# Amazon CEO: $220 Billion Still Can’t Meet AI Demand The a16z Show · 2026-10-23 · 61 min · https://www.youtube.com/watch?v=-YMMRQ1qscI ## Transcript Andy Jassy Even spending $220 billion this year, we’re leaving a meaningful amount of demand on the table. We always leave a certain amount for on-demand, but 2026 is obviously sold out. 2027 is largely sold out. A good chunk of 2028 is sold out. The financial markets are saying, “Oh, there’s not enough money in the world to finance these things.” Everything we do in the world today—the stock market, banking, health care—is using data centers. Our job is to try to figure out how to have as much capacity as we can. We have an opportunity to help across so many more customer experiences and make people’s lives better and easier every day. The big labs are willing to pay almost anything for power, cooling, and chips. Well, maybe we should buy out the old guys and sell to the new guys, because they’ll pay us whatever. Erik Torenberg Andy, welcome to the podcast. It’s great to have you. Andy Jassy Thanks for having me. Erik Torenberg So you’re five years into running Amazon, and you’ve done a lot since being here. A lot is coming at you as CEO. Right now, you have $220 billion in CapEx spend, and AWS is growing 28%. You’ve achieved a lot. How do you think about the ambition in 2026 as it relates to Amazon, compared with 2021, when you formally took over as CEO? Andy Jassy Oh, 2021 feels so long ago. It feels like— Erik Torenberg Eight lifetimes. Andy Jassy A decade ago. Yeah. The world changes fast, and if you compare it with 2021, our overall top line is about 80% higher. The AWS business is almost three times bigger. Our ads revenue is about two and a half times bigger. The business has grown quite a bit, and I think a few things have really happened. Pretty quickly after I started in the role, on the tail end of the pandemic, our retail business ended up with a real cost-to-serve challenge. We had basically doubled the size of our fulfillment network—which we’d built over 25 years—in about 18 months. Then we built a last-mile transportation network the size of UPS in about 20 months. Whenever you build that quickly, you end up with all sorts of things that aren’t optimized. Our cost to serve and our fulfillment network were not where we wanted them to be. That created a lot of challenges and angst. About a year—or a little less than a year—into the job, people were saying, “Boy, what a terrible time to start,” and, “What bad luck.” Erik Torenberg Mm-hmm. Andy Jassy In retrospect, I think it was really a great time to start, and a real blessing, because it allowed us to consider a bunch of things we probably wouldn’t have considered for many years. They turned out to be really good ideas for our customers and for the business. The biggest one in our fulfillment network was that we regionalized a flat, national fulfillment network in the U.S. We got items much closer to customers, which lowered our cost to serve because transportation distances were shorter, while also getting items to customers much more quickly. That reevaluation of our most closely held beliefs, during a time when we were struggling with cost to serve, really unleashed speed of delivery and a lot of demand in our retail business. It’s a big part of why we’ve grown as fast as we have there. Pretty quickly after that, AI exploded further onto the scene. AI wasn’t new; we’d all been doing machine learning for a long time, and there had been lots of waves of AI that were supposed to be the real tipping points. But by 2023, you really saw a huge difference with generative AI. Our AWS business was growing really quickly at that point. We had the broadest functionality by a fair bit, stronger operational performance and security, and a different kind of partner ecosystem. We were the clear leader, and we felt that business could be a few hundred billion dollars in revenue over a long period of time—figure over 10 years. Now we look at what’s happening with AI and how every customer experience we know of is going to be reinvented with inference and AI. We think the AWS business is growing really rapidly. Last quarter, we grew 36.7% year over year on a $169 billion annual revenue run-rate business. Erik Torenberg Wow. Andy Jassy The opportunities today are really striking. And if you layer on top of that the seeds of new businesses—whether it’s how the team has progressed our Prime Video business, which is now a large and profitable business; the low Earth orbit satellite we’re building with Amazon Leo; the autonomous-driving ride-hailing service with Zoox; our health care business; or what we’re doing in robotics—we have an opportunity to help so many more customers across so many more customer experiences and make their lives better and easier every day. That mission of trying to make customers’ lives better and easier every day really unifies everything we’re doing across the company. Erik Torenberg A couple of follow-ups on that. You mentioned that when you’re building out the fulfillment network so fast, it’s going to be a little inefficient. When you go that fast, there’s some sloppiness to it. And you mentioned there was angst around that internally. From a CEO perspective, how do you communicate what you’re doing without lowering the cultural standard? Amazon is known for its excellence in operating efficiency, being frugal, and all these kinds of things. But here you are saying, “We’re going to build UPS,” and— Ben Horowitz And so how— Andy Jassy Yeah. Ben Horowitz —do you reconcile those? Andy Jassy What’s interesting is that we have this set of leadership principles that a lot of people know about. We have 16 leadership principles, and a lot of them are supposed to—and do—govern the way we think about operating as a team and culturally. But several of them are in tension with one another, and sometimes people are confused about that. The reality is that it reflects the way the real world works. Nothing is perfectly clean. Ben Horowitz Yeah. Andy Jassy You’re always making trade-offs. If you go back to the time when we made the decision to build out the fulfillment network and the last-mile transportation network that aggressively and that quickly, the question the team was facing was: We know that if we go this fast, it’s going to be really