Jensen Huang — Will Nvidia’s moat persist?

Summary

Jensen Huang sits down with Dwarkesh Patel for a remarkably contentious 100-minute conversation about whether NVIDIA’s moat is real, what its real shape is, and whether US export controls on China are advancing or undermining American technology leadership. Jensen frames NVIDIA’s job as one of transforming “electrons into tokens” — a journey he insists is far from commoditized because every layer of his “five-layer cake” of AI (chips, systems, models, applications, data centers) requires extreme co-design that nobody else is currently doing. He argues the moat is not just $100 billion of locked-up supply at TSMC, HBM, and CoWoS packaging, but the velocity of a stack that ships a new architecture every year, a CUDA ecosystem that took 20 years and hundreds of millions of dollars of losses to build, and a downstream demand position so dominant that suppliers are willing to take huge upstream bets on NVIDIA’s behalf.

The middle of the conversation is a striking dispute over export controls. Dwarkesh presses repeatedly on the case for keeping advanced chips out of China — pointing to potential cyber-attack uplift from frontier models, the gap between H200 and Huawei 910C bandwidth, and the strategic value of America’s compute lead. Jensen responds by repeatedly calling the premise “wrong,” arguing that China already has more than enough compute, that energy abundance compensates for older nodes, that 50% of the world’s AI developers are Chinese, and that conceding the second-largest market in the world will only accelerate Huawei’s ecosystem and force open-source AI to optimize for non-American tech stacks. The exchange is unusually combative — Jensen at one point tells Dwarkesh “you speak in absolutes” and “your loser premise makes no sense to me” — and it ends only when Dwarkesh agrees to move on.

The final stretch covers NVIDIA’s investment philosophy (“as much as needed, as little as possible”), why NVIDIA refuses to become a hyperscaler, why Anthropic ended up on TPUs and Trainium (“100% Anthropic” is the answer to TPU and Trainium growth, Jensen claims), why NVIDIA doesn’t pick winners among foundation labs, and a fascinating glimpse into a market that is just emerging: extremely high-ASP “premium” inference tokens for software-engineering agents that NVIDIA is now segmenting around with the Grok-style fast-inference accelerator folded into CUDA. Jensen closes by noting that even without deep learning, NVIDIA would still be a giant accelerated-computing company — because Moore’s Law’s general-purpose scaling is over, and domain-specific acceleration is the only path forward.

Highlights

”Something has to transform electrons into tokens”

“Well, in the end, something has to transform electrons to tokens. The amount of artistry, engineering, science, invention that goes into making that token valuable, obviously we’re watching it happening in real time.” — Jensen Huang, 0:32

yt-dlp --download-sections "*0:32-1:30" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "electrons-to-tokens.mp4"

”The biggest bottleneck? Plumbers.”

“Yes, you know, by the way, great idea. But that’s a good condition. You want a market, you want an industry where the instantaneous demand is greater than the total supply of the industry.” — Jensen Huang, 9:49

yt-dlp --download-sections "*9:46-10:45" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "plumbers-bottleneck.mp4"

”Without Anthropic, why would there be any TPU growth at all?”

“Anthropic is is a unique instance and not a trend. Without Anthropic, why would there be any TPU growth at all? It’s 100% Anthropic. Without Anthropic, why would there be any Trainium growth at all? It’s 100% Anthropic.” — Jensen Huang, 37:06

yt-dlp --download-sections "*37:06-38:00" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "anthropic-100-percent.mp4"

”Don’t pick winners — 60 graphics companies, NVIDIA was at the top of the list NOT to make it”

“When NVIDIA first started, there were 60 graphics companies, 60 3D graphics companies. We are the only one that survived. NVIDIA would be the top of that list not to make it. NVIDIA’s graphics architecture was precisely wrong. And here we are. So I have enough humility to recognize that, you know, don’t pick winners.” — Jensen Huang, 47:29

yt-dlp --download-sections "*47:29-48:40" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "dont-pick-winners.mp4"

”The day DeepSeek comes out on Huawei first, that is a horrible outcome for our nation”

“DeepSeek is not an inconsequential advance. And the day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.” — Jensen Huang, 1:10:34

yt-dlp --download-sections "*1:10:34-1:11:30" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "deepseek-huawei-first.mp4"

”Architecture matters. Computer science matters. Moore’s Law is dead.”

“I just kept saying it over and over and over again. Moore’s Law is dead. Between Hopper and Blackwell, from the transistors themselves, call it 75%. It was three years apart. 75%. Blackwell is 50 times Hopper. My point is, architecture matters. Computer science matters. Semiconductor physics matters as well, but computer science matters.” — Jensen Huang, 1:32:30

yt-dlp --download-sections "*1:32:18-1:33:30" "https://www.youtube.com/watch?v=Hrbq66XqtCo" --force-keyframes-at-cuts --merge-output-format mp4 -o "moores-law-dead.mp4"

Key Points

• NVIDIA as electron-to-token transformer (0:32) - Jensen frames the company’s mission as making token generation more valuable over time, with a five-layer AI cake (chips, systems, models, apps, data centers) that resists commoditization
• As much as needed, as little as possible (1:58) - The core operating philosophy: do only what nobody else will do, partner with the rest of the ecosystem
• Tool use will explode software companies, not commoditize them (3:18) - Synopsys, Cadence, Excel-like tools should see exponential agent-driven instance growth, not collapse
• NVIDIA’s $250B in upstream commitments aren’t the moat — the velocity is (5:04) - Suppliers make implicit commitments because Jensen personally aligns CEOs across the supply chain on demand forecasts
• Plumbers as the real bottleneck (9:21) - Instantaneous demand is greater than total industry supply; CoWoS is now scaling alongside logic and memory because TSMC was forced to swarm it
• Energy and electrical infrastructure are the long-term constraints, not chips (1:02:44) - Hardware bottlenecks resolve in 2-3 years; energy is the structural limit
• Accelerated computing vs ASICs (30:36) - GPUs are like F1 racers requiring expertise to push to the limit; CPUs are Cadillacs; NVIDIA’s optimization expertise routinely 2-3x’s customer model performance
• NVIDIA’s TCO leadership challenge (33:00) - Jensen invites Trainium and TPUs to demonstrate their cost claims on Inference Max benchmarks; says “nobody wants to show up”
• Anthropic is the only ASIC growth story (37:06) - Without Anthropic, TPU and Trainium growth would be zero; this is “fairly well-known and well-understood”
• Why NVIDIA missed investing in Anthropic (40:31) - Jensen didn’t realize at the time that VCs couldn’t supply $5-10B; Google and AWS could, so they captured the lock-in — “I’m not going to make that same mistake again”
• NVIDIA refuses to be a hyperscaler or financier (44:16) - The company supports CoreWeave, Lambda, and Nebius because they need NVIDIA to exist, not because cloud is a desirable business model
• First-in, first-out chip allocation — not highest bidder (54:15) - “It’s a bad business practice. We set the price, you decide to buy or not. You can count on us.”
• One agent surrounded by thousands of safety agents (1:01:31) - Jensen’s vision of the AI security future requires a vibrant open-source ecosystem
• The China policy debate — the conversation’s centerpiece (1:03:33) - Dwarkesh presses on cyberattack uplift; Jensen pushes back that energy abundance plus 7nm Hopper-equivalent capacity already passes the threshold of concern
• 50% of AI developers are in China (1:19:29) - Conceding that ecosystem will force foreign open-source AI to optimize for non-American tech stacks
• Moore’s Law is dead — 75% transistor improvement Hopper to Blackwell, but 50x system gain (1:32:30) - Architecture and computer science matter more than node shrinks
• Premium-ASP inference tokens are a new market segment (1:37:15) - NVIDIA folding Grok into CUDA and creating a faster-response, lower-throughput Pareto-frontier segment for high-value customers like coding agents
• NVIDIA without deep learning would still be huge (1:39:49) - Accelerated computing is the answer to the end of Moore’s-law-driven general-purpose scaling, regardless of whether AI took off

