NVIDIA’s Jensen Huang: AI Demand Is Exploding

Jensen, I did not see in the report or hear a single indication that you or any of the A.I. players you deal with are stumbling, let alone taking a breather in this A.I. race. If you had to use one word to describe, not the recent quarter, but the rest of the year coming up, what word would you use? Demand accelerating. Not only is demand already strong, it's actually accelerating. It's really quite extraordinary. And the reason for that is this, Liz, it's great to see you. The first thing is that A.I. has become useful. It's doing productive work. And it's doing productive tasks. And the tokens that these A.I. labs are generating are profitable tokens. And when you're generating something profitable and the demand is so extraordinary, you want to generate more of it. So the thing that's limiting their growth today is not customers, is not their technology. Customers are, the demand is incredible. What's holding it back now is just compute. We need to bring more compute online so that we can help these A.I. labs grow. The second thing, and this is the part that people don't see, A.I. is far, is way beyond more than just a few top A.I. labs. They're doing incredible work there. But A.I. is literally everywhere. It's being used in enterprises, it's being used all around the world, Neo and Neo clouds and sovereign A.I. clouds. The other half of our business that most people don't see, that is also growing incredibly. And so we're just seeing growth everywhere. And next year, we're launching a new product called Vera Rubin, and the success of it is incredible. This is going to be the fastest product launch we've ever had in history. It was incredible last time with Blackwell, but Rubin's going to be even faster. So it's going to be a huge year next year. You had predicted Vera Rubin would be massive. Can you say that every single Vera Rubin chip that comes out of the fabrication plant is spoken for? And Jensen, how far out do those orders go? Customers have to plan with us at least a year out these days, because these are not, you're not buying a computer, you're building a factory. And these factories are 50, $60 billion in investment. And so you're planning it pretty far down the road. You know, we have good, we have complete visibility into our demand through all of next year and beyond. But for next year, every single chip that we can make has already been sold. And so we have, we're predicting at least 70% growth next year, and that's supply constraint. Our demand is really quite extraordinary. And speaking of looking a year out, you had said that next year, I believe, revenue opportunity could receive, could, I guess, exceed $1 trillion. That would be double the $500 billion that you were expecting this year total. So I'm just wondering, I mean, with such massive demand that you've just articulated, is that $1 trillion number already old news? Is it on the move? Could it be higher? Well, we're forecasting 70% growth year over year. And so not quite a trillion, but our demand is greater than 70%. It's just that we only have supply chain support for 70%. I'm working with all the suppliers. We have the largest supply chain in the world, and we have great partners, but the demand is just too strong. And so I need to make sure that I continuously work on the supply. Otherwise, our customers are going to be disappointed. Their demand is higher than the supply that we currently have. And so what do you do with Taiwan Semi, your big manufacturer? Do you pick up the phone and start saying, guys, I need faster, more? Well, it's TSMC. It's packaging, co-auth packaging, it's wafers, it's packaging, it's memory. It's everything. It's networking, it's transceivers, it's silicon photonics. It's power generation in the data centers. It's availability of high quality land that's powered and ready to receive these computers. You've got bottlenecks literally everywhere, because this is the largest infrastructure build out in human history. We're creating jobs in chip plants, packaging plants, computer plants, and of course, all of these AI data centers. And so we're creating jobs everywhere. And those are also a challenge, because there's just not enough skilled labor around the world and surely not here in the United States to be able to support my entire demand. But we have enough support for the 70 percent that I forecasted. Bottlenecks. I'm glad you mentioned that, because the memory chip bottleneck is certainly an issue. One of the big questions that have been hanging out there was actually answered on the conference call by your CFO, Colette Kress. She said that come the first quarter of next year, the price of your chips will go up. The rumor has been that it would be up about 15 percent. What are you telling customers it will be? Is that around the correct number? Well, this is one of the concerns that, in fact, most investors have had. They've seen that memory prices have gone way up. And so this quarter, we decided that we would rip the Band-Aid off, reset expectations about our gross margins, and just explain to people that, in fact, we have now absorbed the cost increases. We have also repriced our products in the marketplace. And based on that, our margins are going to come down from 75 percent, but it will be between 72 and 73 percent next year. And so we've now reset the expectations and, you know, removed this anxiety that everybody had about our gross margins that's kind of been, you know, a bit of a cloud on our stock for some time. And I'm just happy that we're able to let everybody know that we now have everything worked out and we rolled out new pricing. We've accepted