WHEN: Today, Thursday, April 11, 2024
WHERE: CNBC’s “Squawk Box”
Following is the unofficial transcript of a CNBC exclusive interview with Amazon CEO Andy Jassy on CNBC’s “Squawk Box” (M-F, 6AM-9AM ET) today, Thursday, April 11. Following are links to video on CNBC.com:
All references must be sourced to CNBC.
ANDREW ROSS SORKIN: We are here in Seattle this morning at Amazon’s headquarters. And earlier this morning, company’s CEO Andy Jassy released his annual letter to shareholders, touching on everything from product-pricing, to cost-cutting, to delivery speeds, artificial intelligence, and so much more. And Andy Jassy joins me now in an exclusive interview. And we’re thrilled to have you here.
ANDY JASSY: Thanks for having me.
SORKIN: Thanks for waking up early to do this. You know, we were just talking about this inflation data. And before we even get into Amazon, I’m just so curious, given that you have your tentacles in so many parts of the economy, from the consumer, to the business-to-business side of things — how you see things and how you see things as it relates specifically to inflation right now?
JASSY: Well, what we see are — consumers are spending but they’re just being really careful about what they spend on and how much they spend. And so, wherever they can, they’re trading down an average selling price. You know, you see — you see consumers wherever they could find a deal, they take the deal. We see — it’s going to take a lot for people not to buy detergent or shampoo or things like that. You can see that in the growth of our everyday essentials business which last, you know, Q4 of 2023 was over 20 percent year-over-year. Some of that is because people are going to buy all time, but some of that it’s just because of the speed that we have and we’re getting to people, so they consider us for more. But people are still, they — they are still buying, they’re just being careful about what they shop for. And they’re not —
SORKIN: But you’re seeing a trade down?
JASSY: We are. We see people trading down on ASP wherever they can. You know and then I’d say on the enterprise side, you know, the last couple years, companies have tried to save money however they could and we saw it a lot in our AWS business and the cost optimization we saw. But I think a lot of that cost optimization has attenuated and we’ve seen a lot of the deals that people were talking about with us for a long time, you know, that were sitting on the back burner have started to move. Migrations have picked up again. The pipeline is really full. And I think people have moved from largely trying to save cost to figuring out how to modernize their infrastructure again, move from on premises to the cloud and then think about how they can use generative AI to change their business.
SORKIN: Are you seeing though on the cost side — on the inflation side, things continue to rise? Are there places where you see price declines? Are you seeing anything on terms of wages either on your own employees or others?
JASSY: Again, we see consumers are spending they’re just trading down. You know, we — a place where we see it real impact is — you know, in discretionary items things like TVs or computers or electronics. You know, we’re growing our market segment share at a faster rate there than others, but still at a lower rate than what we see in a healthy economy.
SORKIN: I want to start — I mean, there’s a hundred places to go here because you touch on just about everything in this letter. But I want to start with AI because it seems to be the buzzword of everything. When we were here last year, we were talking about AI but it was unclear I think completely where — where Amazon was going to play in this AI world. And you now say it is going to change everything and that you have really almost shifted the whole company towards AI in almost every space now.
JASSY: Yeah, we — we’ve been saying it for a while and I do think that AI is going to transform every customer experience that we know. And you know, I think a lot of the discussion has been around applications. And you know, starting really with ChatGPT which really caught people’s attention, but we think there are three big macro areas of the generative AI stack, each of which are gigantic and each of which we’re investing deeply in. And you know, I think of it as at the lowest layer are for people that are building their own large language models, and the two things that really matter there is the compute, to train the model, and to run the predictions of the inferences. And what matters in that is the chip.
SORKIN: Right.
JASSY: And then the services that you have to actually build models. And we’ve — we’ve got very significant investments there. We — you know, we — most of the early AI models have been built by Nvidia, but supply is more scarce and people are concerned about cost. And so, we’ve built our own custom silicon and AI. We’ve built a training chip called Trainium and an inference chip called Inferentia that are meaningfully more price performant than what you can find out. There’s a lot of the training, a lot of the predictions are going to be done on those chips. And, you know, and then I would say at that middle layer of the stack are for people that are going to actually leverage somebody else’s model but customize it with their own data and then use our features to make it easy as a service and we built the service called Bedrock. And a lot of the generative AI applications are being built on top of Bedrock. It’s the easiest way to build a high-quality generative AI application. Then there’s the applications of which, you know, we’re building a lot of them ourselves, but the vast majority of those will be built by third parties and we’re optimistic a lot of them will be built on our enterprise.