hard. It’ll probably be more expensive than if we went slower. We probably won’t get everything as optimized as we would if we went slower. But customers and sellers really need it. Are we actually going to shade on the side of enabling them? Remember, in the early part of the pandemic, no physical stores were open. If you look at the role Amazon played in everybody’s lives during the pandemic, it was very seminal. We had to ask ourselves: Are we going to build fast and maybe not as optimally, with the trade-off of serving customers and sellers who are trying to have businesses that sustain themselves, or not? That’s the first decision you have to make, and we decided, as we almost always do, to shade on the side of customers. Then it was fairly obvious as we were building that not everything we were planning was going perfectly. Our fulfillment team was more in the office than most of the rest of the company, but a lot of people who had to participate in the designing, planning, and assessing were all over the place and remote. Now you’re balancing the tension between bias for action, or speed, and relentlessly high standards, which means delivering at the cost structure, delivery, and design you’d always planned for. Again, you have to make trade-offs. We decided to side with customers and move quickly so we could serve sellers and consumers, and we thought there would be some amount of cleanup. Did we know exactly how the topography would work out? No. There were certain things we didn’t have optimized that I’m not sure we foresaw wouldn’t be optimized. But if you have good metrics and a truth-telling culture—which is another really important part of our culture—you get the data and issues on the table, and you’re direct with one another. We don’t allow people to paper over issues. With that type of culture, it’s fairly obvious in the metrics and in the discussion. Then you just have to figure out how you want to improve it, and which pieces are small productivity things and which are architectural. Andy Jassy We went back and really re-architected key parts of the network after we got that capacity installed. At the time, that was probably eighth or ninth on that team’s list of priorities, but it became a much higher priority as a way to get the combination of speed and cost. Ben Horowitz Yeah. Oh, fascinating. At the beginning, when you started AWS, it was kind of “build it, and they will come.” You were building all this infrastructure, and then you were going to rent it, so they needed to come or you’d be in a lot of trouble. But this is at a much bigger scale. You and others—but I’d say primarily you, or at least you’re certainly one of the biggest players—are building data center and chip capacity so fast that the financial markets are saying things like, “There’s not enough money in the world to finance these things.” We just saw that skittishness in the CoreWeave debt market and equity markets, and so forth. How do you know how much is too much on that build-out, given that demand is so steep and you’re growing so fast? And there’s contention for the supply chain you’re fighting for to get your cooling, power, chips, and all that. Andy Jassy Yeah. It’s a very good question, and I wish there were an equation that would make it simple. It would save us a lot of time as a team. In the first wave of cloud, when we started AWS, there was a little bit of “We didn’t know if people would use it,” so it was “build it, and they will come.” But we had very good signals inside the company. This was in the early part of the 2000s. We had teams that weren’t moving nearly as quickly as we wanted them to, even though we felt that, relatively speaking, we moved quickly. We just couldn’t understand why it took so long. It turned out that our product teams were spending 70 to 80 percent of their time on infrastructure pieces that didn’t differentiate them, and only about 20 percent of their time on the pieces that differentiated the product. We realized every team was doing it and reinventing the wheel. We originally conceived of AWS to allow ourselves to move more quickly internally. We figured that if we built a set of high-scale, high-availability, low-cost, highly flexible building-block services—compute, storage, database, content distribution, messaging, and things like that—we’d be able to build much quicker in our consumer business. We figured we were pretty good, relatively speaking, and that other companies probably had the same issues. Even before we launched externally, so many teams internally were interested that we knew we were on the right track. Ben Horowitz Mm-hmm. Andy Jassy I don’t think any of us anticipated it would grow as fast as it did externally. On the AI side, there have been these different waves of people saying, “This is the time. AI is finally here.” Even 6 or 7 years before the generative AI wave, people felt like it was AI’s time. We built a whole bunch of AI services that were actually a pretty good business for us on the AWS side, but nothing like what we’re seeing right now. In the early days of AWS, we felt like we were growing really fast, and we felt good about it. After 3 years, we were about a $60 million annual run-rate business. In the first 3 years of this generative AI wave, our AI business in AWS is over $25 billion in run rate, so it’s several hundred times faster. Ben Horowitz Yeah. Andy Jassy We just haven’t seen anything like it. Then the question becomes: How much CapEx do you want to invest to go after the opportunity? We’ve said that this year, in 2026, we expect to spend about $220 billion in CapEx. I’ll tell you, I remember being at a leadership team breakfast 15 years ago when we couldn’t believe the telco company spent $50 billion in CapEx. We were like— Ben Horowitz Wow. Andy Jassy “I'm only spending that much on CapEx.” These are mind-blowing numbers. Even spending $220 billion in CapEx this year, we’re leaving a meaningful amount of demand on the table. The thing that helps us is that we have really good visibility into customer demand. If you think about the CapEx investments in AWS and the way the cash cycle works, there are 2 macro areas: the investment in data centers, and the investment in servers and networking. For the data center piece, you have to invest about 2 years in advance of when