WHEN: Today, Wednesday, February 26, 2025
WHERE: CNBC’s “The Exchange”
Following is the unofficial transcript of a CNBC interview with Amazon CEO Andy Jassy on CNBC’s “The Exchange” (M-F, 1PM-2PM ET) today, Wednesday, February 26. Following is a link to video on CNBC.com: https://www.cnbc.com/video/2025/02/26/amazon-ceo-andy-jassy-on-business-strength-and-new-ai-powered-alexa.html.
All references must be sourced to CNBC.
KELLY EVANS: It’s also another busy day on the AI front. This time, Amazon is launching its new generative AI product. It’s a voice product, and the twist is, it costs 20 bucks a month if you’re not a Prime member. The shares are down 10 percent from their record high just three weeks ago, and they’re on pace to snap a five-month win streak with their worst month in two years. Now, it’s the splashiest move we have seen from the tech giant in quite a while, so we will see if that turns the tide here. And joining us now to talk about that in a first-on-CNBC interview is Amazon CEO Andy Jassy, along with our very own Jon Fortt. Welcome to both of you, Jon.
JON FORTT: Alright, Kelly, thanks. Andy, thanks for coming all the way out to New York. You had a big announcement today. Alexa+. Want to talk about that. I also want to mark 2025 as 10 years since you and I first sat down at re:Invent 2015. So—
ANDY JASSY: It’s amazing.
FORTT: Yes.
JASSY: It feels like it’s been 20.
FORTT: We will see how you feel after the interview.
JASSY: Yes.
FORTT: No, so tell me about Alexa+. There’s this shift in what Alexa can do. And as we have got this era of new models in AI, you’re in this agnostic, use the best model for the job. How’s Alexa+ going to work better and differently for people?
JASSY: Well, we’re really excited to announce Alexa+ today. It’s our next generation of personal assistant. She’s more — she’s smarter. She’s more capable. She’s more useful. And you can do all the things you have been doing with Alexa for 10 years, but just every single one of those functions gets better with generative AI, the way we built it in. And I will kind of give you some ways I have been using it myself playing around for the last few weeks to give you an idea of why I like it so much. I’m a big music fan. You and I have talked about this before. So the other day, I was trying to remember. I went to the opening night of the Guns N’ Roses “Use Your Illusion” tour in 1991. In Alpine Valley, Wisconsin, with my college roommate. And I was trying to remember who the opening act was for it. And so I asked Alexa, and she reminded me it was Skid Row. She said, “Do you want to play Skid Row?” I said, “Maybe not.” I did want to play a Guns N’ Roses song, but I wanted to play the song that I couldn’t remember the name of. I said, it’s the song that begins — it has the word apples in it. She’s like: “Bad Apples. Do you want to play it? I said yes. Like, that’s — think about that versus what Alexa has been. She wouldn’t have known the opening act. She wouldn’t have known when it was. She wouldn’t have been able to get “Bad Apples” from Guns N’ Roses. So, much smarter, takes action. Or if you use Alexa for Smart Home, you can now say, hey, we have guests coming to the house at 7:00 tonight. Please open the drapes. Turn the lights on the porch and the driveway, raise the temperature five degrees, and play mellow music in the dining room, like all just verbally, no app, no complexity. It just all happens, again, taking action. Or, in New York, I can ask, where are good restaurants around where we are right now? She will give me ideas. She will ask if I want to make a reservation. She can make a reservation for me. So, it’s not just that she’s really smart and can be conversational, but she can also do something that you just don’t see today in these chatbots, which is take real actions and get real things done.
FORTT: Multistep awareness, context, that kind of thing.
JASSY: Yes.
FORTT: Now, tie that in for me to the business overall, because we know about AWS, huge data center presence, huge scale. You guys are the biggest. You have also got Trainium, Inferentia. You got chips that are tuned specifically to do the tasks that you want done in the data center. Combine that with what you’re doing now with Bedrock and the choice of models on top of that. How does that give you a business advantage and maybe even longer term a profitability advantage over the other players in the space?