SORKIN: Let me ask you about this. Let’s talk about large language models because there’s just so much focus on it. You have now made investments in Anthropic, which Google’s also made an investment in. You’re building your own large language model called Olympus. What — how are you thinking about investing in things like an Anthropic versus trying to build your own?
JASSY: I think the one thing that we’ve seen pretty consistently in this relatively early stage of generative AI is that customers want choice. You know, they’re learning how to build models. They want different model types for different types of applications and use cases. They want different model sizes because it changes the latency and the cost structure. And so, you know, part of the attraction for so many people of Bedrock I was just mentioning is just — it has the largest selection of models available. It’s got not only models that we build but also Anthropic, and Meta with Llama 2, and Mistral, and Cohere, and Stability AI. So you have a lot of choices and then Bedrock makes it really simple to move between models and move between model sizes, so that you can actually experiment and get the right quality and latency. We’re going to — for us and we see it with our own — we’re building dozens of generative AI applications right now. Some of which are launched, others are in development. We find it with our own model builders and our own application builders that sometimes they want to use our own app — our own models, sometimes they want to use third party models, and we’re going give people choice so they can make those decisions.
SORKIN: Right. How important though is it for you to catch up on a — on having your own generative AI large language model? Meaning this Olympus project which has been reported on, how important is that? How far behind you think you are? What do you have to do to get that to the same level as where Anthropic may very well be or where you believe OpenAI is today?
JASSY: Well, I mean, first of all, what I would tell you is that — this is — this is going to transform virtually every experience we know and it’s a gigantic space, and they going to be a lot of very successful players in it. And if you go talk to enterprises or companies that are using generative AI, they are very excited about what we’re building and more and more customers are moving to our services. And, you know, we’re going to have partnerships — we have a very deep partnership with Anthropic. Claude 3 which they released just a few weeks ago is the best model on the planet right now, with the best performance, and it runs, you know, best on top of AWS. And so, our customers are very excited about that. But at the same time, we — you know, we have a lot of experience building models at the company. We’ve been doing it for a long time, and we’re — you know, we’ve been building our own models and these titan models we have. We’re going to keep building those and we’re going to give people choice.
SORKIN: Big issue with building these models is the data input that’s being used to train. Do you know for example with your new large language model what data you’re using to train it on? There were reports that OpenAI was in search — desperate for data, they started — they actually built a program just so they could create transcripts of YouTube videos so they could train off the transcripts because they didn’t think they had enough data. And that, of course, raised questions about copyright and the like. But I’m so curious how you guys are doing it.
JASSY: Well, we have a group of people that spend a fair bit of time, all the time by the way because the landscape continues to evolve but thinking through the best way to build the models. We’re very thoughtful and very careful about what data we use. We, of course, have, you know, some proprietary data and some licensed data, and some publicly available data. And we use all of it. All sources —
SORKIN: You know, some people ask because and someone asked me this, you have all the books in the world, many on Kindle. Do you, are you able to train on that data?
JASSY: Again, there, there’s some of that data we can train on, and you know, again we try to be thoughtful about what data we train on. We also do licensing deals in some of these cases and in different types of data. And so, you know, it — to — these models, they’re getting smarter and smarter, and I think you’ve probably read about, you know, as you build these really big models, we have not yet — yet gotten to a point where the models are so smart that they wouldn’t benefit from more data. And so, you know, I think people are looking for data, but we’re being thoughtful about how we use it.
SORKIN: Does it matter that something like an Anthropic also is operating with Google? I mean, they have an investment in them as well. They’re — they’re on their cloud as well. I asked because — I look — you have $70 plus billion in cash, would you just prefer to buy the whole company if you could?
JASSY: We have — you know, we have a great partnership with Anthropic. And you know, it’s true with most technology, if you operate in the technology space, you get pretty comfortable with coopetition. It’s just the way it works. And so, we have a great relationship with them and we’re very comfortable with the arrangement. We’re excited about the investment. Remember, as part of our relationship, it’s not just that we offer their models in an easy to way use on Bedrock, they’re training their future models on our chips, on Trainium and they also named us as their primary cloud partner. And so, there’s a lot of collaboration there. We’re really happy about the relationship.
SORKIN: Do you think you could buy a firm like Anthropic in this environment from a regulatory perspective?
JASSY: I don’t know. I mean, it’s — I think we got to be careful right now in Western countries in the way that we’re handling regulation. I think a really good example of that is what happened with iRobot, if you follow that, you know?
SORKIN: Yeah.