you can put servers in the data centers and start to monetize them. The good news is you only make that investment once. It’s a useful-life asset of about 30 years, but you make that investment a couple of years in advance. For the server and networking gear, depending on the component, you only invest a few months in advance. So you have very good visibility into customer demand before you pull that trigger. If it turns out you don’t actually need the capacity to fill those data centers, you won’t buy the server and networking gear. On average, those investments tend to break even in about 3 years, and they have useful lives of about 5 to 6 years. Most of the deals we’re doing are about 5-year deals. When you’re building as fast as we are, with so many data centers simultaneously, you’re going to spend a lot of capital, and your free cash flow is going to have headwinds in those years until you get to those later stages, a few years down the road, when all those data centers are online and being monetized, and you’re making back your money from the server and network investment. Because of the way the cash cycle works, you have headwinds in the early years when you’re growing that fast. Ben Horowitz So, actually, on that: When you started AWS, data centers were probably the last thing in the world you would believe would be a political issue. But here we are. Particularly in the United States, it’s funny, and I’m sure you’ve experienced this, but nobody anywhere in the world is as anti-data-center as US politics, for whatever reason. How does that end up affecting this? I guess, is there any merit to the anti-data-center argument? And how does that inform your strategy for what you build here and what you build in the other regions? ### Data Centers Face Political Pushback Andy Jassy Yeah. I’ll give you some of the context that I have, or at least what I believe. If you think about everything we do in the world today—stock market, banking, healthcare, entertainment, shopping, education, business, ride-hailing—everything we do is digitally driven. They’re all using data centers in a very similar way. Ben Horowitz Yeah. No, everybody is: “I’m anti-me.” Andy Jassy Yeah, it— Ben Horowitz I don’t want to use my phone anymore. Andy Jassy And they all use data centers, and that’s before AI. AI is the biggest technology transformation of our time, which I think is saying something, since we’ve lived through the internet, the cloud, and mobile. The reality is that the type of infrastructure build-out we need to support the rapid growth we were just talking about on the AI side—there hasn’t been anything like it in the world, and in this country, since maybe when we built out the highways and bridges in the 1950s. That’s the type of build-out we’re going through. The country—and the world in general, but I’ll say the US specifically—cannot afford not to build data centers, or we’ll find that another country like China establishes all the leadership and all the norms there. I don’t think any of us are going to like that, either domestically or globally. So we have to build data centers. Do I think there are ways all of us, as an industry, can do it better than we have? Of course. It’s happening quickly, and I think everybody’s learning together. But there are a number of things I’d consider somewhere between myths and mistruths being propagated right now. It’s one of those issues where everybody has an easy villain, and politically, sometimes opportunists like to find that villain to point at. I’ll give you examples. The golf industry or the almond-farming industry consumes way more water than data centers do. Just as a data point, golf courses consume 200 times more water than all our data centers, and they consume more in 2 days than our data centers do in a year. That’s a good example. Or people talk about pollution. What happens is, when you register new sites and have backup generators, you have to register and show the amount of pollution there would be if you ran all of those generators full-time, at full capacity. But of course, that never happens. You only run backup generators 0.1% of the year, like 10 hours a year for maintenance. Ben Horowitz Right. Andy Jassy When you talk about energy rates, if you look at the different geographies where we have a big data center presence, some of the energy rates have gone up and some have gone down. But the real issue is that, as a country, we have not invested in the infrastructure of our grids the way we’ve needed to, given the meaningful growth that’s coming. Something was going to push us to invest, and it turns out that AI, because it’s growing so rapidly and so significantly, is what pushed us over. Once you actually get that capacity, in most geographies, you’ll find that rates flatline or go down, as they have in Indiana, Georgia, and Mississippi. Another thing people forget is that the big data center providers pay a lot of tax in these geographies. We’ve paid about $3 billion in tax in St. Joseph County, Indiana. The communities get that money and can spend it on things that really matter to them. We’ve built programs like Built Together, and we just committed another $1 billion to the communities where we have data center presence over the next 5 years. We have to keep working with the communities. It has to be a public-private partnership, but the build-out is so important to the future of the country. Imagine if we hadn’t built highways and bridges, and what that would mean for the country’s economic prosperity and capabilities. We have to keep it in perspective and be partners in the communities where we build. Ben Horowitz Let’s go back to the CapEx for a second. The $220 billion is $600 million a day, just to put it in perspective. Andy Jassy Yeah. Ben Horowitz I’m curious how you think about AI demand right now. Offline, we were talking about how you see it as a barbell. Maybe you can situate or contextualize the state of AI demand at the moment. ### AI Demand Meets Supply Limits Andy Jassy The first context piece is what I said earlier: I can’t believe how big and fast AI has grown. It’s grown a few hundred times faster than AWS did in its first 3 years. We have a $25 billion run rate for AI revenue in AWS in the first 3 years. It’s really large, and yet demand really is barbelled. On one end of