Mentions

Companies

• TSMC (0:00) - 30-year partnership without a legal contract; NVIDIA’s primary foundry partner across N3, N2, and beyond
• SK Hynix, Micron, Samsung (0:00) - HBM memory suppliers; Micron called out by name for early commitment to LPDDR and HBM roadmaps
• Anthropic (37:06) - The “100% Anthropic” driver of TPU and Trainium growth; ended up on Google and AWS chips because NVIDIA wasn’t able to write the multi-billion-dollar check at the time
• OpenAI (47:00) - NVIDIA invested ~$300M; Jensen committed to a $30B-scale investment given they needed it before IPO
• CoreWeave, Lambda, Nebius, Crusoe (46:03) - Neoclouds NVIDIA helped seed; ~$2.3B CoreWeave backstop reported
• Huawei (1:08:11) - Just had its largest single year in company history; shipping millions of 910C chips; “Huawei’s a networking company”
• SMIC (1:05:45) - China’s leading-edge foundry, stuck at 7nm because of EUV export controls
• Google / AWS / Microsoft Azure / OCI (33:00) - Top-five hyperscaler customers; most NVIDIA capacity in their clouds serves external customers
• AMD (37:31) - OpenAI’s announced AMD deal; Jensen says OpenAI is “vastly NVIDIA”
• Broadcom (38:51) - ASIC margins are ~65% (only marginally below NVIDIA’s 70%)
• Mellanox (1:34:53) - “That’s why NVIDIA bought Mellanox. Networking matters.”
• Lumentum, Coherent (12:00) - Silicon photonics supply-chain partners NVIDIA reshaped
• Synopsys, Cadence (3:00) - EDA tool makers Jensen predicts will see exponential agent-driven instance growth
• Jane Street (1:14:26) - Sponsor; trained backdoors into LLMs and challenged the audience to extract trigger phrases
• Cursor (50:24) - Sponsor; Dwarkesh used Composer 2 to build an AI co-researcher in a weekend

Products & Technologies

• Hopper (1:07:51) - 7nm; “today’s models are largely trained on Hopper”
• Blackwell (1:32:18) - 50x Hopper at the system level despite only 75% transistor improvement
• Vera Rubin / Vera Rubin Ultra / Feynman (55:08) - The annual cadence: Rubin this year, Rubin Ultra next, Feynman after
• CUDA / CUDA-X libraries (45:00) - 20 years of investment “while losing money most of that time”; cuLitho for computational lithography called out by name
• NVLink 7.2 (1:09:09) - The interconnect that lets NVIDIA gang chips into one supercomputer
• CoWoS / HBM2 / HBM (9:49) - Packaging and memory tech that was specialty just two years ago, now mainstream
• TPU / Trillium / Trainium (37:06) - Anthropic’s chips of choice; Jensen welcomes them onto Inference Max
• Inference Max / MLPerf (33:00) - Public benchmarks NVIDIA challenges TPU and Trainium teams to demonstrate their cost claims on
• Grok (1:37:15) - Recently added accelerator being folded into CUDA for the premium-ASP fast-inference token segment
• MoE (Mixture of Experts) (1:09:58) - “A great invention” — example of algorithm advances reducing compute requirements
• DeepSeek (1:10:34) - “Not an inconsequential advance”; the headline scenario Jensen worries about is DeepSeek launching on Huawei first

People

• Sanjay Mehrotra (Micron) (11:01) - Doubled down with Jensen on LPDDR and HBM roadmaps five years ago
• Larry Ellison & Elon Musk (53:23) - “They begged for GPUs. That never happened” — Jensen denies the famous dinner story
• Dario Amodei (1:17:22) - Quoted: “It’s like Boeing bragging that we’re selling North Korea nukes”; Jensen calls the analogy “lunacy”
• Liang Wenfeng — referenced indirectly via DeepSeek; not named directly in this transcript

Surprising Quotes

“Listen. Why are you causing one layer of the AI industry to lose an entire market so that you could benefit another layer of the AI industry? There’s five layers. And every single layer has to succeed.” — Jensen Huang, 1:13:41

“I’m the evidence. You take a model that’s optimized for Nvidia and you try to run it on something else. And they don’t run better.” — Jensen Huang, 1:11:39

“You’re not talking to somebody who woke up a loser. And that loser attitude, that loser premise makes no sense to me.” — Jensen Huang, 1:20:18

“Comparing AI to anything that you just mentioned is lunacy. Because it’s a lousy analogy. It’s an illogical analogy.” — Jensen Huang, 1:17:37

“Even if AI doesn’t exist today, NVIDIA will be very, very large.” — Jensen Huang, 1:39:49

Transcript

Dwarkesh Patel: 0:00 We’ve seen the valuations of a bunch of software companies crash because people are expecting AI to commoditize software. And there’s a potentially naive way of thinking about things, which is like, look, NVIDIA sends a GDS II file to TSMC, TSMC builds logic dies, it builds the switches, then it packages them with the HBM that SK Hynix and Micron and Samsung make, then it sends it to an ODM in Taiwan where they assemble the racks. And so NVIDIA is fundamentally making software that other people are manufacturing. And if software gets commoditized, does NVIDIA get commoditized?

Jensen Huang: 0:32 Well, in the end, something has to transform electrons to tokens. That transformation, there’s no… the transformation of electrons to tokens and making those tokens more valuable over time, I don’t… I think that that’s hard to completely commoditize. The transformation from electrons to tokens is such an incredible journey. And making that token… you know, it’s like making one molecule more valuable than another molecule, making one token more valuable than another. The amount of artistry, engineering, science, invention that goes into making that token valuable, obviously we’re watching it happening in real time. And so the transformation, the manufacturing, all of this science that goes in there is far from deeply understood and it’s far from the journey is far from over. And so I doubt that it will happen. We’re going to make it more efficient, of course. I mean, the whole thing about NVIDIA, in fact, the way that you framed the question is my mental model of our company. The input is electrons, the output is tokens. That is in the middle, NVIDIA. And our job is to do as much as necessary, as little as possible to enable that transformation to be done at incredible capabilities. And what I mean by as little as possible, whatever I don’t need to do, I partner with somebody and I make it part of my ecosystem to do. And if you look at NVIDIA today, we probably have the largest ecosystem of partners both in supply chain upstream, supply chain downstream, all of the computer companies and all the application developers and all the model makers and all the… you know, AI’s a five-layer cake if you will, and we have ecosystems across the entire five layers. And so we try to do as little as possible, but the part that we have to do, as it turns out, is insanely hard. And I don’t think that that gets commoditized. In fact, I also don’t think that the enterprise software companies, the tools makers… you know, most of the software companies today are tools makers… Some of them are not. Um, but, uh, some of them are workflow codification, you know, systems. Um, but for a lot of companies, they’re tool makers. For example, you know, Excel’s a tool, PowerPoint’s a tool, uh, Cadence makes tools, Synopsys makes tools. I-I actually see the opposite of what people see. I think the number of agents are going to grow exponentially, the number of tool users are going to grow exponentially, and it’s very likely that the number of instances of all these tools are going to skyrocket. It is very likely the number of instances of Synopsys design compiler is going to skyrocket and the number of-number of agents that are going to be using the floor planners and all of our layout tools and our design-design rule checkers, the number of agents that- today were limited by the number of engineers, tomorrow those engineers are going to be supported by a bunch of agents and we’re going to be exploring the design space like you’ve never seen explored before and we’re going to use the tools that we use today. And so, so I think- I think tool use is going to cause- cause the software companies to skyrocket. The reason why it hasn’t happened yet is because the agents aren’t good enough at using their tools yet. And so either these companies are going to build the agents themselves or agents are going to get good enough to be able to use those tools. And I think it’s going to be a combination of both.