all of our cost increases and everything is, you know, from this point forward, let's go execute. Yeah. Well, I ask because the 15 percent would make a single Vera Rubin rack cost around $8 million, which I know there's a lot of money being thrown around here, but certainly a question. Does it frustrate you when you talk about bottlenecks? There's that. There is the grid buildout, which, of course, is highly dependent on things like permitting, that it's almost as if your own roadmap of growth now is helmed in by something that you don't even control when it comes to pushing out chips. It's somebody else's capacity. Well, we announced this quarter, we reported 100 percent growth year over year, and that will supply constraint. Next year, about this time, will be 70 percent, of course, and that will be supply constraint. And of course, it's also 70 percent off of a much, much larger base. I wish we could do more. We're going to keep on working hard to do more. But some constraints in the system is not a bad thing, and it causes the best technology to emerge and rise. It causes all of the best options to emerge and rise. So I think it makes everybody sharper and makes all of the market sharper. AWS, Amazon's cloud business, and they will buy an additional two million GPUs from you guys, and they'll be deployed 2027, 2028. Is Amazon your best friend now? That's a big number. Half of our business comes from the hyperscalers. And so there's like six hyperscalers sometimes, and they change, you know, who is the largest changes from quarter to quarter, but half of our business comes from six hyperscalers. Now, the businesses that we sell to them, some of it is used internally, but a lot of it is rented externally. And when they rent it, they rent it to essentially NVIDIA developers all over the world. That's one of the great things about our architecture. We have so many end customers that our platform, our technology is the most rentable of any that they have. It captures the highest pricing of anything that they have. And so that's one of the reasons why we're so desirable. We're productive. The infrastructure is productive, it's fungible, it's durable, it has long life, and it's really super rentable. And that's half of our business. The other half of our business has a large number of customers from industrials and enterprises and sovereigns and neoclouds, and so they're all over the place. But half of our business is concentrated within six partners, and they're partly customers of ours, end customers of ours, but they're also partly, if you will, partners of ours as we rent our platform to developers all over the world. You're wearing a lot of hats these days. I mean, your original DNA has been that of a chip creator, but that role has truly expanded. It feels like, Jensen, over the past couple of years, it's almost like you guys have become somewhat of a, in a different way, a bank, and I mean, I'm thinking of the Wall Street Journal article with the headline, NVIDIA HAS BECOME A BANKER TO THE AI BOOM, PUTTING IT ON DANGEROUS GROUNDS. I mean, you've long invested in startups, but this is different. You've become this direct investor in a whole bunch of, whether they're hyperscalers or neoclouds, and that is interesting and maybe a bit worrisome to some people. What does the market have wrong when they worry about that? There's two parts to that. There's one part of it, which is we're investing in the ecosystem. We're just like a venture capitalist in this way, except these, when I invest, when we invest in these companies, these companies use NVIDIA technologies. We only invest in companies that are extremely high-quality team, great business opportunity, important technology, and they use NVIDIA's platform. And so when we're investing in these companies, we are highly informed investors, and the results of our investment shows. I was one of the, we were early investors, and recently, with SpaceX going public, you know, we have some $20, $30 billion return on our investment, and we're delighted by that. And there are many examples of that. We're highly informed investors, but we're investing in our ecosystem for developers that in the future will use NVIDIA's technology. The second part of it is what people would call backstopping or guaranteeing. And this is really centralized around a couple, a few of the AI labs, and really primarily most of it is going to be around a couple of AI labs. And the reason for that is because obviously these are extraordinary companies going through an extraordinary time. Their businesses are turning a corner. They're inflecting and becoming quite profitable. And what they need is compute. But they don't have the balance sheet to go secure the loans. They don't have the, they're not investment-grade yet. They're neither public yet. And so they don't have the ability to secure low-cost capital. And therefore, because we understand exactly what's going on with their business, because we know compute is vital to their growth, and also because, you know, these companies are on the way of becoming public companies, this is really a perfect time for us to help them with their flywheel. And there we backstop them. My sense is that this is going to be a very, very good investment, and it's going to be a fantastic support to two of the most consequential companies in technology history. I just want to get your reaction to Bill Gates' missive that he just put out yesterday, where he said this is going to be a problem for much of America, and the job losses are important enough that he suggested to tax tokens. What do you make of that? What do you think of that? Well, I see things very differently