JASSY: So I think it’s — it’s a really kind of a sad story. It’s a great entrepreneurship story where you had this — you know, this American company that invented this product, invented — invented the category and built a business that was almost a billion dollars in revenue. And whenever you build a good business, you end up with company and competition. They attracted these two very large Chinese companies as competitors, and they needed scale because scale lets you buy components at the right price and invest in R&D. So they merge with Amazon. And the European Commission blocks it because they worry that we’re going to feature our vacuum cleaner, you know, their Roomba versus others, which, of course, is not our model because we make actually at least as much money selling third party items as our own. But they block it and then immediately after they block it, the FTC comes out and says, oh, we would have blocked it if they hadn’t. And so, what happens we abdicate the acquisition, iRobot lays off a third of its staff, the stock price completely tanks. And now, there’s a real question whether they going to be a going concern. Turns out with these in-house vacuum cleaners, they have to map the inside of your house. That’s way they, way they don’t run into a table or a chair. So, really what Western regulators were saying was that they trust these two large Chinese companies with maps of the inside of U.S. consumers’ homes more than they do Amazon, even though we’ve been an amazing steward of customer data in our retail business, and for AWS, where customers of both those businesses will tell you that. That’s — that can’t be what we were going for. And so I think people don’t know what they can do right now, and I just think we got to try and find a way to be reasonable in what we’re doing.
SORKIN: You know, the FTC’s looking into even the partnership that you have with Anthropic. They’re looking at the way Microsoft’s relationship exists with OpenAI. I’m sure they’re going to look at Microsoft’s sort of “acqui-hire” deal. I don’t know what you think of that, as it relates to inflection. I mean, it’s very interesting what’s happening. Nobody’s making outright acquisitions but in some ways, these partnerships are in lieu of that. No?
JASSY: Yeah. I mean, there — there are some acquisitions happening but we’re, you know, we’re consuming a lot of time and taxpayers money with what we’re doing right now, and I think a lot of it is outside the bounds of the law right now.
SORKIN: Outside the bounds of the law in terms of how far you think the regulators are going?
JASSY: I think that, you know, some of these organizations are making decisions that are outside the bounds of the law.
SORKIN: Let me ask you this since we’re on the topic of regulation. I want to get into what you’ve done with sellers over the past year, and just what’s happening on the consumer side of the platform which is a remarkable thing given how you change the distribution model and the like. And you now have more sellers on the platform than ever. But at the same time, you have the FTC out there saying that you are a monopoly, that one out of every $2 that a seller on the platform is getting is going back to you in the context of advertising, servicing the distribution and the like. What do you, what do you make of that?
JASSY: Well, I think you got to start with facts, you know? And if you look at on Amazon, over 60 percent of the units we sell are sold by third party sellers. Seller — you know, it’s not hard to actually create software to put up an e-commerce website or storefront. It’s much harder to get distribution and access to customers, which is what Amazon gives sellers. You know, our sellers on average sell about $230,000 in our marketplace. We have thousands of sellers who sell over a million dollars a year in our marketplace. So sellers are making a lot more money selling on Amazon than they could on their own. You know, then if you look at things like fulfillment by Amazon, which is our service that allows you to store your products with Amazon, we’ll pick back and ship it for you and it also is available for prime shipping, that saves sellers over 70 percent from having a do it themselves. So we — it costs money, so we charge a fee for it, but it’s much more cost effective for sellers. And while there’s always things we can be doing better for sellers and we work really hard at it, we have a great relationship with sellers.
SORKIN: Is it a better business you think to have a third-party marketplace or to have a first party marketplace? Give — given — no, and it’s very — you know, if you look at just the margin and the fees that that you’ve been able to accumulate from third parties, is that a better business? Is it less capital intensive? Is it more capital intensive?
JASSY: It’s a very good question. They’re both really interesting businesses. You know, on the — one of the main reasons that we have first party business is that it allows us to make sure for the items that customers care a lot about that otherwise wouldn’t happen, that that you can keep the prices as low as possible for customers and keep the in-stock levels the way you want. But wherever we have third party sellers who are selling, it’s always a great value proposition and you know, they sell the majority of units for us now.
SORKIN: One of the things that you write about in this — in this letter is this idea of primitives. And I just wanted to talk about it for a second so that the — those who follow Amazon understand it’s really a way you think about building new parts of the business. Can — can you give me sort of a 60-second overview for those who don’t haven’t had a chance to read this letter yet what that even means?