Dwarkesh Patel: 4:31 Mm. I think in your latest filings, you had almost $100 billion in purchase commitments with people, foundry, memory, packaging. And then SemiAnalysis has reported that you will have $250 billion of these kinds of purchase commitments. And so one interpretation is NVIDIA’s moat is really that you’ve locked up many years of these scarce components that are- you know, somebody else might have an accelerator, but can they actually get the memory to build it? Can they actually get the logic to build it? And this is really NVIDIA’s big moat for the next three years.

Jensen Huang: 5:04 Well, it’s- it’s one- it’s one of the things that we can do that is hard for someone else to do. The reason why we could- we’ve made enormous commitments upstream. Um, some of it is explicit, these commitments that you mentioned. Some of it is implicit. Um, for example, a lot of the investments that are upstream are made by our s- supply chain because I said to the CEOs, let me tell you how big this industry’s going to be and let me explain to you why and let me reason through with you and let me show you what I see. And so as a result of that- that process of- of informing, inspiring, um, aligning with CEOs of all different industries upstream, they’re willing to make the investments. Now why are they willing to make the investments for me and not someone else? And the reason for that is because they know that I have the capacity to buy it, buy their supply and sell it through my downstream. The fact that NVIDIA’s downstream supply chain and our downstream demand is so large, they’re willing to make the investment upstream. And so if you look at GTC and and to, you know, people are marveled by the scale of GTC and the people that go. It’s a 360 degrees, the entire universe of AI all in one place and the- they’re all in one place because they need to see each other. I bring them together so that that downstream could see the upstream, the upstream could see the downstream, and all of them could see all the advances in AI and very importantly, they can all meet the AI natives and all the AI startups that are all, you know, being being built and all the amazing things that are happening so that they could see firsthand all the things that I tell them. And so I spent a lot of my time informing directly or indirectly our supply chain and our partners and our ecosystem about the opportunity that’s in front of us.

Dwarkesh Patel: 7:04 Now most of your keynotes…

Jensen Huang: 7:06 Yeah, some people always say, you know Jensen, in most keynotes it’s like one announcement after another announcement after another announcement after another announcement. Our keynotes are there’s always a part of it that’s a little torturous in the sense that it almost comes across like an education. And in fact, that’s exactly on my mind. I need to make sure that the entire supply chain upstream and downstream, the ecosystem understands what is coming at us, why it’s coming, when it’s coming, how big is it going to be, and be able to reason about it systematically, just like I reason about it. And and so I think the the moat as you describe it, we’re able to of course build for a future, if our next next several years is a trillion dollars in in scale, we have to supply chain to do it. Without our reach, the velocity of our business, you know, just as there’s cash flow, there’s supply chain flow, there’s turns. Nobody’s going to build a supply chain for an architecture if the architecture the business turns is low. And so our ability to sustain this scale is only because our downstream demand is so great and they see it and they all hear about it, they see it all coming and so that’s- allows us to do the things that we’re able to do at a scale we’re able to do.

Dwarkesh Patel: 8:41 I do want to understand more concretely whether the upstream can keep up. For many years now, you guys have been 2Xing revenue year over year, you guys have been more than tripling the amount of FLOPS you’re providing to the world year over year.

Jensen Huang: 8:59 Exactly.

Dwarkesh Patel: 9:00 And 2Xing at the scale now is really incredible. It’s going to be 86% next year, according to SemiAnalysis. How do you 2x if you’re the majority? And how do you do that year over year? So are we in a regime now where the growth rate in AI compute has to slow because of upstream? Do you see a way to get around these, you know, how do you build 2x more fabs year over year ultimately?

Jensen Huang: 9:02 Yeah.

Dwarkesh Patel: 9:03 Exactly. So then you look at logic say, you’re the biggest customer on TSMC’s N3 node, and you’re one of the biggest on N2, AI as a whole this year is going to be 60% of N…

Jensen Huang: 9:21 Yeah, at some level, the instantaneous demand is greater than the supply upstream and downstream in the world. And it could be at any instant we could be limited by the number of plumbers, which actually happens.

Dwarkesh Patel: 9:46 The plumbers are invited to next year’s GTC?

Jensen Huang: 9:49 Yes, you know, by the way, great idea. But that’s a good condition. You want a market, you want an industry where the instantaneous demand is greater than the total supply of the industry. The opposite is obviously less good. If we’re too far apart, if one particular item, one particular component is too far away, obviously the industry swarms it. So for example, notice people aren’t talking very much about CoWoS anymore. And the reason for that is because for two years we swarmed the living daylights out of it. And we double-double-doubled on several doubles, and now I think we’re in a fairly good shape. And TSMC now knows that CoWoS supply has to keep up with the rest of the logic demand and the memory demand. And so they’re scaling CoWoS and they’re scaling future packaging technologies at the same level as they scale logic, which is terrific because for a long time CoWoS was rather specialty and HBM memory was rather specialty, but they’re not specialties anymore. People now realize they’re mainstream computing technology. And then of course, we’re now much more able to influence a larger scope of our supply chain. In the past, in the beginning of the AI revolution, all the things that I say now, I was saying five years ago. And some people believed in it and invested in it. For example, Sanjay and the Micron team, I still remember the meeting really well where I was clear about exactly what’s going to happen and why it’s going to happen and the predictions of today. And they really doubled down on it and we partnered with them across LPDDR, across HBM memories, they really invested in it. And it obviously has been tremendous for the company. Some people came a little bit later, but now they’re all here. And so I think each one of these bottlenecks gets a great deal… And now we’re prefetching the bottlenecks years in advance. So for example, the investments that we’ve done with Lumentum and Coherent and all of the Silicon Photonics ecosystem the last several years, we really reshaped the ecosystem and the supply chain of Silicon Photonics. We built up an entire supply chain around TSMC. We partnered with them on COOP, invented a whole bunch of technology, we licensed those patents to the supply chain, keep it nice and open. And so we’re preparing the supply chain through invention of new technologies, new workflows, new testing equipment, double-sided probing, investing in companies, helping them scale up their capacity. And so you could see that we’re trying to shape the ecosystem so that it’s ready—the supply chain—so that it’s ready to support the scale.

