AWS CEO on AI, Return-to-Office Policy and What’s Next in the Cloud | WSJ

Sure, yeah. Look, AWS has been going for, it's about 18 years now, and our vision is not that dissimilar from when we started, which is largely to help companies of all sizes and shapes, and governments now, to be able to go and focus on the things that they do best. And so our vision is we wanna provide compute infrastructure and storage and services, so that companies, whether they're startups or large enterprises, can innovate fast and do the things that they are most focused on. And so if you're a bank or a healthcare company or a two-person startup, running infrastructure and building storage services or database services or compute services is likely not the thing that makes you special and makes you different than your competition. Those are just the ways that you get there. And so our vision is to help those companies be able to do that. To help them go fast, help them be able to scale globally, and help them be secure and have great operational availability, and so that they can innovate and focus on what makes them unique and special. And so, that's our focus, and we wanna do that with a partnership mentality too, where we really partner with our customers, and we want our customers to love us and not have that old school kind of vendor relationship with their tech providers, where it was kind of antagonistic, but we help 'em grow together.

Yeah, I think, well, look, I firmly believe that AI, and particularly generative AI, is a transformational technology that is gonna change every single company, every single job, every single workflow out there. And so, when I think about where we are today and where I think the technology will be, then over the next couple years, I really think that everything, every one of those businesses is going to be disrupted. The way that workflows are gonna be done, the way that businesses will be done, the customer experiences that will be out there, I think they'll be fundamentally different. And so, I think this is a foundational technology that will change many of those businesses that are out there today, existing ones and new ones. And so, I don't, you know, there'll be ups and downs, there will be stories about how it's overblown, and now it's a best thing since sliced bread, and then once again, it's driving tons of value. But I think it is a fundamental technology that is gonna have the transformational impact on a wide variety of industries out there.

Sure. Yeah, I think, look, if you think about, there's a whole range. So you can take somebody, like a technology company like Adobe, who, on AWS, saw that this was gonna be, have a big impact on how people use their creative suite. And so, they trained a new model, and they launched it, and now they run Firefly, which is the way that allows you to use generative AI. That instead of the old way of kind of using Photoshop and going in and trying to update things, you can actually use AI to create images inside of their creative suite, adjust images and be much more creative as you're going and building images and videos inside of their suite. Then you look at somebody like GE Healthcare, who's using generative AI to completely change how workflows are done inside of a hospital. And so if you go into a hospital, and you talk to doctors and nurses, they would love to spend more time with patients and less time doing the administrative tasks that actually take up much of their time. Together With GE Healthcare, we've announced a partnership to work together on that problem to try to automate many of those workflows that do take up much of their time today. And then you can go look at some of our financial services, where Nasdaq is using AI models today in their market, and analytics, to look after fraud or to find if there's any sort of market manipulation things going on. And they've been able to respond to any sort of questionable events 30% faster, which has a huge impact on making sure to shut those things down quickly. So, and that's just where things are today. I think in the future, there'll be more and more bigger impacts.

I do agree that that is a general perception, and I totally agree... Yeah, so look, I think, I think the whole world was amazed when ChatGPT first came out. And it was a fantastic application. One, in that OpenAI did a great job and built a great model, but also, it kind of triggered a creative renaissance, where everyone realized what these models were capable of doing. And a lot of people in the tech industry had been working on some of these things and trying to kind of crack that code for the last couple of years. And we knew the generative AI models were powerful and had the potential to do a lot of amazing things. And I think this broadly opened it up to a wide swath of people who understood what these models could be capable of. And so what you saw was a bunch of the providers rush out and try to get something out to market quickly. And I'm sure many of you all, 'cause companies ran and did hundreds of proof of concepts, and they had chat bots on their website very quickly. And we took a bit of a different approach, mainly because of how we think about how we support our customers. And so we said, I do think this is gonna be a transformational technology, but the things that our customers are gonna care about, our enterprise customers are gonna care about, are, we think, fundamental to making sure that you get this right, which is, they're gonna wanna make sure that their data is secure, they wanna make sure that they, 'cause we firmly believe that your data is the thing that ultimately differentiates one company versus the other as you go and actually apply these models inside your businesses. And so companies wanna make sure that that was very secure, that it wasn't leaked back to models, and that they had control over how that data was accessed when they're using it together in an application. And then we also thought that companies... We wanted to enable people to go build interesting models and interesting technologies, really. And so if you're thinking about putting generative AI into an application, we kind of had this long-term view that you wanted access to all of the capabilities that were gonna be there. And we had this fundamental view, which was controversial probably two years ago, and is not so controversial today, that there wasn't gonna be one model that was gonna be 'the rule them all' type of a thing. But there's gonna be a lot of different models. There's gonna be small models and big models, and actually used in concert with each other to go build interesting applications. And the people were gonna need a rich set of tools in order to actually go build interesting experiences and go deliver interesting business results. And so we took a step, we basically said, 'We're gonna go build that platform so that our enterprise customers, our startup customers, our government customers can go build those applications and build what's gonna be unique to them, and not just get to market something quickly.' I think we're now seeing, that's now been out for, it's probably been about nine months now, and we're seeing our customers really love that platform, build real enterprise use cases on that platform. Not just proof of concepts, but things that are delivering actual value. And we're quite excited about the trajectory that we're seeing that business on.

