this coming wave of change and technology. And a lot of them are thinking about and realizing that they either have to get ahead of this technology or they're going to be left behind. And so it starts with some of these customers have already migrated to the cloud, but many are on that journey. And it all starts with helping them figure out how they get their data in a cloud world and organized in such a way that they can take advantage of it. I think increasingly enterprises and startups and everyone are realizing that their data is going to be their IP. That's the thing that's going to differentiate what they can do versus others. And if that's not in a cloud world, a bunch of these things around agentic workflows, around content summarization, around AI improvements aren't going to really materialize. And so number one is they really are looking at us to get that cloud estate in place as quickly as possible. And so this is actually a huge tailwind to getting more people to move their critical data and workloads into the cloud. And then number two, and this is often in parallel because again, things are moving so fast is they want to figure out how are they going to get real business returns for this? I think as I talked to a broad swath, and this is across almost every single industry, it's fascinating. People are finally realizing that it is their industry is going to change, their business is going to change, the jobs of every employee that they have are going to fundamentally change and they have to change along with that and move quickly. But they're looking for us to see how do they actually make that change? How can we enable them to make that change? And how can we make sure that it actually has a really positive ROI so that when they look at their financials and they look up, they have benefited from this change and they're not behind the eight-ball by spending too much money on things that didn't actually deliver value.
I always joke with a customer the other day of yours, and they're also looking at a whole on-prem, basically distributed computing paradigm, which is here, finally it's large scale. They say, 'Our agent strategies are about data and tokens out. We need more tokens. Agents eat tokens for breakfast, lunch and dinner.' This dynamic of tokens becomes a really big deal in this age of reasoning because it puts more pressure on tokens and they don't have unlimited budgets. So you have these enterprises saying, 'Okay, I'm in the cloud, I'm running workloads, but now agents need to feed off these tokens, they need high performance, I need a new software kind of glue layer.' They're looking at a kind of a, not reset, but like a re-architecture because the demand for tokens is putting pressure on more large scale systems. What's your reaction to that? What's your opinion?
Yeah. Well look, I think the more these systems are able to think, they obviously require lots of compute capacity. Tokens is really just a proxy for compute time. It's just a proxy really for cycles. And that is what powers these, and these get bigger and bigger models and they think longer, they require more compute time. Now there's a couple of things that we need to do. One is we've got to bring down the cost of that compute. It's still too expensive today. It's a couple of orders magnitude cheaper than it was a couple of years ago and still needs to keep getting cheaper. And it is, and we're on that path and we're driving a bunch of things, some of it with our Silicon roadmap and Trainium too, but also we're building tools and technologies that allow people to cut costs in other ways. And so think about something like Q Developer and our QCLI, which is actually wildly popular with customers right now. When they're developing with that, sometimes I hear developers saying that they get 2, 3, 4 X performance boost gains because we're building agentic workflows across everything that they do. It's not just code completion, but it's writing documentation for them. It's writing unit tests for them, it's actually helping them reason about what the architecture of a system looks like and when you get somebody who can deliver something three X faster, the fact that it uses a little bit more compute time is actually turns out to be a rounding error. And so agentic is one of those things that's changing that math equation, but I think you're absolutely right. We've got to help them and help customers understand where they can get that value. And we're doing a ton to roll out tools and capabilities as well as infrastructure that helps customers make sure that they can get the value out of those things.
Yeah, I'll get into that a little bit later because I think that's a huge factor. I mean, EC2 made its bones in the fact that hit databases, people were running stuff on their applications, on their SaaS apps, on the cloud, agents changed the game. Let's talk about the business side. Obviously your growth, still phenomenal numbers. Talking to Andy back in the day, the number's still on a big number. So the growth rates is a function of the number. So talk about what's driving your business now. You can continue to see growth. Every earnings, we see the performance. Talk about the impact of the enterprise and startup ecosystem to the numbers.