JASSY: Well, I think one of the interesting things about what we’re doing with AI and at Amazon is, we have this very unusual AI flywheel, where if your mission is to make customers’ lives better and easier every day, which is ours, and if you believe every customer experience is going to be reinvented by generative AI, which we believe, then you’re going to be building a lot of generative AI apps. And if you’re building a lot of apps, it means you’re getting a lot of feedback from builders on what they need in the building blocks for generative AI And then, if you invest in those building blocks, like we have, our own chips with Trainium2, which are 30 to 40 percent more price performative than what’s out there, our own frontier models in Amazon Nova, our own model building services in SageMaker AI and generative AI, application building services in Amazon Bedrock. If you’re investing in those and get a lot of feedback, it can’t help but making those investments and those services much better more quickly. And as they get better more quickly, it can’t help but allow builders to more easily and quickly build generative AI apps, which means they can’t help building more apps on top of the platform. So there is that very unusual flywheel we have, and it impacts really all the areas of our business in a way that’s unusual.
FORTT: We saw it in smartphones really. We saw it with Apple and iOS and the App Store built on top of it, and then, if you get the right customers and you have the right tools, and it’s easier to build, et cetera, et cetera. But there’s a difference between iOS and Android in profitability and in willingness to pay, right, for the customer. How do you make the customer experience, the businesses who are building their applications on top of AWS, tapping into those models that you make available on Bedrock so much better than what they can get on other platforms that AWS excels?
JASSY: Well, remember, what AWS ultimately is, it’s a set of technology infrastructure services that others build applications on top of. And the more that we make companies successful building better customer experiences and improving and growing their business, the more AWS succeeds. And so really what all those I mean, if you’re building generative AI applications, which every company virtually in the world is doing, and you start to get to scale, you realize really quickly that inference or the predictions at a scale generative AI application get expensive. So you, the key to that inference is the chip that’s in the compute. And so, if you can save 30 to 40 percent in price performance on the chip and on the compute, like you do with Trainium2 over what you can do today with GPUs, that’s a big deal. And that means more people will actually start using Trainium and Trainium2 to run their applications. And the more people that run Trainium 2, remember, Trainium2 or chips inside of our EC2, which is our compute service. Inside of that of those instances, the more people run those, the more that they drive the AWS business, and AWS has a reasonable profitability profile. So the more we get people using Trainium2, using, or really any chips on top of our compute infrastructure, using our services to build models that they then deploy to applications that run on AWS, the more those applications run on AWS, the more it grows the AWS business with a reasonable profile margin.
FORTT: I do want to talk some more about that and AI, but, first, I want to touch on some other things, and actually first something that AI doesn’t yet do so well, which is craft high-end entertainment. Recently, fair to say the Hollywood world has been rocked by the news that Amazon MGM is taking full control over the James Bond enterprise. It had been, let’s see, Michael G. Wilson, Barbara Broccoli who sort of post their dad were guiding that franchise. And people are concerned that Amazon is going to over-commercialize or something. Tell me about what Amazon will and won’t do. How much attention are you paying to those notes that Ian Fleming took about what’s Bondian and what’s not?
JASSY: Well, I mean, first of all, I don’t believe that AI is going to write a Bond movie any time soon. So I think that we are, we’re incredibly honored to have the ability to help shepherd the Bond franchise for the next generation. And we have amazing filmmakers that we have hired over time in MGM, Amazon Studios. And we haven’t, we don’t have a plan yet on what the next theme is going to be. We haven’t, nobody’s written the story yet. But we will take great care. It’s an amazing franchise that we care a lot about and that we expect to develop over a long period of time.
FORTT: Now let’s talk about retail and the consumer a bit. Give me your sense now that we’re a couple months into 2025. You were talking about consumers really looking for bargains and even trade-downs, I think, late last year. How is it looking into 2025 as, for some consumers, credit continues to be stretched?