than he does, and there are several things. Of course, there are jobs that will be changed. Everybody's jobs will be changed because of AI. My job has changed. And some jobs will be eliminated. Many new jobs will be created. And if you go back and look at history, when technology, you know, revolutionary technology comes along, incredible advances comes along. And so I believe, again, that this will be a net job creator. However, there are going to be many jobs that will be disrupted. And so we have to be sensible about that. We have to be sensitive about that and be supportive of that. However, this very important point, we are going to be more productive. When companies are more productive, they don't lay off people. They hire more people. And the reason for that is because companies have ambitions. Most companies do. And I would say the vast majority of the world's companies do. They have ambitions for growth. When they're more productive, when they're more profitable, allows us to invest more and go after more growth. Nonetheless, overall, this is going to be a net job creator at a scale that we've never seen. And now coming back to even long before the AI gets deployed, don't forget chip plants are being created, packaging, computer plants, all of the AI factories being created, hundreds of thousands of jobs are being created as we speak. The shape of our economy will be very different in this new era. We have lots and lots of white-collar workers, but we're also going to have a lot of skilled labor. And having a large population of skilled labor and people who build things and make things with their hands is tremendous for the United States. We want to reindustrialize the United States. We want to create more jobs. And all of that's going to happen, you know, right now as we speak with AI. So before we go, is that a thumbs up or thumbs down on Gates's tax the tokens idea? Well, we'll see. First of all, I'm in favor of taxes. And I think that's for anybody who is productive, it's a great way for us to contribute back to the society and the economy. But the fact of the matter is, there are probably lots of different ways to approach this. I don't see what he sees. I love the heck out of Bill. But I don't I don't see what he sees. I see something very, very different. And so, you know, my remedies will be a little different. Jensen, very, very fascinating quarter of growth, and we look forward to what happens in the current quarter and beyond. Thanks so much for joining us. Let's break down what he said and what it actually means. The first thing worth understanding is what Jensen said about demand. He was directly asked if you had to use one word to describe demand for the rest of the year, accelerating or already strong. And here's exactly what he said. Demand isn't just accelerating. It's accelerating beyond what NVIDIA can physically supply. You see, that's a critical distinction that changes everything. It's not demand is strong and we're taking all we can get. It's demand is so extraordinarily strong that we're supply constrained by our own manufacturing capacity, our own ability to source components, our own ability to deliver product that fundamentally changes how you should think about NVIDIA's growth trajectory. The company could grow faster than 70 percent if it had the supply. But every single piece of the supply chain, wafers from TSMC, packaging capacity, memory allocation, power systems, even available land with power ready to go has hit a constraint somewhere. That's not a problem. That's actually a good problem to have. The second thing Jensen emphasized is that AI has finally become useful and profitable. And this is crucial to understand. The tokens that AI labs are generating are now profitable tokens. They're making money off AI. That's not theoretical anymore. That's operational reality. When you're making money on something, you want to generate more of it. The only thing stopping you is compute capacity, not lack of customers. There's infinite demand, not broken technology. It works. Just compute. You see, that's why NVIDIA is in such a powerful position. They're the essential bottleneck for profitability across the entire AI industry. If you want to make money on AI right now, you need NVIDIA chips. And there's not enough of them. That dynamic is going to last as long as this cycle runs. Now, here's where it gets really important. Jensen talked about who's actually using AI infrastructure. Most people think AI is just about the top AI labs, OpenAI, Anthropic, Google, Meta. But Jensen's saying that's only part of it. You see, AI is literally everywhere now. It's in enterprises. It's in sovereign AI clouds. It's in neoclouds all over the world. That's actually the half of NVIDIA's business that most people don't see. Half their revenue comes from the six major hyperscalers. And the other half comes from enterprises, industrials, and sovereigns building their own AI infrastructure. That's a much broader moat than people realize. Vera Rubin is launching next year. And Jensen said it's going to be the fastest product launch they've ever had. Every single Vera Rubin chip that comes off the fabrication line is already spoken for. Customers have to plan a year in advance because they're not buying computers. They're building factories. We're talking about $50 to $60 billion investments per customer. When you're spending that kind of money, you plan it far out. And Jensen said they have complete visibility into demand through next year and beyond. Every chip they can make next year is already sold. Here's the pricing angle that matters. Jensen confirmed that chip prices are going up about 15% in