Yeah, well look, yeah, the business is going great and I think as we think about how do we help customers move to the cloud, we're still in the early stages of that. And we keep saying this and it keeps being true and I think that the number keeps getting bigger. But it's less than 20% of workloads at this point have moved to the cloud, which it just really ... I mean, stepping back, it's a huge opportunity for both our customers and for us to actually still take advantage of that core capability of getting their data in the cloud and taking advantage of EC2 and S3 and our database services, let alone all of the great new things that we're launching now. And so as we look at that potential, there's just massive potential there. And so I would say that if you look at the adoption curve, it's not like any other industry where there's the early adopters that jump in quickly and then you've got these kind of early enterprises who are kind of leaning forward. I actually think that generative AI and agentic is pushing a bunch of that kind of middle adoption people to realize it is now an imperative. Before, it was a nice to have and maybe even a little bit more nice to have, it was going to give you an advantage, but there was work and you saw lots of customers take advantage over the last 10, 15 years of being in the cloud and they accelerated more rapidly and they were more agile and able to adapt to changes. But there was still some kind of laggards that wanted to go slower or for good reason, sometimes good or bad. I think their customers now are realizing it's a business imperative. They are going to probably cease to exist if they can't take advantage of this technology wave. And so for us, that's a really big tailwind in helping customers of all sizes, frankly, governments the biggest enterprises in the world, banks, healthcare companies, retail companies, all the way down to one person, startups that are building the coolest technology aspects.
There's two things out there I want to get your thoughts on and reaction to because I think this is kind of a tail sign of the current wave. On the startup side, I remember back in the SaaS days, well, I'm programming and I was in that boat when I started SiliconANGLE. I didn't want to put a data center out there. I just used AWS. Startups then had SaaS apps, they would build it, they'd host it on AWS, throw the app in the App Store, and then off to the races and they become big brands. Now you got the startups doing agentic things, so it feels like the same vibe. So growth in the ecosystem of say inventors and startups. And the second thing is that in the enterprise side, I hear the term unlock a lot, unlocking value from the data that they have. Talk about those two things. One, the dynamics of the developer because that developer also sits in the enterprise. Now, in the old days, you'd launch on AWS, use higher level services. Now there's all kinds of services, but they're still innovating. So the startup ecosystem looks kind of the same, but it's different. It's not as easy. I won't say easy, but the old days, it seems easy. You throw it in the App Store and you're on the cloud, but now it's like you're enterprise by default because you have all these things to deal with, if you're going to build an agent, for instance, if you were a startup. Now the enterprise has all this data, that's an unlocking opportunity. So talk about that developer mindset of this generation. And then the enterprises are sitting there asking themselves, 'How do we unlock the value of our data?'
Okay. Yeah. Yeah, it's interesting. I think more and more we're seeing a bunch of startups be interested in selling to enterprise as you mentioned. And I think they see a lot of this. I think everyone sees this potential, and I think in some ways it goes back to your unlock is an interesting phrase that people say. And I hear that a lot too. I think a lot of this goes back to the companies realizing that their data is the key piece of value that they're sitting on that can really change the game of how they drive more value from their enterprise. And in a world where technology is changing so fast, the barriers of moving from enterprise A to enterprise B or whatever, the value that they really get is that data that they have because that is where they're going to differentiate. And so there's a bunch of startups that are moving from not just building consumer applications where a lot of startups were back in the day, but just really focusing on how do I help enterprises move quickly and get value from that data? And some of that is bringing sources of data together. There today it's been locked up in these kind of vertical data stores that they couldn't combine together. And these companies are really realizing that if you can combine the data from my ERP system with the data from maybe my HR system, which the data from my internal wikis from my sales system, et cetera, there's some super interesting information there. I can make my employees all much more effective and move more quickly and I can go deliver massive value to customers. And so it's a cost reduction play, it's an efficiency play, and it can go faster. And I think that you're right. I think a bunch of startups are building really interesting, fascinating things to try to help that unlock.
And props to you. You talked about startups before re:Invent and seeing some momentum. You guys are investing in startups. It's great to see the DNA back, the original Amazonian vision. In terms of the business infrastructure, large scales come up, sovereign cloud is a topic. You have countries who want to have certain protections. What does the footprint look like right now? I know there's news breaking recently about North Carolina expansion, $10 billion. You've had region deployments. I mean, on LinkedIn, almost every other week, you're announcing another region. Where are we on the footprint? What does that mean for customers?