Well, we spend a lot of time worrying about that, so I actually am able to sleep at night, mostly because we spend a lot of time thinking about that during the day. But we take that responsibility very seriously, and we have layers upon layers of protection to try to as best protect, our own infrastructure and our customers, from that. And so there's a couple of things to remember there. It's both, and by the way, there's kind of like two number one priorities for us. It really is security and operational excellence, and they kind of go hand in hand. And as you think about security, there's an enormous amount we invest from the very lowest layers of what we build on, from hypervisors to kernels to operating systems to our networks, to how we've architected all of AWS to ensure that we have the very best security. And we employ many of the best security mines in the world and spend billions of dollars every year to ensure that the AWS environment is protected. And we really start from, every time a software developer builds a new application, they start with security. They think about, 'Okay, how am I gonna build this in a secure way?' They have checkpoints to make sure it's secure. We actually have AI models that are constantly kind of sending up signals and say, 'Hey, you may wanna think about coding this in a different way.' And then we have a very rigorous process before anything gets close to production to make sure that we are adhering to all of our best practices. But the other important thing is how we architect our systems. And so, the example you're referring to, CrowdStrike was actually not a security issue, it was actually an operational issue, right, where a deployment went out. And so we also have systems that ensure that deployments can't hit everything at the same time, and they go out, and they roll out in a coordinated fashion so that they go to a little bit at a time, and only certain sets of our infrastructure at a time to try to isolate any potential issues that do get through all the various checks and balances that we have. So we spend an enormous amount of time thinking about, how do we really secure the environment, how do we really ensure that we're operationally excellent? And then we then go build tools along with our broader partner ecosystem to help customers secure their environments as well. And it's a hard problem, by the way, and it's getting harder all of the time.

Yeah, I think it's from, look, the security is an issue from all over the place. I mean, security can be, it can be nation state actors, it can be your own developers that accidentally lift your keys in a repository and published it to a public GitHub repository, and some ransomware people then took control of your account. Like, it can be a wide range of things that can impact customers. And so we try to help across that whole stack of things. And it's, and again, security is not one of these things that's easy to bolt on after the fact. And there's others out there that are scrambling to try to get that security piece in place, and it's hard. You have to really build it from the ground up, and that is key to ensuring you really have that strong environment.

Yep. I think it's 2030, hopefully not 2036. We're more optimistic... they can come a little sooner than that. But they're just part of the portfolio. And so I think there's a number of things that as we think about both energy now, energy in the next, call it three to five years, and then energy 10 years out, I think we're trying to think about all of those different timelines, because it's really impactful. As you think, particularly if you think about these generative AI models that folks are building, estimates are that kind of two to three models out, an individual model may require somewhere between one to five gigawatts of power, which is like a small to medium, maybe a large city. It's like a million households or something like that. And so, it's an enormous amount of power that these models could require in the future. And so, we're thinking about... In the near term, there are projects in place that, you know, can meet all of the needs in in the next couple of years. And many of those are renewable, by the way, whether they're wind, whether they're solar, et cetera. We've invested, Amazon over the last five years, every year for the last five years has been the biggest investor in renewable power projects all around the world. We've done more than 500 renewable power projects, adding funding brand new projects, bringing new power onto the grid to power, what we think, is an important renewable source of power. And we think that nuclear can be a big part of that. And so some of that, by the way, is bringing on, there's a bunch of nuclear around the world that's been shuttered, because people didn't need actually that source of power. And so the very first project we did was a partnership with Talen in Pennsylvania to bring almost a gigawatt, or eventually be almost a gigawatt, of power back online that just wasn't being used in the world today. And so, some of that is bringing some of these plans back online. Some of it is thinking about how you can refurbish, you know, called dirty energy plants or whatever, and bring them back online as clean energy plants. And some of them are investing long term. And some of these small modular reactors, you're right, they're not gonna solve anything in the 2020s, but in the 2030s, you know, in 2030 plus, they could be a excellent source of energy as we're looking to solve the energy problems.

What I'll say is we have successfully worked with every administration that's been in place so far. And will continue to do so. No, look, like that is the business from an AWS perspective in particular. We supported both conventions, we support elections all around the world. Not just in the US, but in Latin America and Asia and other places. And I think it's an important... And we are, the US government is an enormous customer of ours, and we do support them. You know, we support the intelligence agencies, we support the DoD, we support the fed and financial services. Like, there are many of the US agencies that run many of their critical workloads in AWS. And so we, it is a political-ish type of answer, but it's true. We support agencies throughout whichever the party is is currently running. And then we take that responsibility very seriously. I think it's an important mission that we support the country on. And so from our perspective, it's super important. I wanna continue to support the country, no matter if it's a Republican or Democrat in office, and that's our plan.