the barbell are the AI labs, which are consuming gobs and gobs of compute, unlike anything I’ve seen, and a few runaway apps like ChatGPT and Cloud Code. On the other end are enterprises, which are largely using AI for cost avoidance and productivity, like automating customer service, fraud detection, or business process automation. In the middle of the barbell are enterprises’ current production workloads. Many are using AWS and using inference today, but I’d say the overwhelming majority still aren’t using it in any pervasive way. I think that’s going to be the largest absolute segment of AI usage. It’s hard for me to predict the rate at which that’ll happen. It’s happening to a greater extent every year, but if you zoom out 5 to 7 years, I think quite a few, if not most, of the enterprise applications that exist today are going to be using inference pervasively. I don’t think it’s going to wait 5 to 7 years; it never happens in one fell swoop. There’s a lot of growth in front of us. Ben Horowitz How do you think about it? There’s an impending shortage of data centers, chips, memory, everything, really in 2027, because everybody’s building now, but it’s not necessarily going to be ready by the end of the year, and demand isn’t slowing down at all. When faced with the big labs being willing to pay maybe anything, or almost anything, for power, cooling, and chips, while you also have customers who’ve been around a long time and are paying their regular rates to run their businesses, is there any business pressure to buy out the old guys and sell to the new guys because they’ll pay whatever, while you can’t double the old guys’ prices? That wouldn’t be fair. How do you think about that? It’s such a— Andy Jassy Yeah. Speaker 2 —weird thing, in that you’ve got people who’ve been there for 20 years doing something that, as you say, is probably going to inference, and then you’ve got these new guys who are insatiable. They’ll buy anything. Andy Jassy Yeah, it’s a good question. We don’t think of it quite the way you’re talking about, but I can completely understand. If you think about it, there’s only so much data center capacity and power today. Speaker 2 Mm. Andy Jassy There’s only so much memory, and I don’t think that shortage ends in 2027. I think— Speaker 2 Yeah. Andy Jassy One of the reasons I think we can invest a lot more and it wouldn’t be a bad use of capital is that 2026 is obviously sold out at this point. 2027 is largely sold out. We always leave a certain amount for on-demand in our inference services, Bedrock or EC2, but largely all the big chunks are sold out. A good chunk of 2028 is sold out. There’s so much demand. You can’t—if you think about the AI business and what we call the core business, meaning non-AI, they’re both growing at really rapid rates right now. If you have only a certain amount of memory, you have to decide how much to allocate to HBM, to AI accelerators, versus DRAM for your core business. These are real trade-offs you have to make, but where we start on this issue is being customer-focused. We have deep relationships and partnerships with our customers in AWS. During the pandemic, unlike our competitors, who I think tried to optimize within quarters to drive revenue as much as possible when companies were going to cost-optimize, we took a different approach. We proactively had our field teams help customers figure out how they could cost-optimize, even though we knew it would diminish our revenue in the short and medium term. Our strong belief is that if we help customers over a long period of time, our business will come along with it. It’s the same thing here with the core business. I guess you could calculate that, because of the scarcity and their willingness to pay, AI customers would pay more and you could squeeze out your core customers. But those customers are running their businesses and all their core applications on AWS. We’re not going to squeeze our customers, and we’re going to make sure they can scale. Our job is to figure out how to have as much capacity as we can, so that not just our core customers but all of our customers who want to use us on the AI side can do so in a way that’s economically sustainable for us as well. I wish I could tell you that I thought the supply constraints would be just a 2027 issue. I think we’ll see them in different components for several years. We’re probably putting in more capacity than anybody else because we’re very passionate about continuing to be the significant leader in the infrastructure and AI space, as we’ve been. We’re spending about $220 billion this year, and we could spend a lot more. It wouldn’t be a bad use of capital in the medium to long term, and I think we still probably wouldn’t satisfy all the demand. There’s incredible demand right now. Speaker 2 Speaking of that, you did something back in 2015 when you bought Annapurna Labs, which was probably before—you probably weren’t really thinking, maybe you were, about giant AI demand in 2015, because there just wasn’t much AI demand then. What drove that decision, and how has it paid off in this new world? Andy Jassy Yeah. I think the Annapurna team is one of the most talented teams and sets of people I’ve ever had a chance to work with. Andy Jassy It’s one of the very best acquisitions we’ve done at Amazon in the history of the company, if not the best. We foresaw some pieces, but we did not make the Annapurna purchase for AI chips per se. When you get to a certain scale, you start to have access and visibility to patterns of customers at very large scale. You realize that if you can vertically integrate and build certain capabilities yourself, you’ll be able to provide more value for customers and better economics for the business itself. What Annapurna started as for us was a solution for the core EC2 server. We felt like we were using too much of the server for virtualization, networking, security, and storage, and we wanted to be able to use nearly 100% of the server for compute for customers. We had been working with Annapurna before we acquired them on a NIC that offloaded those components, allowing us to use as much of the server for compute as possible. We really liked the team, and we really liked the product we collaborated with them on. We realized it was going to end up in virtually every EC2 