Yeah, there's a couple of things I'll say and I'll touch on the sovereign bit as well. So we're still rapidly expanding both into new regions as well as growing the regions that we have and growing that footprint and the investment that we're making in North Carolina is part of obviously growing our U.S. footprint to support again the massive growth that we're seeing from both AI footprints as well as our core computing environments. And as you said, we're growing internationally as well. We recently launched a region in Thailand. We recently launched a region in Mexico. We announced a new region in Chile. We have many regions coming online and they continue to add more and more of them because customers for a couple of reasons. One is they want to operate locally. And whether you're a bank or a utility or a government workload, they want those workloads to run in-country. And the customers that are actually from a latency perspective get better performance if their applications run and support the local citizens there in that country. There's one other region that we're launching, which we will be launching at the end of this year, which I'm very excited about, and it'll be the first of its kind called the European Sovereign Cloud, and that is a bit of a different angle, which is in Europe we saw a lot of countries and companies in those countries concerned about sovereignty at another level. They're really worried about how do I ensure that this region can really run independently if disruptions happen anywhere? And so for that, we actually launched this year, we're building this European Sovereign Cloud, and the idea there is that everything is fully self-contained. It has its own account structure, it has its own login information and IAM stack and pieces like that. So all of the data in every region, customer data stays in the region, they put it in, but our billing data goes back to a central place and there's a global login so that you can log into all these different places. This is going to be fully self-contained and operated. It's going to be operated fully with EU citizens. We actually have an independent fully-owned subsidiary that's a European entity that actually has a European governing board that's subject to EU laws. And so it really is, we've worked together with many of the governments in Europe to put this together as a way that they can then get comfortable that they can put their critical government data or critical what they consider to be systemically important workloads in this cloud and be confident that they have the right sets of controls that whether disruption happens anywhere and it's the first of its kind and-
So built from the ground up?
Built from the ground up and no one has a region like this at all. The other cool thing is that it is a full AWS region. Has all the services, all the capabilities, and so when you go put a workload there, you're not making any sort of trade-offs on having to have a less capable cloud.
You've seen that movie before. I have to ask, in covering the CIA cloud and the GovCloud stuff, is there any experience that's learned from that, that applies here? Is it a completely different animal?
No, I mean we learned a lot from some of our isolated government clouds like we do for the intelligence agency, which is just how do you operate something like that. And I think that we learned a lot from building that type of infrastructure. It's different because this has internet access and it's public and people can come use it, and we expect many of our customers, even our global customers, our ISVs are interested in coming and deploying here.
It's some of the same concerns, 'Hey, my stuff's going to run on this.'
But the operational controls, how we operate it and things like that, we definitely learned from our decades of running some of the government clouds and GovCloud and other things like that, yeah.
So I know you got all kinds of meetings going on, you guys are spending a lot and get the checkbook open. I think last time we talked about nuclear, we riffed on that a little bit, but the CapEx spend that you guys are executing on, it's unprecedented, at least in my lifetime, I've never seen this kind of spend. Talk about the rationale there. Is it competitive forces? Is it demand? Is it something that you're thinking, I hate to say it, but build it, they will come because they want it? There's demand. What is the difference on the demand side versus being proactive and preemptive on the market, on the CapEx spend, on the data center, obviously new regions coming online, tons of more GPUs, XPUs, you guys are spending a lot on this.
Yeah, you're right. And it's demand-driven. We look to where the demand is today and part of our job is to think forward about where demand is going to be from customers a year out, two years out, three years out. Because a lot of times building data centers, acquiring power, things like that are pretty long lead times. And so that's a lot of where we focus our time today is thinking about how we balance that risk. But it's a very capital-intensive business. We knew that when we got into it and as you scale, the numbers get bigger and bigger, but as I said, we're also in the very early stages of adoption and so the potential is massive. Part of the promise we make our customers is that we'll scale for them and when we run out of capacity or we run short of capacity, it's painful for our customers. And so that is part of the promise that we make is being able to scale ahead of that. And as we know, much of the AI world has been short of capacity probably for the last two years and still is today, and we're still ramping as fast as possible.
