Okay, Leslie, thank you so much. Andy, thank you very much for spending this time with us. It's good to have you.
Yeah, thanks for having me.
I really want to start by getting your reaction to the news of the moment out here, and that is what's gone on with Jensen Huang and Nvidia, how you see it given his importance in this entire AI story.
Well, you know, I have deep respect for Jensen and I have deep respect for Nvidia, and they've been very significant partners for Amazon for a long time, and they will be for as long as I can foresee. And I know they're, you know, I don't know all the details of what's happened, but I know they're talking through it and see what happens.
You have such a unique relationship in that you're a customer and you're a competitor. Let's talk about on the customer side of things. Do you see that changing in any way?
No, I mean, we have this deep partnership. We've led with Nvidia in all of their launches. We have a lot of customers who are going to ask to run on top of Nvidia chips and our infrastructure for a very long time, as long as I can foresee. And to your point, we also have our own custom silicon in Trainium, which is doing really well. You know, the last time we talked about it, in the Q2 call, we shared that it was over a $25 billion annual revenue and run rate, our chips business between Trainium and our CPU chip, Graviton. And so I think both of those things in the technology space, you have this co-opetition where you're a customer of a company on one thing, you partner on other things, and there's some overlap competitively, and these market segments are so large, there's room for all of us to be successful. And that's what I think will be the case.
If one of your competitors, though, is wounded, so to speak, does that not help Amazon?
I just don't think of it that way. I mean, again, if you just think about our chips business with Trainium and Graviton, it is wildly successful now and growing triple digits year-over-year. And so I think it, you know, it is successful, and I think it will be successful regardless of what happens around us competitively. And you never wish ill on a competitor, particularly one that you're a partner of. And, you know, I think of ourselves more as a partner to Nvidia than as a competitor, even though there's some overlap. And so we, I believe Nvidia's business will be very strong for as long as I can foresee, and we'll be partners with them.
Does it help you sell Trainiums to customers who otherwise might have wanted Nvidia? Does it give you any kind of competitive advantage in that sense?
I don't see it that way. You know, I think that people will choose the technology and the services based on the capabilities of that technology, and regardless of what's happening around the company.
Is there going to be a time where you no longer need Nvidia chips yourself because you've developed Trainiums fast enough that you no longer rely on them at all?
I don't foresee a day where we're not partners with Nvidia. I think they have amazing products. I think we're going to always have customers who want to run Nvidia in our infrastructure. I think in the same way we're going to have a lot of, I mean, if you look at Bedrock, which is our big inference service, the vast majority of inference in Bedrock is on top of our own custom silicon and Trainium. So we're going to have customers who want our silicon, we're going to have customers that want Nvidia silicon, and there may be others too.
The market's trying to figure out, it feels like, where we are in the whole AI cycle, where we are from an equity trading standpoint, and just from where sentiment is. How does it feel to you right now? Where are we in terms of this buildout? Are we early innings still? Are we middle, late?
Incredibly early innings, in my opinion. Just, I mean, it's, you know, if you look at the history of AWS, we grew AWS what we thought was at a very fast rate in the first three years. After three years, we had a $57 million annual revenue run rate, and in the first three years of this generative AI wave, our AI revenue in AWS is over $25 billion. It's 440 times faster. It's, you know, so it's growing unbelievably quickly. And yet I would also say that if you look at how the demand, what it's resembled, it's been pretty barbelled to date. You've had one end of the barbell that has been, you know, the AI labs consuming gobs and gobs of compute and some runaway generative AI apps like ChatGPT and Claude Code. And the other end of the barbell have been enterprises having success in AI with cost avoidance and productivity, so automating customer service or fraud or business process automation. In the middle of that barbell are all the current enterprise production applications, more of which use inference than they did before, but still a small percentage of them using inference as pervasively as they will. And so there is so much growth in front of us. We are in the very early innings. I think it's, I understand why people are nervous because the numbers are so big, like the growth and the pace of it is so big. But if you have the demand signals and the visibility to it like we do, if we look at, you know, we'll always keep a fair bit of capacity available on demand for our customers, but the big chunks of compute are largely reserved for us in '27 and a good chunk of that in '28, and actually a fair bit in '29. If you have those demand signals, you realize that it is just, it is very early innings. And if you understand the cash cycle and how it works in a business, an infrastructure business like AWS, where you have to lay out that capital for data centers at least a couple years in advance before you can put servers in and monetize it, and then you invest in advance on the server and networking gear before you actually start monetizing it. When you're growing this quickly at this rate, you're deploying a lot of capex, and it has impact on your free cash flow. But when you get a few years out into monetizing and breaking even, and then driving positive free cash flow, the free cash flow and the return on invested capital in this investment is very compelling. And so we will be a much larger revenue business, a much larger free cash flow business in the coming years because of this investment.
I think you blew people away on earnings where you guided out AWS growth for two years. That stunned some people because you're just not used to doing such a thing. Is it the demand visibility that you think you have that led you to be confident enough to do that?
You know, I think we, from time to time over the years, we've guided a little bit further in advance than others. But I think in this case it's what I was saying. We do have what we consider very strong visibility into the demand, and we also are very aware that we're spending an amount of capex that's unusual. And so being able to share with people that we, you know, details around why we believe that investment is working and paying off and will pay off in a very significant way, I think is helpful for investors and shareholders.