Dwarkesh Patel: 12:57 It seems like some bottlenecks are easier than others. And so, scaling up CoWoS versus scaling up—

Jensen Huang: 13:03 I went to the hardest one, by the way.

Dwarkesh Patel: 13:05 Which is?

Jensen Huang: 13:06 Plumbers.

Dwarkesh Patel: 13:07 That’s true.

Jensen Huang: 13:08 Yeah. Yeah, I actually went to the hardest one. Yeah, plumbers and electricians. And the reason for that is because—and this is one of the concerns that I have about all the doomers describing the end of work and killing of jobs. You know, one of the things that if we discourage people from being software engineers, we’re going to run out of software engineers. And the same prediction 10 years ago, some of the doomers were saying—were telling people, ‘Whatever you do, don’t be a radiologist.’ And you might hear some of those videos are still on the web. ‘Radiology is going to be the first career to go. Nobody’s… the world’s not going to need any more radiologists.’ Guess what we’re short of? Radiologists.

Dwarkesh Patel: 13:54 Oh, but okay, so going back to this point about, well, some things you scale, other things, like how do you actually manufacture 2x the amount of logic a year? Ultimately that’s bottlenecked by—memory and logic are bottlenecked by EUV. How do you get to 2x as many EUV machines a year, year-over-year?

Jensen Huang: 14:11 None of that’s impossible to scale quickly. You just need… you could do all of that is easy to do within two or three years. You just need a demand signal. It’s not about… once you can build one, you can build 10, and once you can build 10, you can build a million. And so these things are not hard to replicate.

Dwarkesh Patel: 14:31 How far down the supply chain do you go? Do you go to ASML and say, ‘Hey, if I look out three years from now, for Nvidia to be generating two trillion in a year in revenue, we need way more EUV machines’?

Jensen Huang: 14:41 Some of them I have to directly, some of them indirectly. And some of them, if I can convince TSMC, ASML will be convinced. And so that’s, you know, we have to think about the critical pinch points and—but if TSMC is convinced, you’ll have plenty of— EV machines in a few years. And so none of that, my point is that none of the bottlenecks last longer than a couple two-three years. None of them. And meanwhile, meanwhile we’re improving computing efficiency by 10x, 20x, in the case of Hopper to Blackwell, some 30-50x. We’re coming up with new algorithms because CUDA’s so flexible. We’re developing all kinds of new techniques so that we drive efficiency in addition to increasing capacity. And so, those are things that none of that worry me. It’s the stuff that’s downstream from us, energy policies that prevent energy from, you know, you can’t grow, you can’t create an industry without energy. You can’t create a whole new manufacturing industry without energy. We want to re-industrialize the United States. We want to bring back chip manufacturing and computer manufacturing and packaging, and we want to build new things like EVs and robots and we want to build AI factories. You can’t build any of these things without energy. And those things take a long time. But more chip capacity, that’s a two-three year problem. More CoWoS capacity, two-three year problem.

Dwarkesh Patel: 16:14 Interesting. I feel like I have guests tell me the exact opposite thing sometimes, and I don’t in this case I just don’t have the technical knowledge to adjudicate. But—

Jensen Huang: 16:23 Well the beautiful thing is you’re talking to the expert.

Dwarkesh Patel: 16:26 Yeah. True, true. Okay, I want to ask about your competitors. So if you look at TPU, arguably two out of the top three models in the world, Claude and Gemini, were trained on TPU. What does that mean for NVIDIA going forward?

Jensen Huang: 16:42 Well, we have a very different, we build a very different thing. You know, what NVIDIA built is accelerated computing, not a tensor processing unit. And accelerated computing is used for all kinds of things. You know, molecular dynamics and quantum chromodynamics. It’s used for data processing, data frames, structured data, unstructured data. It’s used for fluid dynamics, particle physics, you know, in addition, we use it for AI. And so, accelerated computing is much more diverse and, although AI is the conversation today and is obviously very important and impactful, computing is much broader than that. And what NVIDIA has done is reinvented the way computing is done from general-purpose computing to accelerated computing. Our market reach is far greater than any TPU, any ASIC can possibly have. And so if you look at our position, we’re the only company that accelerates applications of all kinds. We have a gigantic ecosystem. And so all kinds of frameworks and algorithms all run on NVIDIA. …and because our computers are designed to be operated by other people, anyone who’s an operator could buy our systems. Most of these home-built systems, you have to be your own operator because it was never designed to be flexible enough for other people to operate. And so as a result of the fact that anybody can operate our systems, we’re in every cloud including Google Home and Amazon and, you know, Azure and OCI and, right? And so whether you want to operate it to rent or operate it if you want operate it to rent you better have large ecosystem of customers in many industries that be the offtakers. If you’re operating it, if you if you want to operate it for yourself, um we can, you know, we obviously have the ability to help you operate it yourself like for example for Elon with xAI. And because we could we could enable operators in any company in any industry, you could use it to build a supercomputer for scientific research and drug discovery at Lilly. And so we can help them operate their own supercomputer and use it for the entire diversity of drug discovery and biological sciences um that that we accelerate. And so there’re just, you know, a whole bunch of applications that we can address that you can’t do so with TPUs because NVIDIA’s built CUDA as a fantastic Tensor Processing Unit as well, but it does, you know, it does every life cycle of data processing and computing and AI and so on and so forth. And so I our market opportunity is just a lot larger. Our reach is a lot greater and because we have such a large um we basically support every application in the world now, you could build NVIDIA systems anywhere and know that there’ll be customers for it. And so it’s a very different thing.

Dwarkesh Patel: 19:30 Um, this is going to be sort of a long question but, you know, you have spectacular revenue. Um, and this revenue is mostly, you’re not making 60 billion a quarter from pharma and quantum, you’re making it because AI is unprecedented technology that is growing unprecedentedly fast. And so then the question is what is best for AI specifically and I’m not in the details but I talk to my AI researcher friends and they say, look, when I use a TPU it’s this big systolic array that’s perfect for doing matrix multiplies whereas a GPU is very flexible, it’s great when you have lots of branching, when you have irregular memory access. But these, you know, what is AI? Just like these very predictable matrix multiplies again and again and again and you don’t have to give up any die area for warp schedulers, for, you know, switches between threads and memory banks. And so the TPU’s really optimized for the majority, the bulk of this growth in revenue and use case for compute that is coming online right now. Um yeah, I wonder how you react to that.

Jensen Huang: 21:00 Matrix multiplies is an important part of AI, but is not the only part of AI. And if you want to come up with a new attention mechanism, or if you want to disaggregate in a different way, if you want to come up with a whole new type of architecture altogether, for example, you know, a hybrid SSM. Uh, if you want to use a, if you want to create a model that, that, uh, fuses diffusion and auto regressive somehow. Uh, you want an architecture that’s just generally programmable. And, and we run everything you can imagine. And so that’s the advantage. It allows for invention of new algorithms a lot more, a lot, a lot more easily. And so because it’s a programmable system. And, and the ability to invent new algorithms is really what makes AI advance so quickly. You know, TPUs like anything else is impacted by Moore’s Law. And we know that Moore’s Law is increasing about 25% per year. And so the only way to really get 10x leaps, 100x leaps, is to fundamentally change the algorithm and how it’s computed every single year. And that’s NVIDIA’s fundamental advantage.