So there's obviously demand.
There's lots of demand. That's right. And we're careful about how we invest long term and we're thoughtful about how we manage risk in that spend, but it's to support demand.
It's not an overbill, it's simply you guys have an intentional plan.
All right. Talk about the agent. This tees up the billion dollars that are available. I mean some say we even ... We're riffing on our CUBE podcast that the agentic will be a multiple bigger than SaaS. If you look at kind of agentic as a size, and that's a little bit over the top, but what's your view on agentic? It's SaaS-like in the sense that it's a whole nother wave of applications, but it's up and down the stack. Is that the infrastructure layer and all the way up to the business logic, do you see that as a multi-billion dollar opportunity for AWS? How do you scope that? How do you thinking about agentic as a core AWS opportunity?
Yeah, look, I think it's an enormous potential for many of our customers and I think if you think about all of the value that we've already seen from generative AI today, from generating of content to summarizations to things like that, I think that that is ... It's led to 10%, 20%, 30% efficiency gains and benefits for customers. I think the promise of agentic is when these models can go and actually accomplish tasks for you and long-running tasks where they can go do stuff, where they can use tools, they can use browsers, they can use capabilities out there, they can use text editors, et cetera. That's when you start to see not 20% gains but 200% gains. And so that's where I think that the potential is massive. It's a huge investment for us. I pulled a dedicated team together that's very focused on just going and building agentic capabilities for our customers so that they can build on us and take advantage of that. I think it's that important for us and for our business and for our customers. I think when I compare it to SaaS, I'm not sure if that's exactly how I would put it because I actually think that it's evolution. It's not like someone's going to take their application and throw it away. I think that they will still want to have those applications. Humans are still going to want to have a UI, we're still going to want to have a system of record, we're still going to want to have a bunch of these applications. But the ways in which you interact with them, the capabilities that they'll have will change. And so just how at my talk at re:Invent this last year, I said that inference is just the new building block and really think about it like that. In some ways, agentic workflows are just an extension of inference. They're just going and running models, taking data, but they can iterate more and do a bunch of these kind of reasoning loops and think for longer and do more capabilities, but at the end of the day, it's a building block inside of the thing that you're trying to go do. And so I think it's an extension of that. It's an adder on to that. I think it actually explodes the size of what the overall market can be, but it's not really a replacement.
It's still an application.
It's still an application at the end of the day, and so it doesn't really replace it. It's just augmenting and changing.
It actually makes it bigger actually. Expands the aperture.
The business logic is at the app layer you get the horizontal scale, you got the global footprint, all the actions and the domain experts and the domain expertise and the domain data.
That's kind of in the vertical. So we started to see healthcare booming. You're starting to see life sciences folks that are kind of, I call first generation democratized super-computing. They're getting the benefits. What are you seeing in these markets? Obviously financial service is always in there, early adopter side, but healthcare specifically, you wouldn't have guessed. Well, you guys have guessed. But they really weren't IT-heavy mainly for all their other reasons, but now they seem to be booming. Healthcare seems to be hot. Life sciences, oh my god, a miracle. I got super-computing. More competition.
This is actually one of the super interesting things, which I think if you would've talked to a healthcare life sciences customer six or seven years ago, they would've talked about this as kind of the bane of their existence. But a number of the regulatory agencies have forced a number of these customers, whether they're healthcare or life sciences customers, to organize and catalog and tag their data for a really long time. They have to keep long-term records, they have to know where it is, HIPAA requirements, et cetera. And because they had this repository of data that was actually really well labeled and organized, they were actually ready to move quickly and take advantage of AI-
Exactly. It was kind of a happy accident for some of these folks, but I agree with you. And by the way, it was probably an industry that, particularly on the healthcare side, not as much on the life sciences, but on the healthcare side, it was an industry that probably hadn't taken advantage of some of the earlier technology innovations where they've been a little bit slower to adopt.
It's just regulated industries almost are well-prepared because of the data hygiene they-