Q1 next year. That puts a Vera Rubin chip at around $8 million each. Now, that sounds expensive. But remember what Jensen said. Customers are building $50 to $60 billion factories around these chips. The difference between $7 million and $8 million per chip when you're talking about billion-dollar infrastructure decisions is almost rounding error. It matters to the margin line. But it doesn't change the customer decision to buy. The margin story is actually really important here. Gross margins are coming down from 75% to 72% to 73% next year. Why? Because memory prices are up. Because supply chain costs are up everywhere. Because NVIDIA decided to rip the Band-Aid off and deal with the cost increases rather than let investors worry about margin pressure indefinitely. Jensen's saying, we've absorbed the cost increases, we've repriced our products, we've worked it all out. Now, let's move forward. That's a confident signal. If they thought demand was weakening, they wouldn't be raising prices. They're raising prices because they can. The supply chain constraints are hitting everywhere. Jensen mentioned TSMC, packaging, memory, networking, transceivers, silicon photonics, power generation, and skilled labor. This is literally the largest infrastructure build-out in human history. And every single part of it has bottlenecks. But here's what's important. Some constraints in the system is actually healthy. It makes technology sharper. It makes the best options emerge. It creates urgency around solving real problems. Now, let's talk about the hyperscalers. Amazon announced it's going to buy an additional 2 million GPUs from NVIDIA to be deployed in 2027 and 2028. 2 million GPUs. Think about that number. That's not a 2027 problem that Amazon's dealing with. That's a commitment Amazon's making right now for infrastructure it'll be deploying years from now. That means Amazon's already planning, already locking in its capacity, already committing capital for multi-year growth. And Amazon's one of six hyperscalers that comprise half of NVIDIA's business. If all six are doing this kind of multi-year capacity planning, that's billions of dollars in already committed demand. Half of NVIDIA's business comes from these six hyperscalers. And that half is partly used internally, but a lot of it is rented externally to developers all over the world. When customers rent NVIDIA infrastructure from hyperscalers, they charge developers the highest prices of anything they offer. That's because NVIDIA is the most productive, most fungible, most durable infrastructure they have. It's highly rentable. That dynamic keeps the hyperscalers coming back for more. Then there's the banker-to-the-AI boom discussion. Jensen explained that NVIDIA's investing in its ecosystem like a venture capitalist, but with one key difference. They only invest in companies that use NVIDIA technology and that are extremely high-quality teams with great business opportunities. They were early investors in SpaceX and now have a $20-30 billion return on that investment. That's venture capital at scale, and it validates their investment thesis completely. But there's a second part that matters even more strategically. NVIDIA's backstopping a couple of AI labs. These are extraordinary companies going through an extraordinary time. Their businesses are inflecting and becoming profitable. But they don't have the balance sheets to secure low-cost capital because they're not public yet, and they're not investment-grade. NVIDIA understands exactly what's going on with their business. They talk to these companies constantly. Compute is vital to their growth. These companies are on the way to becoming public companies, which means they'll eventually become trillion-dollar revenue machines that need enormous amounts of NVIDIA chips. So NVIDIA's helping them with capital to accelerate their growth and survive until they can go public. That's incredibly smart positioning. They're investing in companies that will become the largest consumers of their chips for decades. That's incredibly smart positioning. They're investing in companies that will become enormous consumers of their chips. On the job creation question, Jensen disagreed with Bill Gates' concern about job losses. You see, Jensen's point is that when companies become more productive, they don't lay people off. They hire more people because they want to grow. There will be disruption, sure. But the net effect is job creation at a scale we've never seen. Chip plants. Packaging plants. Computer plants. AI data centers. Hundreds of thousands of jobs are being created right now. And the U.S. needs skilled labor. NVIDIA's infrastructure build-out is creating exactly the kind of manufacturing base and skilled labor jobs that strengthen an economy. Here's what Jensen said that ties it all together. Every constraint in the system is real, but NVIDIA's got visibility into demand through next year and beyond. Every chip they can make is spoken for. Growth is going to be 70% next year, but that's supply constraint. Demand is actually higher. Prices are going up because they can. Margins are stable. The hyperscalers are locked in. The ecosystem is diversifying. And this is just the beginning of the largest infrastructure build-out in human history. That's the picture Jensen painted. It's not a bubble bursting. It's not demand weakening. It's constraint supply meeting explosive demand, and every single part of that supply chain hitting pressure points. But NVIDIA's walking through this with prices stable, demand locked in, and visibility into at least a full year of growth ahead. Subscribe to stay on top of how this actually plays out.