JASSY: I think we’re seeing pretty similar things, Jon. I think that consumers are still spending. They’re being careful on how much they spend on each product. Where they can trade down a price, they do so. Where they can find deals, they continue to do so. We spend a lot of time trying to help with our third-party selling partners. Customers have lower prices. We were ranked the lowest prices in U.S. retail for the eighth year in a row by Profitero right before the holiday season. And so I think we see a lot of those same things. I think there are certain countries where maybe the economics or the macroeconomic factors are just more challenged, Germany being an example of that. But we, I think what happens in times where consumers are a little bit uneasy about different economic conditions is that they, customers tend to figure out who are they going to buy from. They tend to pick a smaller number of merchants they buy from. And because we have such broad selection and we have such low prices and we get items to people so quickly, and we take care of customers, first and foremost, I think customers are continuing to choose Amazon for more of their purchasing.
FORTT: I want to zero in on that, getting things to people quickly, that logistics piece. The stat I have seen is that Amazon Logistics processed $5.9 billion U.S. delivery orders in 2023, 27 percent of all packages shipped in the U.S. It was a close second to the U.S. Postal Service then, which had 31 percent of parcel volume. Do you think you’re going to pass USPS in 2025, if you didn’t already in ’24?
JASSY: Well, you know, it’s funny. We don’t spend any time thinking about that or talking about that, Jon. We have, we are unusual probably relative to most retailers, where we have built our own last mile transportation network that’s at least the size of UPS at this point. And so most of our shipments go through our own last mile transportation network. But we, the USPS is still an important partner of ours. And we do work with them, and for various geographies. And I don’t, I mean, it’s hard for me to tell what their growth will be like, but we expect to continue to grow our own network.
FORTT: Because they have been shrinking. At least they were heading into that last stat. How important is scale in that piece, the actual last mile delivery piece of the business? Last time you and I talked, we were talking about those local sortation centers. And when you scale up with those, it actually gets easier to deliver faster. How is that working out with the scale that you’re achieving?
JASSY: Well, scale matters in almost everything you do. When you have scale, it helps with all of your various cost structures and all of your inventions. And there’s so much technology innovation in our fulfillment network gets amortized over a larger number of nodes. Sometimes, when you scale, it can create complexity. And so if you remember, for the pandemic, we took a fulfillment network that we’d grown over 25 years and doubled the size of it in 18 months and then built out a last mile transportation network the size of UPS in 20 months. And when we were done with that, we had a different network than we had before and it took us time to really optimize that network and get the cost to serve right. But what it does is, as you reach different elbows of the curve and scale — this is true in a fulfillment network. It’s true in technology services. As you get to different elbows of the curve and scale, you have to go back into your architecture and figure out where are the points that may not scale the way you want. And for us in our fulfillment network, we completely rearchitected where we had our big hubs. We built a regional network. We completely redid the architecture around our inbound network. And so now we’re able to get items located in fulfillment buildings much closer to end users, which means shorter delivery distances, which means we get to customers more quickly at lower cost.
FORTT: Now, also I want to talk safety when it comes to that. In about a week-and-a-half, it’ll be a year since the last Amazon safety report that talked about the progress that you had made on injuries in the warehouses. Now, that progress as of a year ago still had you, I think, with more than twice the injury rate of Walmart. I don’t know if you have an update on that for the more current period or how you’re hoping that’s going.
JASSY: Yes, well, the first thing I would tell you is, just in terms of trying to compare injury rates to different companies, if you actually look at the detail, which I have spent quite a bit of time along with the team doing, everybody measures it differently. So it’s very hard to do an apples-to-apples comparison. But if you look at the safety data over the last few years or all of our key safety metrics have improved about 25 to 30 percent, so very significant. We will share our 2024, first look at our 2024 safety metrics, they got better again year over year. And so, for us, safety is priority zero. It’s the very first thing that we care about. And if there’s something amiss about it, we will stop what we’re doing and rearchitect. And but I’m really proud of the progress we’re making.
FORTT: Well, speaking of Walmart, Amazon quarterly revenue past Walmart for the first time recently, and I know it was a Q4, so it’s a little bit different. You were talking about scale before. Do those benchmarks matter at all? I think they probably matter to investors to kind of trigger a sense of how big Amazon is getting overall in terms of the amount of money that customers are trusting you with.