Dwarkesh Patel: 22:01 Mhmm.

Jensen Huang: 22:02 The only reason why we were able to make Blackwell to Hopper 50 times… You know, I said it was 35 times and and when I first announced it was going to Blackwell is going to be 35 times more energy efficient than Hopper, nobody believed it. And and then Dylan wrote an article, he said he said in fact in fact I sandbagged, it’s actually 50 times. And you can’t reasonably do that with just Moore’s law. And so the way that we solve that problem is new models, MOEs, uh, parallelized and disaggregated and distributed, uh, across a computing system.

Dwarkesh Patel: 22:54 Yeah.

Jensen Huang: 22:56 Uh, and without the ability to really get down and come up with new kernels with CUDA, it’s really hard to do.

Dwarkesh Patel: 23:07 Mhmm.

Jensen Huang: 23:09 And and so the combination of the programmability of our of our architecture, uh, the fact that NVIDIA’s an extreme co-design company where we could even offload some of the computation into the fabric itself, NVLink for example, into the network Spectrum-X, uh, and that we could affect change across the processors, the system, the fabric, the library, the algorithm… All of that was done simultaneously. Without CUDA to do that, I wouldn’t even know where to start.

Dwarkesh Patel: 23:53 My sponsor Crusoe was among the first clouds to offer NVIDIA’s Blackwell and Blackwell Ultra platforms. And they just announced their NVIDIA base… Ruben deployment scheduled for later this year. But access to state-of-the-art hardware is only part of the story. For example, most inferences already do KV caching for a single user’s forward passes. But Crusoe does it across users and GPUs. So if a thousand agents are running on the same system prompt, Crusoe only has to compute the KV cache once for it to become available to every single GPU in the cluster. This is especially important as systems get more agentic and require much longer prefixes in order to use tools and access files. In a recent benchmark, Crusoe was able to deliver up to 10 times faster time-to-first-token and up to 5 times better throughput than vLLM. This is just one among many reasons that you should run your inference workload with Crusoe. And if you need GPUs for training, you don’t need to switch clouds. Crusoe’s got you covered there too. Go to crusoe.ai/thor-cash to learn more. So this gets at an interesting question about Nvidia’s clientele, where if 60% of your revenue is coming from these big five hyperscalers, you know, in a different era with different customers, let’s say it’s professors who are running experiments and they are helped a bunch by they need CUDA, they can’t use another accelerator. They need to just run PyTorch with CUDA and have everything optimized. But if you’ve got these hyperscalers, they have the resources to write their own kernels. In fact, they have to to get that extra last 5% that they need for their specific architecture. Anthropic, Google are mostly running their own accelerators or running TPUs and Trainium. But even OpenAI, using GPUs, has Triton, which they’re like, ‘We need our own kernels.’ So they’ve down to CUDA C++. They’ve instead of using cuBLAS and NCCL and everything, they’ve got their own stack which compiles to other accelerators as well. And so if most of your customers can and do make replacements for CUDA, to what extent is CUDA really the thing that is going to make frontier AI happen on Nvidia?

Jensen Huang: 25:58 CUDA, CUDA is a rich ecosystem. And so if you want to build on any computer first, building on CUDA first is incredibly smart. And because the ecosystem is so rich, we support every framework. If you want to create custom kernels, if you need, for example, we contribute enormously to Triton. And so the backend of Triton, huge amounts of Nvidia technology. We’re delighted to help every framework become as great as it can be. And there’s lots and lots of frameworks. There’s Triton, there’s vLLM, there’s SG-Lang, and there’s more, right? And now there’s a whole bunch of new reinforcement learning frameworks coming out. You’ve got Ver-all, you’ve got Nemo RL, you’ve got a whole bunch of new and then the now with with post-training and reinforcement learning, that entire area is just exploding, right? And so if you want to build on an architecture, building on CUDA makes the most sense because you know that the ecosystem is great. You know that if something happens, it’s more likely in your code and not in the mountain of code underneath. You know, don’t forget the amount of code that you’re dealing with when you’re building these systems. When something doesn’t work, was it you or was it the computer? You would like it always to be you and to to be able to trust the computer and and you know obviously we still have lots and lots and lots and lots of bugs ourselves but but our system is so well rung out that you could at least build on top of the foundation. So that’s number one is the richness of the ecosystem, the programmability of it, the capability of it. The second thing is is um if you were a developer and you were building anything at all, the single most important thing you want more than anything is install base. You want the software that you run to run on a whole bunch of other computers. You don’t want to build a software you’re not building software just for yourself, you’re building software for your fleet or for everybody else’s fleet because you’re a framework builder. And NVIDIA’s CUDA ecosystem is ultimately its great treasure. We are now I don’t know how many several hundred million GPUs. Every cloud has it. Goes back to A10, A100, H100, H200, you know, the L series, the P series, I mean there’s a whole bunch of them and and they’re they’re they’re in all kinds of sizes and shapes and if you’re a robotics company you want that CUDA stack to actually run in the CUDA in the robot itself. We’re literally everywhere. And so the install base says that once you develop the software, once you develop the model, it’s going to be useful everywhere. And so the install base is just too incredibly valuable. And then lastly, the fact that we’re in every single cloud makes us genuinely unique because you know you’re an AI company and you’re an AI developer, you’re not exactly sure which CSP you’re going to partner with and where you would like to run it and we run it everywhere including on-prem for you if you like. And so so I think that that the richness of the ecosystem, the expansiveness of the of the of the install base and the versatility of where where we are, that combination is is um makes CUDA invaluable.

Dwarkesh Patel: 29:16 That makes a lot of sense. I guess the thing I’m curious about is whether those advantages matter a lot to your main customers, like there’s many people who they might matter for, for the kind of person who can’t actually build their own software stack, who make up most of your revenue. And especially if you go to a world where AI is getting especially good at the things which have tight verification loops where you can RL on them and then this question of how do you write a kernel that does attention or MLP the most efficiently across a scale-up? It’s a very verifiable sort of feedback loop and so oh can everybody can all the hyperscalers write these custom kernels for themselves? Um and they might still NVIDIA has It still has great price performance, but they might still prefer to use NVIDIA. But then the question is, does it just become a question of who is offering the best specs, the best flops and memory and memory bandwidth for a given dollar, where historically NVIDIA has just had and still has, you know, the best margins in all of AI across hardware and software, 70% plus, because of this CUDA moat? And the question is, oh, can you sustain those margins if for most of your customers, they can actually afford to build instead of the CUDA moat?

Jensen Huang: 30:33 The number of engineers we have assigned to these AI labs is insane.

Dwarkesh Patel: 30:35 Mm.