JASSY: Yeah, so think of a primitive as a foundational building block that does one thing really well and that people can stitch together however they see fit. And when we had — when we came up with this idea of primitives about 20 years ago, most of the software and technology services out there were these all singing, all dancing services that did lots of things but none of them very well. And what we ran into — and we were building that way too. We — you know, we had a number of these e-commerce services like payments or ordering or item catalog or search, that when we started doing deals with companies like Target where target.com used Amazon’s e-commerce technology under the covers, to try and provide the individual services through API so they could customize however they saw fit was way harder than we imagined because all of our services had depended on each other and become jumbled up. And so, when we — around the same time when we were building AWS, we were really convicted that we’re going to build the set of foundational building block primitive services like storage or compute or database that people could stitch together however they saw fit. And the two big advantages are, it really fosters creativity because, you know, you can combine the components however you want, and it fosters speed because if you want to actually use one component, you don’t have to ask five services to make a change. You just use that component.
SORKIN: And now, you’re using that across the board for just about everything that you build.
JASSY: Yeah.
SORKIN: Including on the consumer side, one of the things I said at the top of this was you really redid the entire distribution model of how you get products to consumers and you’ve been able to get them more quickly than ever before. We were shocked, up 70 percent in terms of one — 70 percent —
JASSY: Same day, one day courier.
SORKIN: Same day, one day — what is that going to look like in two or three years from now? And can you continue to bring the cost down because that’s one of the pieces of this?
JASSY: I think yes. You know, I think — we’ve been — last year, we’re the fastest shipping speeds for Prime customers in the history of the company. We’ve been even faster in the first quarter of this year. We have a lot of ideas and how we can continue to take delivery times down. I mean, I talked about we have these same day facilities where we have, you know, we store about our top 100,000 skews there, but we also have nearby fulfillment centers that can inject selection in there. And they’re in our major metro cities. And we can those, those fulfillment centers are so streamlined that we can get from the time you pick an order to when we’re ready to ship, in as low as 11 minutes. We’re doubling the number of those. And then when you think about what we’re doing with drones, with Prime Air, you know, I think in several years when they’re more pervasive, we’ll be able to get items to customers in less than an hour.
SORKIN: Okay. I got to ask you about returns, because this is also a big issue and there’s a whole lot of folks that are very frustrated about what they call fraudulent returns I didn’t even realize that this was a thing. So people are buying stuff on Amazon, often times from third parties, and then they return it, but they don’t return the actual product so it would be, you know, I’d buy these sneakers and then I’d return a fake version of the sneakers back and they’re grappling with that. Is that — is that something — how do you deal with that?
JASSY: Well, you know, in our scale, you find — you get some of everything. You know and so and we’ve seen that in the past, but we have a pretty substantial team that works on making sure that returns are appropriate, that we keep counterfeit really low. And so, you change a little bit of some of your policies on returns, on when you actually refund people, and you know, you try to make sure you know you got the right product before you actually authorize the refund.
SORKIN: What about stolen goods? Because that’s another big issue that a lot of — lot of brick-and-mortar retailers worry about. You hear the complaints all the time they say people are robbing us and then they’re taking that product and they’re selling it as a third-party merchant on an Amazon, on an eBay, on these other kinds of services.
JASSY: Yeah, it doesn’t happen — I mean, again, at our scale, you find some of everything but we have a — we have a really significant team that that looks for that and to get off — it’s not — you know, you have to get authorization as a seller. We have a group people looking for items. Whenever we get reports, we act on it. So we have a group of people that are watching that very carefully.
SORKIN: One of the stats this year that’s just mind-bending is just how big an advertising business you have. I mean, just — it’s really grown quite remarkably, growing 24 percent year-over-year, $38 billion in 2022, now $47 billion in 2023. How should we think about Amazon in that regard? And what is it — where does this go?
JASSY: Well, you know, look advertising doesn’t work unless you provide the right experience for customers and they respond to it and then it actually benefits the brands. So, we — you know, most of our advertising are actually machine learning practitioners who are actually figuring out the right products to put in front of people. So it doesn’t even feel like advertising to them. It actually feels like the item that they were searching for. And, you know, we have a product called Sponsored Products that has been very successful in this realm that we’ve expanded with sponsored TV. And you know, it’s — you know, the sponsor TV offering allows people to have, you know, self-service access to entities like Freevee and Twitch and things like that, but we’ve recently added the ability for people to do advertising on Prime Video shows and movies, and that’s very early stage. But, you know, I expect that we will continue to very thoughtfully find ways to place advertisements in the different entities we have.
SORKIN: When you think though about Prime Video, one of — one of the sentences that stood out to me in this letter was that you now think that Prime Video could be effectively a profitable and completely standalone business. What does that mean?