JASSY: Well, I have a lot of respect for Walmart, always have. We spend zero time talking about those benchmarks. I have never been in a meeting where we have talked about where we are size-wise or that we might have a higher amount of revenue than Walmart. We just don’t. Last time I checked, it doesn’t improve the customer experience any. So it’s just not a factor for us.
FORTT: Something else I’m tracking and watching, your advertising and subscription businesses together, I think as of the last fiscal year, passed $100 billion a year in revenue. Advertising now has a $12 billion dollar lead on subscription. How much of that signifies how the business is shifting, broadening, including media, including so many third-party sellers trying to get seen?
JASSY: Yes. Well, the first thing I would say is that we don’t think of advertising as a business, per se. I mean, there would be — if there was no retail marketplace, there wouldn’t be any retail advertising. So advertising is something that’s a part of a number of our businesses. It’s part of our stores business, our retail business. It’s part of our Prime Video business. It’s a part of our grocery business. It’s part of our audio businesses. And then we try we spend most of our resource on the advertising side. It’s — most of the team are really — they’re scientists, particularly AI scientists who are trying to come up and refine products and algorithms that show customers relevant products, so that customers are exposed to a number of new products that they otherwise might not have known about or gotten exposure to. And that’s what we spend a lot of our time on. I think it’s improved every single one of those experiences that way in allowing people to learn about products they otherwise would not have. And we’re constantly working at it.
FORTT: How has that changed, though? Because, from the earliest days I can recall using Amazon, it’s, well, people who bought this might also like—
JASSY: Yes.
FORTT: There’s always been a recommendation engine that it seemed like AI at the time, and it was. We just have sort of talked about AI in different terms now. How is the capability different? How is the work force that you have able to make better use of that now?
JASSY: Well, it’s a really what we’re doing, everything that we’re building and trying to expose is for customers. And so we still have what you were defining, we call similarities, which is, people who have bought these products most often by these products. And that’s still a very important feature that customers use. It’s very prominently displayed. It tends to be products that have broader history over time. That’s how they get enough purchases to make those algorithms. But there are so many new products all the time from sellers inside country or outside of country that just otherwise don’t have a way to get above the fold and get exposure and awareness for customers that a lot of customers value learning about. And so that’s what we’re trying to do with our advertising algorithms in our retail marketplace, which is give exposure in a relevant way based on what you search for some of those products that don’t have the history yet to naturally get into similarities.
FORTT: How are you model, speaking of exposure, how are you modeling Amazon’s exposure to this policy rift between the U.S. and China right now? So many third-party sellers are either in China or relying on products coming in from China. Tariffs going up is going to affect them. How do you model that out?
JASSY: Well, it’s hard to predict. And it’s hard to know what’s going to actually where we’re going to net out on tariffs between China and the U.S. or any other countries that the world has been talking about. But we have a very robust, substantial number of sellers, China sellers, who sell in our marketplace. And they have done very, very well. And they work really hard to have great product, to deliver on time, to make people aware of their products. If there end up being tariffs, my suspicion is, they will still sell in our marketplace. Some of the prices may be higher. But that will be true for everybody who has sellers or merchants or partners in China.
FORTT: Okay. You and I have a tradition going back now probably at least seven or eight years where I ask you whether you’re going to spin out AWS. And you always give me the same answer. But I want to ask it with this framing now. Given that we’re so deep in the AI conversation, and we have just been talking about advertising and data and how closely that is tied in to the retail conversation, how does that factor in to the answer you’re going to give me when I ask you, are you going to spin out AWS?