server, and it’s turned out to be in every server we have. We originally bought Annapurna because that was going to be a core capability of our compute capabilities, and because the team was so talented at building chips, we felt like they could probably build other chips, too. At that time, we had a very deep partnership with Intel, which we still have. Intel had virtually all of the CPU share, and when you have a partner with virtually all the share, they’re not often thinking about how to drive price-performance. They’re trying to think about other features that allow them to get more share and drive more revenue. By the way, it’s totally understandable. Those are some of the dynamics of how markets work. It’s part of how the innovator’s dilemma comes up. We had tried to push on price-performance, and it was hard for us to get attention there. But we had customers who wanted to move more and more of their on-premises workloads to the cloud, and they wanted better price-performance. The next thing we turned the Annapurna team loose on was building a CPU chip. They built Graviton, which 99% of our top 1,000 EC2 customers use very pervasively today. It’s about 40% more price-performance-efficient than other x86 processors. That has been hugely impactful for our customers and for our business. A few years ago, we saw that same story in the AI space, where you really had one dominant provider who’s very good and with whom we have a deep partnership. We still have a deep partnership, and we will for as long as I can foresee. But they have giant markets and lots of different products and priorities, and we knew our customers needed better price-performance. That’s why we spun out a separate team in Annapurna to build an AI accelerator, which we call Trainium. It’s also about 30% to 40% more price-performance-efficient than comparable processors. When you talk about that middle of the barbell we were just discussing, what are the things that will change that middle of the barbell, so more of those current enterprise production apps use inference? There are a couple of things. Some of it is having talent that’s comfortable doing post-training and fine-tuning, and having the talent and organizational muscle to know what to prioritize and how to do it. But a big chunk of it is the cost of inference. The cost of inference has come down quite a bit in the last few years, but it’s still an order of magnitude too expensive, in my opinion, for it to be used as pervasively as possible. If you’re asking enterprises to rewrite their applications that use CPUs today to use AI accelerators, and the difference in cost is significant, then the incremental benefit they have to get from using AI accelerators is a much higher bar for rewriting those apps. But if you can take the cost of inference down, that bar is much lower, and all applications are going to be better, smarter, and more capable with AI accelerators and inference. Driving the cost of inference down is a really important part of enabling customers to use AI pervasively, and our chips business is a big piece of that. Ben Horowitz Yeah. I can tell you from our own firm that it can get very expensive. We had one employee spend $2 million. Anyway, he was doing good work, but— Andy Jassy Yeah. Ben Horowitz I was like, “Okay, this inference is not that cheap.” Andy Jassy Yeah. Speaker 2 One of the things you’ve done is make some investments, particularly in Anthropic and OpenAI—very large investments. I think, like, $18 billion in Anthropic and $50 billion in OpenAI or something like that. What got you to want to do that? Similar investments from Nvidia and others got Wall Street up in a tizzy because it reminded them of the bad old days of the dot-com boom. How did you think about that, particularly since they are also customers? Why did that make sense? Andy Jassy Both of those companies are really impressive teams and really impressive companies. The rate at which they’ve grown their businesses, at least on the top line, is something I don’t know if we’ve ever seen before. It’s unbelievable. For us, it’s a combination of who these teams are, the technology and models they’ve built, some of the applications they’ve built, our belief in how transformative AI is, and our belief that these are going to be two of the most successful companies in this space over a long period of time. Not everyone’s going to be a winner, but there are going to be a lot of very successful companies in this space. We believe these will be two of them. Then there’s the fact that we partner so closely. Starting with Anthropic, where we’ve had the longer partnership, we have so many customers that run Anthropic in Bedrock, our inference service. I think customers are most successful running Anthropic in Bedrock. We’ve had a deep partnership with them on chips, where they’ve been building on top of Trainium for a while. I think it’s been very good for Anthropic and for our Trainium team. We’ve both learned quite a bit from each other. We have a deep partnership with them, and our belief in the future of AI, our belief in those companies, and the partnership made it sensible for us to make investments as part of those partnerships. On the OpenAI side, we had so many customers who wanted to use OpenAI on AWS, and they weren’t able to do so because of prior business arrangements. For a long time, I think the companies had a lot of mutual respect and a desire to work together. Over a series of conversations and months, we developed the ability to run OpenAI in Bedrock. We built a unique product together in the stateful runtime environment, which is how I think all agents are going to be built moving forward. You’re going to need state. Like Anthropic, they made a multiyear, multigigawatt commitment to Trainium. We believe that OpenAI is going to be one of the most successful companies in the space over a long period of time. All of those things made it make sense for us to have deeper partnerships and investments. ### Rebuilding Amazon For AI Ben Horowitz Switching over to how you use AI to build a company: I think you said the Bedrock team rebuilt your inference engine with 6 engineers in 76 days, and that would have taken 40 people a year. That’s pretty different. Amazon is kind of famous for the two-pizza teams. What is your current organizational