Jensen Huang: 30:36 Working with them optimizing their stack. And the reason for that is because because nobody knows our architecture better than we do. And these architectures are not not as general purpose as a CPU. The reason why a CPU is so, you know, a CPU’s kind of like like a Cadillac, you know? It it just always, you know, it’s a nice cruiser. It never goes too fast. Everybody drives it pretty well. You know, it’s got cruise control, you know, and everything is easy. But in a lot of ways, NVIDIA’s GPUs or accelerators are kind of like F1 racers. And yeah, I could imagine everybody’s able to drive it at a hundred miles an hour. But it takes quite a bit of expertise to be able to push it to the limit. And we use we use a ton of AI to create the kernels that we have. And I’m pretty sure we’re going to still be needed for quite some time. And so our expertise helps our our our AI labs partners get another 2X out of their stack easily. Oftentimes, it’s not unusual that we, you know, by the time that we’re done optimizing their stack or optimizing a particular kernel, their model sped up by 3X, 2X, 50%. That’s a huge number, especially when you’re talking about the install base of the fleet that they have, of all the Hoppers and Blackwells that they have. When you increase it by a factor of two, that doubles the revenues. That directly translates to revenues. NVIDIA’s computing stack is the best performance per TCO in the world, bar none. Nobody can demonstrate to me that any single platform in the world today has better performance TCO ratio. Not one company. And in fact, in fact, the the benchmarks are out there. Dylan’s, right? Inference Max is sitting out there for everybody to use. And not one TPU won’t come, Trainium won’t come. I I encourage them to use Inference Max and demonstrate their incredible inference cost. It’s really, really hard. Not nobody wants to show up. MLPerf? I would - I would welcome Trainium to demonstrate their 40% that they claim all the time. I would - I would love to hear them demonstrate the - the cost advantage of TPUs. It makes no sense in my mind. It makes absolutely zero sense. On first principles it makes no sense. And so I - I think the - I think the - the reason why we’re so successful is simply because our TCO is so great. There’s a second - you say, um, 60% of our customers are the top five, but most of that business is external. For example, most of AWS’s… most of NVIDIA in AWS is for external customers, not internal use. Most of our customers at Azure, obviously all of our customers are external, all of our customers at OCI are external, not internal use. The reason why they - they favor us is because our reach is so great. We can bring them all of the great customers in the world. They’re all built on NVIDIA. And the reason why all these companies are built on NVIDIA is because our reach and our versatility is so great. And so - so I think - I think the flywheel is really installed base, the programmability of our architecture, the richness of our ecosystem, and the fact that there’s so many AI companies in the world. There’s tens of thousands of them now.

Dwarkesh Patel: 34:21 Mm.

Jensen Huang: 34:22 And if you were one of those AI startups, what architecture would you - would you choose? You would choose an architecture that’s most abundant - we’re the most abundant in the world. The one has the largest installed base - we’re the most - largest installed base, and one that has a rich ecosystem. And so that’s the flywheel that - that’s the reason why between the combination of one, our perf per dollar is so great… um… that - that - they have the lowest cost tokens. Second, our perf per watt is the highest in the world. And so if - if - um - one of these companies, if - our partners built a 1-gigawatt data center, that 1-gigawatt data center better deliver the maximum amount of revenues that - and number of tokens, which directly translates to revenues. You want it to generate as many tokens as possible, maximize the revenues for that data center. We are the highest tokens per watt architecture in the world. And then lastly, if your goal is to rent the infrastructure, we have the most customers in the world.

Dwarkesh Patel: 35:24 Mm.

Jensen Huang: 35:25 And so that’s the reason why the flywheel works.

Dwarkesh Patel: 35:27 Interesting. I - I guess the - the question comes down to what is the actual market structure here because even if there’s other companies… there could have been a world where there’s tens of thousands of AI companies that have roughly equal share of compute. But if even through these five hyperscalers, really the people on Amazon using the compute are Anthropic, OpenAI, and these big foundation labs who can themselves afford and have the ability to make different accelerators work…

Jensen Huang: 35:56 No, I think your - your assumption is - premises wrong.

Dwarkesh Patel: 35:58 Maybe. Um, but I… Let me let me ask you a slightly different question, which is…

Jensen Huang: 36:02 Come back and make me correct your your um your premise.

Dwarkesh Patel: 36:04 Okay, I’ll let me just ask you a different question, which is… okay, if every everything you’re saying is true…

Jensen Huang: 36:08 But still make sure to make me come back and and fix because it’s just too important to AI. It’s too important to the future of science, it’s too important to the future of the industry. That that premise, the premise, look.

Dwarkesh Patel: 36:22 Let me just finish the question and then you can address it together.

Jensen Huang: 36:24 Okay, yeah.

Dwarkesh Patel: 36:25 So, what do you think if if all these things are true about price performance and performance per watt etc are true, why why do you think it is the case that say Anthropic for example just announced a couple days ago they have a multi-gigawatt deal with Broadcom and Google for TPUs and majority of their compute obviously for Google it’s TPUs and majority of compute, so if I look at these big AI companies, it seems like a lot of their compute, there was some point where it was all Nvidia, and now it’s not. And so I’m curious how to square… if these things are true on paper, why are they going with other accelerators?

Jensen Huang: 37:06 Yeah, Anthropic is is a unique instance um and not a trend. Uh without Anthropic, why would there be any TPU growth at all? It’s 100% Anthropic. Without Anthropic, why would there be any trainium growth at all? It’s 100% Anthropic. And I think that’s fairly well known and well understood. It’s not that it’s not that there’s an abundance of ASIC opportunities, there’s only one Anthropic.

Dwarkesh Patel: 37:31 But OpenAI’s deals with AMD, they’re building their own Titan accelerator.

Jensen Huang: 37:35 Yeah, but they’re mostly, I think we could all acknowledge they’re vastly Nvidia. And and we’re going to still do a lot of work together. Yeah. Yeah. And we’re not we’re not I’m not offended by other people using something else and trying things. If they don’t try these other things, how would they know how good ours is, you know? And sometimes you got to be reminded of it. And and um we got to and we have to continuously earn earn um the position that we’re in. Uh there’s always big claims and look at the number of ASICs that have been canceled. Just because you’re going to build an ASIC, you still have to build something better than Nvidia. And it’s not that easy building something better than Nvidia. It’s not sensible actually. You know, it’s we Nvidia’s got to be missing something seriously, you know, and because our scale, our velocity, we’re the only company in the world that’s cranking it out every single year, big leaps every single year.

Dwarkesh Patel: 38:32 I guess their logic is that hey, it doesn’t need to be better, it just needs to be not more than 70% worse because they’re paying you 70% margins.

Jensen Huang: 38:40 No, no, no, don’t forget. Uh even an ASIC’s margin’s really quite high. Nvidia’s margin’s 70% let’s say, but an ASIC margin’s 65%. What are you really saving?

Dwarkesh Patel: 38:51 Oh you mean from Broadcom or something like that?

Jensen Huang: 38:53 Yeah, sure. You gotta pay somebody.

Dwarkesh Patel: 38:56 Yeah.

Jensen Huang: 38:57 And so so I think the the ASIC margins are are incredibly high. Good. From what I can tell, and they believe it so too. And so they’re quite proud of their incredible ASIC margins. And so you ask the question why? A long time ago, we just didn’t have the ability to do it. And this is… and at the time, I didn’t deeply internalize how difficult it would be to build a foundation AI lab.

Dwarkesh Patel: 39:30 Mmm, like OpenAI and Anthropic.