JASSY: Well, Jon, if you have been listening all these years that has always factored into the answer. And so, oftentimes, I start with explain that usually companies spin out business units that they don’t want them on their financial statements or if they can’t afford to fund the new business. Neither of those are true for us. In the AWS business, we have so many customers and partners who want to have a partnership with us, not only in AWS, but sometimes in our retail business, or sometimes in our entertainment business, or sometimes in our devices business, or sometimes in our advertising capabilities, where the companies with whom AWS competes have the same sorts of strategic relationships. So it’s very useful to be part of the same company. But on your data piece, I think AI presents kind of another good example of how it’s useful for us to be part of the same company. It comes back to this, what I think is a very unusual and compelling AI flywheel that we have at Amazon, which is, if your mission is to make customers’ lives better and easier every day, which is our mission, and if you believe that virtually every customer experience is going to be reinvented by generative AI, which we believe, then you’re going to be building a lot of generative AI apps. And if you’re building a lot of generative AI apps, it means you’re going to be getting a lot of feedback on the components that can make those generative AI apps better and easier to build. And if you’re willing to invest in those components, which we are, as you know, we talked about chips with Amazon Nova and our own frontier models — I’m sorry — chips with Amazon, with Trainium and Trainium2, our own models, frontier models with Amazon Nova, model building services like SageMaker AI or generative AI, app building services like Bedrock. And if you’re willing to invest in those services like we are, and you get all the — that feedback, you can’t help but improve those building blocks for generative AI much more quickly. And as we improve those building blocks more quickly, you can’t help but make it easier for first-party and third-party application builders to build generative AI apps. And when you make it easier and quicker for them to build, you end up with more of those generative AI apps on top of your platform. And that flywheel is very unusual, given the nature of both of what we do in AWS and then the breadth of businesses that Amazon pursues. And having those connected, as you can see, drive each other in a very powerful way for customers and for our business.
FORTT: So, no, you’re not spinning it out?
JASSY: No plan.
FORTT: I don’t know if you have heard, though, sometimes companies spin out gems. Comcast is spinning out CNBC. That’s, anyway. DeepSeek let’s talk about that. It was up in your presentation this morning among the menu of options on top of Bedrock that customers have to choose from. There’s been a lot of talk about Jevons paradox and this idea that, when something gets cheaper in the technology world, more efficient, it’s going to drive up the usage of it. Have you seen that even since the emergence of DeepSeek, even since your collection of Nova models? What have you seen happen with usage?
JASSY: Well, I would say that, first of all, I’m a huge believer that the less expensive you’re able to make technology, people really appreciate having a lower cost per unit of that infrastructure, but they don’t tend to spend less. It tends to open up the floodgates to all the other things they want to do, and they end up, in absolute, spending more. So the more cost-effective that we can make inference, whether it’s through our own models in Amazon Nova or models like DeepSeek or any of the models that we provide, the easier and more cost-effective it’s going to be for customers to be successful with generative AI and the more that they’re going to build and run on top of AWS, and that will be good for customers and that will be good for our business. On DeepSeek, it’s early. We got the DeepSeek models into Bedrock within a week. We have real usage there. It remains to be seen how many Western companies will choose to run applications with sensitive data on top of an open-source Chinese model company. I don’t know. If they are, if people are comfortable with that, great. We will make it really easy and very performant on Bedrock. And if they don’t, they will use other models.
FORTT: But is it actually opening up the conversation about Nova? Because you have got an efficiency story and different sizes and different levels of capability of Nova based on how heavy-duty the task is.
JASSY: Yes, I would say DeepSeek doesn’t open up that conversation. But the fact Nova has a lot of buzz and a lot of early usage in Bedrock because it not only has comparable intelligence to the leading frontier models in the world, but it’s meaningfully lower latency and cost-effective. In Bedrock, it’s about 75 percent less than the other Bedrock models. And so the dirty little secret when you actually have generative AI applications at scale, which we have several is the inference gets expensive. And so, if you can find ways to make the cost meaningfully less, that’s very compelling. And, by the way, you — it’s great to be intelligent and to be able to ask an application intelligent things. If it takes 20 seconds, takes 10 seconds to come back with an answer on something that typically takes less than a second or less than two seconds, it doesn’t work. So you need the right ballpark of intelligence, but then low latency and low cost are huge, huge considerations. And Nova is leading on those.
FORTT: I know you said this morning that Nvidia continues to be an important partner, and you expect that for the foreseeable future. How much has your demand for AI chips, for GPUs, for accelerators shifted as you have rolled out these more efficient models in Nova?
