Every CEO wants to move faster, but what does it really take to move at the speed of agentic AI? AWS CEO Matt Garman knows. He has catapulted the world's largest cloud company onto the leading edge of generative and agentic AI. While most people were still getting used to the term agents, Matt had already named agentic AI as AWS's next multi-billion-dollar bet. In this episode of BCG's CEO Moments of Truth, Matt and I talk about the conviction it takes to make a call that bold. We also examined the creativity and the discipline required to turn that bet into a booming new business.
Matt, you took over as CEO of AWS in June 2024. That was the point in time when the general narrative in the market was that AWS had missed the boat when it comes to AI relative to competitors. How did you look at that situation?
Yeah, it was, it's a super interesting story. It's as you remember, a bunch of the technology providers kind of rushed in to offer AI solutions, right? And they had a whole bunch of kind of really early solutions where they could try to get people up and running with chatbots and other things. And there was everybody was running these proof of concepts everywhere. And for us, our strategy was a little bit to take a half a step back and say, what is it that customers actually are going to want? And from our view, customers were going to want a lot of choice. And we made this bet that customers are going to want to use lots of different models. And we made this other bet that customers were really going to care about their data and their IP. And so we were actually a little bit slower to go build that platform. We were building it fast under the covers, but we didn't rush a solution to market. And so this narrative kind of bubbled up to the surface that we're behind. Fast-forward to today, I think everybody would agree that we're in a much better place. If you see workloads starting to move to production, a lot of those land on AWS. When I ask them, why would you launch on AWS or Bedrock as opposed to anywhere else? The thing that they mostly come back to is, I trust that my data is secure inside of AWS. And that wasn't a thing that people were worried about three years ago. It's one of these things where we kind of had to choose to be misunderstood for a little bit of time to build this right platform. And this is one of the reasons we called it Bedrock, because we knew this was going to be a transformational technology for a lot of different industries and a lot of different places.
You mentioned the word agentic AI, and I remember in summer 2024, people started to talk about it, but no one really understood what it is. I think you put a stake in the ground as a new CEO that this is going to be AWS's next multi-billion-dollar business. That's quite gutsy at that moment, but what gave you that conviction and how did your team react to you putting that stake in the ground?
When I looked at where the landscape was going to go and where I thought AI was really going to transform businesses, if you don't really get to an agent world where work is actually being done by the AI, then you're still in a content creation world. I felt very convicted that if this business was going to be big, it was going to be in the agent world. And that meant we had to put a stake in the ground. We also had to build investments and we had to build building blocks, we had to build services, we had to build capabilities that help customers have a framework so that they could go toward that world. That's where I made the bet. I think in retrospect, it seems like an easy bet, but where it's now we're at, I think $25 billion in our AI business, just kind of three years in, which is that's much more growth than I would've anticipated back then. And we like to think big about what's possible for the business, what's possible for our customers, what's possible for all of these services. And if you really think big, it's a multi-trillion-dollar opportunity to really go after. And so if you see it as a multi-trillion-dollar opportunity, it doesn't seem crazy to dedicate resources, dedicate focus and really go after what is going to be today is already a tens of billions of dollars opportunity and is frankly just getting started.
Matt, so you had articulated the ambition, a multi-billion-dollar new business for AWS. You set the direction, but now you needed to make changes. How did you think about those changes and what exactly did you do?
The very first thing that I did was if we're going to put our money where our mouth is, we've actually got to go dedicate a team. And one of the things that we believe deeply in at AWS and at Amazon is having single threaded teams where your top people are absolutely focused on solving the biggest problems you have. And so I took one of my biggest leaders in the organization, Swami, and I took away the other things that he was working on and told him, "I want you to be single threaded focused on going and helping us solve agents for customers." And so he went from having a team of tens of thousands or whatever it was to a couple hundred people and very focused on how do we go win in the agent world? And then we also took Swami and the team and said, "We want to have a robust set of underlying capabilities that we can use and our customers can use to build agents." What are the other kind of purpose-built agents that we could go make for customers to both that they could easily adopt, they could just grab these things and run and get use of right away, but also to inspire them to see what's possible and what's able to build?
Amazon is known for many things, but also for very rigorous resource allocation. You allocated resources very strongly into this new agentic road. Yeah. How did you feel about the whole resource allocation at the time?
It used to be where if you needed to go build a service, sometimes we would have 200 or 300 people that would go after kind of building a service. But with AI, we've actually been able to accomplish so much that particularly in the early start phases of building a service, when I asked the team to go rebuild that inference engine under Bedrock, we had like six people that were doing that.
And as you set that direction, how did you think about organic internal investments versus important investments externally, partnerships, etc.?
Well, I will say typically we like to build internally. It allows us to own the operational excellence. It allows us to own the security. And in this space in particular, we knew that a super important partner in this world was going to be the frontier model providers. And so we were very explicit about leaning in and finding, and we were fortunate enough to work early on with the Anthropic team, fantastic incredible group of scientists, incredible group of inventors. And this year we, again, did a similar partnership with OpenAI and really leaned in with that team there to understand how they could supplement. Our customers were asking for OpenAI models in addition to Anthropic models so we could have kind of both of the leading frontier models inside of Bedrock.