JASSY: Well, I think that when we started with Prime Video, it was — it was very much about driving people to find value in our overall prime offering. And we found that actually a lot of people would sign up for Prime because of our exclusive content, and then once they become Prime members, they would shop in our e-commerce offering. And so, it really drove our business downstream and that continues to very much be the case. But we now have conviction that apart from the value it drives for the rest of the business, it will be a good economic business on its own as well.
SORKIN: The NBA, does it have a place on Amazon?
JASSY: I assume you’re talking about their upcoming —
SORKIN: Yes.
JASSY: Well, we’ll see. I mean, it’s hard not to be impressed with what Adam Silver and the NBA has done. It’s just an incredible product and it’s an international product, and we have a lot of respect for them and we’ll see what happens.
SORKIN: I mean, how big are — how important are sports though for this? And could you ever see yourself spinning this business off? Is it separately or is it so intertwined for the rest of the business?
JASSY: Did Jon Fortt ask you to ask me that?
SORKIN: No, no, I’m genuinely curious.
JASSY: No, I don’t foresee us spinning it off because it’s so tightly integrated into our Prime offering. But I do think that sports — I mean, if you look at the ratings of shows every year, sports occupies 90 plus percent of the top ratings. And so we’ve seen with Thursday night football just how many customers love that offering and how many people come to it and how many prime subscriptions it drives. And so, you know, Thursday night football has been a huge success for us. We’ve got NASCAR coming. We’re going — we’re going to have six races right in the middle of the year. We have soccer through UEFA and Champions League. If you watch, I don’t know if you watched the women’s Final Four this past weekend —
SORKIN: Yes.
JASSY: — which was just unbelievable. You know we have — we’re the exclusive streaming partner on the WNBA, so we’ll get to see Caitlin Clark on Prime Video as well. And so, sports are very attractive for our customers and you can expect us to do more.
SORKIN: Would you do more deals. We keep talking about Paramount, by the way, all morning uh because there’s lots of news going on there. You bought obviously MGM. Could you — would you — would you want to buy Paramount if you could?
JASSY: I think we’re pretty comfortable with the offering we have right now and we’re continuing to produce, you know, I think really great exclusive content. I don’t know if you’ve seen Fallout yet — yet or not?
SORKIN: Yeah.
JASSY: It’s — it’s got great reviews. And so I’m really pleased with the content we have and we’re just starting.
SORKIN: Kuiper, let’s just talk satellites real quick, because we — I’m trying to hit everything we can.
JASSY: I like that you said let’s just talk satellites real quick.
SORKIN: It’s very hard to talk satellites real quick.
JASSY: Yeah.
SORKIN: When do you think we’re actually going to see this product and this would compete, of course, with Starlink?
JASSY: Well, you know, Kuiper is our low earth orbit satellite business that we’re building. There’s 400 million to 500 million households across the world that don’t have broadband connectivity. They can’t do the basic things we take for granted. And in October, we got two end to end prototypes up where we were able to exercise all the functionality, which is really rare and a prototype test like that. And so, we’re putting the satellite up. The first big production pieces will be the second half of ’24 and we expect to have the service up in the next year or so.
SORKIN: And then finally, real quick, I talked last week to Steve Cohen, the investor, and we were talking. He has a very interesting thesis. He thinks we are completely moving to a four-day workweek that will be like truly established and will change economies as a result of it. I know you were — we were all struggling to get people back to — back to the office to begin with. What are you — what are you seeing right now? And do you actually think there is a complete change in how it therefore affects the — affects the economy.
JASSY: I think — you know, my own view is that I think we’re on a journey and we still don’t know yet where we’re going to end up. And you know, I do — what I’ve noticed since we’ve gotten people back in the office at least three times a week is that the collaboration is much better, that if you do a lot of innovation, which we do, just — you know, the way you innovate, it’s not like you schedule an hour and you innovate and you invent something and you’re done. It’s — it’s messy, it’s meandering, it’s wandering, it often takes 90 minutes. When you’re done, you don’t quite get there and two people get on a whiteboard and work it out, you know, after the meeting. And so, I noticed the collaboration is better. The innovation is better. People understand the culture better. They’re more connected to one another. And I just noticed that we’ve kind of gotten back to our culture more since we’ve been back in the office.
SORKIN: Okay. We’re going to leave it there. Andy Jassy, it has been quite a remarkable year, the stock up 22 percent. It’s near an all-time high. And we appreciate seeing you.
JASSY: Thanks for having me.
SORKIN: Thank you for touching so many different topics this morning.