Jensen Huang: 39:32 And the fact that they needed huge investments from the suppliers themselves. We just weren’t in a position to make the multibillion-dollar investment into Anthropic so that they could use our compute. But Google and AWS were, and they put in huge investments in the beginning so that Anthropic, in return, used their compute. We just weren’t in a position to do so at the time. Nor did I, I would say my mistake is I didn’t deeply internalize that they really had no other options, that a VC would never put in 5, 10 billion dollars of investment into an AI lab with the hopes of it turning out to be Anthropic. And so that was my miss, but even if I understood it, I don’t think we would have been in the position to do that at the time. But I’m not going to make that same mistake again, and I’m delighted to invest in OpenAI and I’m delighted to help them scale and I believe it’s essential to do so. And then when I was able to, when Anthropic came to us, I’m delighted to be an investor, delighted to help them scale, but we just weren’t at the time able to do so. If I could rewind everything, NVIDIA could have been as big back then as we are now, I would have been more than happy to do it.

Dwarkesh Patel: 41:07 This is actually quite interesting, which is for many years NVIDIA has been this… the company in AI, making money, making lots of money. And now you’re investing it. It’s been reported that you’ve done up to 300 million in OpenAI and 100 million in Anthropic. But now their valuations have increased and I’m sure they’ll continue to increase. And so if over these many years, you know, you were giving them the compute, you saw where AI was headed and then they were worth like 1/10 what they are now a couple years ago, or even a year ago in some cases. And you had all this cash. There’s a world where either NVIDIA themselves becomes a foundation lab, does the huge investment to make that possible, or has made the deals you made now at current valuations much earlier on. And you had the cash to do it. So I am curious actually… Why not have done it earlier?

Jensen Huang: 42:02 We did it as soon as we could have. We did it as soon as we could have and and if I could have, I would have done it even earlier. At the time that Anthropic needed us to do it, we just weren’t in a position to do it. It wasn’t, you know, it wasn’t in our sensibility to do so.

Dwarkesh Patel: 42:22 How so? Like a cash thing or just?

Jensen Huang: 42:23 Yeah, the level of investment. You know, we’d never invested outside the company at the time and not that much. And we didn’t realize we needed to. You know, I always I always thought that they could just go raise VCs for God’s sakes, like like all companies do. But but what they were trying what they were trying to do couldn’t have been done through VCs. What OpenAI wanted to do couldn’t have been done through VCs. And I recognize that now. I didn’t know it then. You know, but that’s their genius. That’s why they’re smart. You know, and so so they realized they realized then that they had to do something like that. And I’m delighted that they did. You know, and even though even though we caused Anthropic to have to go to somebody else, I’m still happy that it happened. Anthropic’s existence is great for the world. I’m delighted for it.

Dwarkesh Patel: 43:21 I guess you still are making a ton of money and you’re making way more money quarter after quarter.

Jensen Huang: 43:27 It’s still okay to have regrets.

Dwarkesh Patel: 43:29 So then the question still arises, okay, well now that we’re here and you have all this money that you keep making, what should Nvidia be doing with it? And there’s one answer which says, look, there’s this whole middleman ecosystem that has popped up for converting CapEx into OpEx for these labs so that they can rent compute because the chips are really expensive, they make a lot of money over their lifetime because the AI models are getting better, the value that they generate through tokens is increasing, but they’re expensive to set up. Nvidia has the money to do the CapEx. So, and in fact, you are, it’s been reported you’re backstopping CoreWeave up to 2.3 billion and have invested in Lambda. But yeah, why doesn’t Nvidia become a cloud themselves? Why doesn’t it become a hyperscaler themselves and rent this compute out? You have all this cash to do it.

Jensen Huang: 44:16 This is a philosophy of the company and I think it’s wise. We should do as much as needed, as little as possible. And what that means is the work that we do with building our computing platform. If we don’t, if we don’t do it, I genuinely believe it doesn’t get done. If we didn’t take the risk that we take, if we didn’t build NVLink the way we built, if we didn’t build the whole stack, if we didn’t create the ecosystem the way we did it, if we didn’t dedicate ourselves to 20 years of CUDA while losing money most of that time, if we didn’t do it, nobody else would have done it. If we didn’t create all of the CUDA-X libraries so that they’re all domain-specific, this is several, a decade and a half ago. Then we pushed into domain-specific libraries because we realized that if we didn’t create these domain-specific libraries, whether it’s for ray tracing or image generation or even the early works of AI, these models, if we didn’t create them for data processing, structured data processing or vector data processing, if we didn’t create them, nobody would. And I am completely certain of that. We created a library for computational lithography called cuLitho. If we didn’t create it, nobody would have. And so accelerated computing wouldn’t advanced the way it has if we didn’t do what we did. And and so we should do that. We should dedicate our company, all of our might, wholeheartedly to go do that. However, the world has lots of clouds. If I didn’t do it, somebody’d show up. And so following the recipe, the philosophy of doing as much as needed, but as little as possible, as little as possible, that philosophy exists in our company today. And everything I do, I do it with that lens. In the case of clouds, if we didn’t support CoreWeave to exist, these neoclouds, these AI clouds, wouldn’t exist. If we didn’t help CoreWeave exist, they would not exist. If we didn’t support Lambda, they wouldn’t be where they are today. If we didn’t support Nebius, they wouldn’t be where they are today. Now they are they’re doing fantastically. Is that a business model? No. We should do as much as needed, as little as possible. And so we’re trying to we invest in our ecosystem because I want our ecosystem to thrive and I want our I want the architecture and I want AI to be able to connect with as many industries as possible, as many countries as possible, and make it possible for, you know, the planet to be built on AI and to be built on the American tech stack. And so so those that vision I think is exactly what we’re pursuing. Now one of the things… …that you mentioned, there are so many great amazing foundation model companies and we try to invest in all of them. And this is this is another thing that we do. We don’t pick winners. And we like we we need to support everyone. And it’s part of our part of our our joy of doing so. It’s an imperative to our business, but we also go out of our way not to pick winners. And so when I when I invest in one of them, I invest in all of them.

Dwarkesh Patel: 47:26 Why why do you go out of your way not to pick winners?

Jensen Huang: 47:29 Because it’s not our job to. Number one. Number two, when NVIDIA first started, there were 60 graphics companies, 60 3D graphics companies. We are the only one that survived. If you would have taken those 60 companies, 60 graphics companies, and asked yourself which one was going to make it, NVIDIA would be the top of that list not to make it. You know, this is long before you, but NVIDIA’s graphics architecture was precisely wrong. It’s not a little bit wrong… We created an architecture that was precisely wrong. And and it was an impossible thing for developers to support, it was never going to make it. We reasoned about it for good reason from good first principles, but we ended up in the wrong solution. And and uh everybody would have count everybody would have counted us out. And and here we are. And so I’m I’m I have enough humility to recognize that, you know, don’t don’t pick winners.

Dwarkesh Patel: 48:29 Mm. Yeah.

Jensen Huang: 48:31 Either let them all take care of themselves or take care of all of them.

Dwarkesh Patel: 48:37 Um, one thing I didn’t understand is you said look we’re not prioritizing these new clouds, just because they’re new clouds and we want to prop them up. But you also said you listed a bunch of new clouds and you said they wouldn’t exist if it wasn’t for Nvidia. And so how are those two things compatible?