JASSY: We have a lot of demand for AI right now and a lot of demand for our instances that have Trainium chips, have Nvidia chips, AMD chips. And I would tell you that, at this stage — and it could change, but, at this stage, if we had more capacity than we already have — and we have a lot — but if we had more capacity, we could monetize it.
FORTT: For Nvidia, you’re saying?
JASSY: For everything.
FORTT: For everything?
JASSY: Yes. I mean, I think, remember, there’s really predominantly been one chip that people have used the first couple years of this revolution around generative AI I think, over time, people are going to use lots of different chips. And how that mix changes, we will have to see. But I have high confidence that Trainium will be a chip that a lot of people use on top of our platform, in addition to Nvidia.
FORTT: Is this potentially like the Logistics network, where it used to be, oh, everybody uses the Postal Service or UPS and FedEx, and now, look, here’s Amazon, right, about to, it looks like, be the biggest out there?
JASSY: I think the thing that’s similar about it, Jon, is that, if you think about the engine inside of Amazon, everything we do, if you were there in the heartbeat, in all the meetings, what’s driving all the conversations is, what do we think will help customers? How can we make customers’ lives better every day? And in this case of people that use AI, they desperately want better price performance, and same thing, by the way, when in our retail business. We spent all this time figuring out, how can we get items to people more quickly and more cost effectively? People don’t want to pay for shipping and they want it fast. And so, over time, and the pandemic accelerated this because of how demand changed in the early part of the pandemic — we just realized that our third-party partners, while they’re still partners and will be, they just didn’t want to expand to the degree that we needed, at the cost structure we needed to serve customers the way I just mentioned. And I think the same thing you’re seeing in chips, which is that we will have customers who use Nvidia chips as part of EC2 instances forever, but, increasingly, as more companies are trying to do generative AI at scale, they want better price performance to be able to afford it. And that’s what’s driving our investment in Trainium2 and in the Trainium series, is that we want to provide leading performance, but at lower prices. And, for us, we have a little bit different business model than maybe some other people. For us, the more people are successful in using these chips and compute, we’re able to drive — better customer experiences, growth in their business, but also we have a different business model because we have a compute set of services that also benefit.
FORTT: I want to zoom out again, get back to the consumer because it seemed like, especially right up until the election in the U.S., there was a lot of reticence around spending. But then, at the same time, post-inauguration, there were some consumers who were ticked off about CEOs they saw showing up at the inauguration, some people also very happy about it. Have you seen any shift in Amazon consumer patterns kind of post-election, post-inauguration, or have things pretty much evened out and normalized?
JASSY: No, we haven’t seen a shift. Look, what customers care most about, and we have learned this over a long period of time, what consumers care most about in a retail business is really broad selection, low prices, and getting items to them quickly, and a company they can trust that’s going to take care of them. And I think we’re unusual along those dimensions, and I think it’s why people continue to choose us. We have been a business now across six administrations. In every single one of those administrations, our primary focus has been trying to take care of customers. We always try to have a productive relationship with the administration here in the U.S. Some have been more receptive to that than others, but we will always try to have a productive relationship, because we’re headquartered here in the U.S., and we want to help the country.
FORTT: And there are always frustrations between business and administrations. How’s this going so far, this relationship with the second Trump administration?
JASSY: Well, I mean, it’s early, obviously. It’s been a month or so. A lot has happened in a month, but it’s been a month. And I would say, so far, I’m optimistic about it. I think that this administration seems to want to partner with businesses, seems to want to hear what’s working for businesses and what’s not working for businesses. I have never understood how governments in any country believe they’re going to get to the best answer for their economy making decisions without closely partnering and hearing what would work and what wouldn’t work with the private sector. And so I — to me, the most productive countries, it’s always a partnership between the public and the private sector. And this administration seems to want to partner with businesses and hear what they have to say.
FORTT: All right, Andy Jassy, CEO of Amazon, with me here in New York, I appreciate the time.
JASSY: Thanks for having me.
For more information contact:
Jennifer Dauble
CNBC
t: 201.735.4721
m: 201.615.2787
Stephanie Hirlemann
CNBC
m: 201.397.2838