design thinking, given the technology? It does seem to really change things. We got super specialized over the years, with designers and product managers, front-end engineers and back-end engineers, and this and that and the other. Now all of a sudden, everybody’s a Renaissance person. How do you think about that in terms of organizational design? Andy Jassy Yeah, I agree with what you’re saying. I think the Bedrock experience you alluded to was very eye-opening for everybody here. We spent a bunch of time sharing that story internally to inspire people and remind them what’s possible. Over time, what ends up happening—I started the company 29 years ago. I started as an assistant product manager, and I was the product manager for, I don’t know, five products. Today, when companies get successful and bigger and the products get bigger, you end up with a product manager who works on one product and owns just one slice of it. They own pricing. Everybody becomes more specialized, and everybody accepts that’s the way you have to do business if you’re building something big and successful. It isn’t. And it’s definitely not the case when you look at what’s happening with AI. Even just this past week, we’ve been working on some things around agents in our consumer business. We have a whole bunch of investments there, but we had an idea we really liked. On Thursday, we agreed we were going to try to have a real working prototype. Four or five people worked on it over the weekend and had it operating in an 80%-shippable way by Sunday. Those are different worlds that you live in. I think that, for the future of organizations, number one, we’re going to have fewer humans doing the work they’re doing today than we have today. Look at the amount of coding being done agentically, the amount of customer service, business intelligence, and research, and think about how spreadsheets are going to be built, and the number of lawyers you have. I just think you’re not going to have no people, but you’re not going to have as many humans as businesses have thrown at scale as they’ve gotten bigger. There are going to be different jobs and new jobs, some of which we know about and some of which we don’t, that will arise as AI, robotics, and space develop. There are going to be all sorts of new jobs, but we’re going to have fewer humans doing what they’ve done today. I think teams will be much smaller than they are right now. You’ll have smaller teams with people who can work cross-functionally. They won’t only own one sliver, because the models are going to get them 80% of the way there on a lot of these different functions. I think every team will start at a much more advanced stage than we have in the past, so you’ll be able to invent much more expansively and quickly. I do think a bunch of things have to change. Startups starting today are mostly starting agentically and agent-native, or AI-native, right from the start. All the larger companies that have been in business for a while are going to have to figure out how to build at that same pace. Some of them will figure it out. You always like it when you have a big business with a lot of capabilities and a loyal customer base. Some of them will move very slowly to figure it out, and they’ll be more vulnerable to startups, in my opinion. I predict companies will have to rewrite almost all of their code to be agentically capable. They’re going to have to allow agents to update different pieces of code, and they’re going to have to build constraints there. If you look at security, it’s kind of interesting. All the nervousness of the last month or two about models has focused on the more fantastical scenarios. By the way, they’re possible. Is it possible an evil botnet will take over everything? I guess it’s possible, and it’s possible you can infuse robots with malicious intent. They’re all possible. I personally think we’re going to figure that out together as an industry and— Ben Horowitz Mm-hmm. Andy Jassy —the private and public sectors. But— Ben Horowitz I think you’ve got to start by paying attention to some of the things. Andy Jassy Yeah. But the interesting thing is that I don’t think there’s enough discussion about the more immediate-term risks, which are really around cybersecurity. Ben Horowitz Yeah. Andy Jassy These models find so many more vulnerabilities than people did before, and that humans weren’t finding. Maybe you could argue whether they could or couldn’t, but they just weren’t finding them. If you fast-forward, I think applications are going to have to respond to imminent security threats in minutes. Today, you find vulnerabilities, people classify them, and people develop patches. Even if you do that part quickly, most companies take a long time to deploy because there are all these dependencies. You’re going to have to deploy in minutes. If you don’t have your applications rewritten so they can be updated by agents, you’ll have no chance in that new world. I think all organizations are going to be reinvented, and the companies that prioritize this are going to be in much better shape than those that don’t. Ben Horowitz Yeah. It seems like we, as an industry, didn’t do a great job of architecting everything, from big systems to embedded devices, with good security architecture. The day of reckoning is here. The good news is that’s a big opportunity for new companies, new efforts, and people to do things. But from what we’re seeing, that’s the threat. The theoretical, thought-experiment threats are, right, maybe they’ll happen, maybe they won’t. But this is clear and present. Andy Jassy Yeah. Almost every partner and customer we talk to on the AWS side, every company of any size, is worried about the security piece. We have a professional services practice where we’re helping a lot of our customers, and we’ve built a service called Continuum that identifies a lot of these vulnerabilities and builds patches. But it’s easier to say than to do. For a lot of companies and engineers, it isn’t exciting to focus on security. It’s easy to wait until there are problems, and if you protect data that matters, that’s a bad idea. Ben Horowitz Well, and also, going back to our earlier conversation, specialization has probably been the worst case in security. You’ve got these very esoteric security engineers who don’t like working with the other engineers, who are a scarce resource, and so on. As a result, in most software companies, their security practices just aren’t that good for writing software. That obviously has to change. The good news is we have tools that will help with that, but— Andy Jassy Yeah. And the good news, to your point, is that you’re going to be able to—you have to hire pen testers, and there are only so many security people you’re going to hire and only so many applications they can get to. The models coupled with automation are going to allow you to better pen test your entire surface. You’re going to find things you might not otherwise have found outside of a fluke. Ben Horowitz One of the things that’s been interesting as we’ve looked at the market is that you can do a lot of things with AI. One is you can cut costs. You can reduce the number of accountants or lawyers you have, or do that kind of thing. There are actually venture capital firms that are like, “Okay, we’re going to buy old companies and cut their costs.” That’s never been that interesting to us, because we’re like, “Okay, if that’s your idea, you have very few ideas.” Then there’s, “Okay, we’re going to build something the world’s never seen before,” and that’s super exciting. But there’s a third thing you’ve done with Alexa, where you’ve had a successful product and said, “Okay, let’s rethink this whole thing from the beginning because we have this new platform.” How do you get to that, and where does that make sense? Andy Jassy Yeah. I think that’s an interesting delineation of the three elements. I think organizations, as we just talked about, are going to change and get more streamlined. If you do it right, it will enable your really good builders in these small teams to build that second bucket you talked about: things people haven’t seen before. You’re going to be able to do it so quickly. When you have experiences like I was talking about earlier, where in four days you completely build a new idea that’s 80% of the way there, it’s so inspiring to people that they spend more of their free time and free thought cycles thinking about new ideas. To me, that’s one of the most exciting parts of what’s happening with AI. But I also think that existing companies with successful customer experiences at scale will very likely not have the winning experiences three to five years from now if they think about AI as primarily just incrementally improving the experience. I think every one of these experiences we know, even the ones that seem most difficult to imagine changing, is going to be meaningfully different with AI in a few years. And so it means that if you want to think more than incrementally, you have to find a way to get teams to reimagine from scratch what those applications would look like if you were building them today with all the new technology available, like AI. In Alexa, I felt like it was fairly obvious that we had to do that. Ben Horowitz Mm-hmm. Andy Jassy The brains of Alexa were really inventive at the time it was built, but generative AI made the capabilities of what those brains could be so different that you had to rewire them if you wanted to provide a great customer experience and be relevant over a long period of time. That was fairly obvious. Even though it was obvious, it was actually difficult, because teams are connected to what they’ve built and proud of it. They have hard-earned scars, and they know the tenets of what works and doesn’t work. Trying to get teams to completely start over and reimagine takes time, because they’ll say, “That’s not how we’ve done it. We’ve tried that, and it hasn’t worked.” What’s harder is taking other experiences—our retail business, Prime Video, Fire TV, or Prime Music. It’s not as obvious that the technology has a much better solution for those experiences today. But I believe we need to reinvent every single one of those businesses, and we have teams of people working on totally reimagined versions of them. Do I think we’ll get it right from the get-go? Probably not. We’ll find ways to iterate our way into it, but it’s already yielding all sorts of new ideas and customer experiences that we think are very compelling. What’ll be interesting is the race between whether you want a completely different experience, almost like a new app, or whether those advances happen quickly enough that you add them into the existing experience. You could end up meaningfully changing the overall customer experience bit by bit over the next few years as you find elements that work. But I do think every one of these experiences has to be different in the next few years if you want to be the successful leader. Ben Horowitz Yeah, that’s such a tricky leadership problem. It reminds me of when I first began my career as an engineer. I got into an argument with somebody about something, and my boss said, “Look, one thing you have to understand, Ben, is everything is up for discussion until the first line of code is written, and then it’s religious.” I guess you experienced that with Alexa. Andy Jassy Right. Ben Horowitz It’s really interesting, because all the knowledge, experience, and things that went into building that original thing—now you’re saying, “Okay, we’re willing to give a lot of that up and go to this new world.” And, by the way, with the customers too, you’re going to have a big set of customers who love the current experience. Andy Jassy Right. Ben Horowitz Yeah. Andy Jassy Every big transformation, at least in my experience—maybe you have some examples that aren’t, but almost every big transformation requires a huge change. The technology, product, customer experience, and business elements have to be compelling enough to make people want to change. But when that’s true, it doesn’t just happen. It’s a huge change-management effort. Moving from on-premises to the cloud was the same type of transition. Lots of infrastructure leaders were really proud of what they’d built, knew it inside and out, and were in the middle of the action. They didn’t want to learn something new. It’s going to be the same thing as we try to reinvent all our customer experiences. We’re all proud of what we’ve built over the last 10 years, and we all have 100 things that could make those experiences better, things we’re excited about and spend all our waking hours working on. But thinking about how to completely start over is intellectually and emotionally hard to do. Ben Horowitz Yeah, no doubt. No doubt. Erik Torenberg We just incubated Hartwick’s Andreessen Academy a few weeks ago to help young people prepare for this transformation we’re talking about. You’ve said that AI makes building easier, but young people still need to learn real expertise. I’m curious: if you were 18 in 2026, what advice might you have for yourself about how to prepare for this transformation we’re entering? Andy Jassy I have a 25-year-old and a 22-year-old, so I guess my 22-year-old was 19 when this was starting to really take off. I learned very quickly that you can’t tell them not to use AI in education. That ship has left the station. I think universities would be wise not to fight it. It’s a tool that people are going to use. At the same time, I think all young people—and I’d say this for us, too, where we have a lot of people in a writing culture who are using AI to write—you can tell when something’s been written by AI. It’s not quite right. The pieces aren’t put together quite right. The advice I would give somebody young, and gave my kids, was, “You can’t run away from it, and you can’t wish it away.” It’s going to be incredibly impactful in your life and in every experience you try to use in your work. It will allow you to invent in ways that just weren’t possible for us before. You need to learn about it, educate yourself about it, use it a lot, and try it. Try doing things that are outside your comfort zone. Neither of my kids was a developer, but they’ve used AI to build applications. Use it as a focal point for how you operate your day and how you use tools at work. But I don’t think it replaces critical reasoning and thinking, and I think this is true for universities, too. If I think about business school as an example, you’d get three cases that you had to analyze every day. You had to do a lot of reading, take notes, figure out what the key issues were, and then you’d get all these exhibits and take notes on those. Then you had to put together what you thought the plan should be for that company. Oftentimes, there were people who just didn’t do the work. They didn’t do the reading, research, or note-taking. All that kind of goes away with AI. AI is going to get everybody to that advanced starting spot much more quickly. But I don’t think AI is going to put the dots and pieces together for most people most of the time—to build a strategy, an approach, a plan, a new product, or a new customer experience. It’s not that it’ll never happen. I still think you’re going to need human judgment, ingenuity, invention, experimentation, and thinking about how all the pieces come together to build amazing experiences and businesses. You want to use the tools, and they’re going to be huge aids, but I still think critical reasoning and thinking are going to be very important in everything we do. ### Amazon's Expanding Ambition Erik Torenberg Yeah. Gearing toward closing here, you guys are in businesses as diverse as satellite connectivity, robotics, and healthcare. The ambition is incredible. It’s amazing to think this business started by selling books on the internet, and here it is now. What threads these different business lines together? How do you think about which business lines are worth committing to for years, and how do you make sure you don’t spread yourself too thin? How do you make sense of the vast range? Andy Jassy I’d say a few things. First, I think we’re living in an era of three of the biggest transformations in our lifetime, and maybe ever: AI, robotics, and space development. They’re all huge, and they’re all going to completely change a number of customer experiences we encounter. We’re in the middle of all three of those big transformations. If you look at what we’re doing in our retail business, how we use robotics, and what we’re doing on the AI side, we’re making it easier for internal and external builders to build successful AI applications and agents, as well as reinventing all our consumer experiences. If you think about Amazon Leo, our low-Earth-orbit satellite constellation that we’re building for the 500 million households around the world that don’t have broadband connectivity, there’s an incredible set of transformations happening. We have a chance to help customers leverage them in a way that makes their lives better. And we do it in a way that helps customers but also leads to long-term, sustainable businesses for ourselves. I think we have a chance to be much more relevant in helping customers across many more areas. I’m not even mentioning what we’re doing in grocery, autonomous driving, ride-hailing, and healthcare. The way we organize—we’ve often had this internal debate over the years, and a lot of external folks have asked the same question: Are we being spread too thin? I think some of that depends on a combination of how you organize yourself and what you’re willing to invest. Because we don’t try to have the same group of people manage multiple businesses, and we organize in a decentralized, single-threaded way, if you have the right leaders and builders on those teams, an experience that’s big enough to move the needle, a differentiated idea, and a space that isn’t being well served, you have the opportunity to pursue a lot of these things simultaneously. We’ve done that over a long period of time. Then it comes down to how much you’re willing to invest. There are some companies, understandably, that say, “We’re going to stick to our knitting. We’re going to stick to the core business that we have, and we’re going to grow that.” That can be a very successful strategy. We’re also very intent on continuing to grow and build our existing businesses. I think we’ve done that pretty successfully, at a good clip. But there are so many other customer experiences that we know customers really want us to help solve, and that we can. They’re going to help our customers and change our business long term, and we’ll keep pursuing those. Erik Torenberg That’s an inspiring note to end on. Ben Horowitz Yeah. Erik Torenberg It’s an endless ambition. Andy, thanks so much for coming on the podcast. Ben Horowitz Thank you, Andy. Andy Jassy Thank you for having me. I appreciate it, guys.