Jensen Huang: 48:53 Oh, um first of all, they they need to want to exist and they come to ask us for help. And when they when they um when they want to exist and they have a business plan and they, you know, they have expertise and you know they have the passion for it, uh they obviously have to have some capabilities themselves. Uh but if at the end of the day they need some investment in order to get it off the ground, we would be there for them. Um but but the sooner they get their flywheel going, you know, your question was do we want to be in the financing business, the answer is no. Yeah, we don’t want to be we want to because there are people in the financing business and we rather work with all the people who are in the financing business than to be a financier ourselves. And so, so I think the our goal is to focus on what we do, keep our business model as simple as possible, support our ecosystem. When someone like OpenAI needs an investment of thirty billion dollar scale because it’s still before their IPO and and uh we deeply believe in them and we deeply believe that I deeply believe that they’re going to be an well they’re an extraordinary company already today, they’re going to be incredible company, the world needs them to exist, the world wants them to exist, I want them to exist and and they have everything on they have the wind at their back, let’s support them and let them scale. And so to those those investments we’ll do because they’re they need us to do it and uh but we’re not trying to do as much as possible, we’re trying to do as little as possible.

Dwarkesh Patel: 50:24 I spend way too much time copy-pasting text back and forth from Google Docs to chatbots. And so I built what’s basically a cursor for writing, which operates the way I think an AI co-researcher should operate. I can tag it and it can talk with me through inline comment threads and help me dig deeper and brainstorm. I built this entire thing over the weekend with Cursor and their new composer 2 model. With a lot of agentic coding tools, I feel like I have no idea what’s going on under the surface. I just have to relinquish control and hope for the best. But Cursor let me try a bunch of different ideas while staying on top of the implementation. I did most of my brainstorming in the agents window. And after I got some basic files in place, I used the diff window to track changes. The few times that I needed I needed to make a quick tweak by hand, I just used the editor. If you want to try my AI code researcher for yourself, I’ve linked the GitHub repo in the description. And if you have a tool that you’ve been wanting to build, you should make it happen. Go to cursor.com/dwarkesh to get started. This, this may be sort of an obvious question, but we’ve lived many years in this situation where there’s a shortage of GPUs and it’s grown now because models are getting better and—

Jensen Huang: 51:25 We have a shortage of GPUs.

Dwarkesh Patel: 51:26 Yeah.

Jensen Huang: 51:27 Yeah.

Dwarkesh Patel: 51:28 And Nvidia is known for divvying up the scarce allocation, not just based on highest bidder, but rather on, hey, we want to make sure that these neo, neo-clouds exist, let’s give some to CoreWeave, let’s give some to Crusoe, let’s give some to Lambda. Um, why is it good for Nvidia, first of all, would you agree with this characterization of fracturing the market?

Jensen Huang: 51:43 No. No. Yeah, your premise is just wrong. Yeah, um, we’re sufficiently, um, mindful about these things. Uh, we’re very mindful about these things. First of all, if you don’t place an or— if you don’t place a PO, all the talking in the world won’t make a difference. And so until we get a PO, what are we going to do? And so the first thing is, is we work, we work really hard with everybody to get a forecast done because these things take a long time to build and the data centers take a long time to build, and so we align ourselves with demand and supply and things like that through forecasting. That’s job, job number one. Number two, everybody who, you know, we try to forecast with as many people as possible, but in the final, in the final analysis, you still have to place an order. And maybe, maybe, uh, for whatever reason you didn’t place your order, what can I do? And so at some point, first in first out. But beyond that, if you’re not ready because your data center’s not ready, or certain components aren’t ready to, to enable you to stand up a data center, um, we might decide to serve another customer first. That’s just maximizing the throughput of our, of our own factory. And so, um, we might do some adjustments there. Aside from that, the prioritization is, is first in first out.

Dwarkesh Patel: 53:22 Mm-hm. Yeah.

Jensen Huang: 53:23 You got, you gotta place a PO. If you don’t place a PO… Now, of course, there’s stories about that, you know, like for example, the all of this kind of started from, from a, there was an article about Larry and Elon having dinner with me where they, where they begged for GPUs. That never happened. We had, we absolutely had dinner. We absolutely had dinner, um, and it was, it was a wonderful dinner. In no time did they beg for GPUs. And so it’s just, they just had to place an order. And once they place an order, we do our best to get the capacity to them. We’re not complicated.

Dwarkesh Patel: 53:58 Okay, so it sounds like— Like there’s a queue and then um based on whether your data center’s ready and when you place your purchase order, you get ‘em at a certain time. But it still doesn’t sound like highest bidder just gets it. Is there a reason to do it?

Jensen Huang: 54:13 We never do that.

Dwarkesh Patel: 54:14 Okay.

Jensen Huang: 54:15 Why not just do highest bidder? Because it’s it’s a bad business practice. You set your price, you set your price, and then and then people decide to buy it or not. And um there there I understand that that others in the chip industry um change their prices when demand is higher, but we just don’t. We just don’t. That’s just never been a practice of ours. You can count on us. You know, I I prefer to be to be um dependable, uh to be the foundation of the industry and I you don’t need you don’t need to second guess. You know, if you if I quoted you a price, um we quoted you a price. That’s it. And if demand goes through the roof, so be it.

Dwarkesh Patel: 55:03 And on the other end, that’s why you have a productive relationship at TSMC, right?

Jensen Huang: 55:08 Yeah. Yeah, yeah. Nvidia’s been in business, we’ve been doing business with them for uh I guess coming up on 30 years. And Nvidia and TSMC don’t have a legal contract. There’s there’s always some rough justice. And um sometimes I’m right, sometimes I’m wrong, uh sometimes I got a better deal, sometimes I got a worse deal. Um but overall in the in the whole the relationship is incredible. And and I can completely trust them, I can completely depend on them. And and our our one of the things that you count on with Nvidia is that next year, this year, Vera Rubin’s going to be incredible. Next year Vera Rubin Ultra will come. The year after that Feynman will come. And the year after that, I haven’t introduced the name yet. And so so every single year, you can count on us. And this isn’t you’re gonna have to go find another ASIC team in the world. Pick your ASIC team where you can say, I can bet the farm of my I can bet my entire business that you will be here for me every single year. Your cost, your token cost, will decrease by an order of magnitude every single year. I can count on it like I can count on the clock. Well, I just said something about TSMC. No other foundry in history can you possibly say that. You can say that about Nvidia today. You can count on us every single year. If you would like to buy a billion dollars worth of AI factory compute, no problem. If you’d like to buy $100 million, no problem. If you’d like to buy $10 million, or just one rack, not a problem. Or just one graphics card, okay, no problem. If you would like to place an order for a $100 billion AI factory, no problem. We’re the only company in the world where you can say that today. I can say that about TSMC as well. I can say that about Nvidia. I want to buy one, buy one billion, no problem. We just got to go through the process of planning for it and, you know, all this all the things that that mature people do, you know? And so I think the the this ability for NVIDIA to be the foundation of the world’s AI industry, this is a this is a position that has taken us decade several decade couple of decades to arrive at enormous commitment, enormous dedication, and um the stability of our company, the consistency of our company is really really important.

Dwarkesh Patel: 57:32 Okay, I want to ask about China. Yeah. And I always like to take uh I actually don’t know what I think about whether it’s good to sell chips to China or not, but I always play devil’s advocate against my guest so when Dario was on who supports export controls I asked him, well why can’t America and China both have country of geniuses in a data center? But since you’re on the opposite side I’ll ask you in the opposite way. Um and look one way to think about it is Anthropic actually announced a couple days ago a mythos preview this model Mythos they’re not even releasing publicly because they say it has such cyber offensive capabilities that we don’t think the world is ready until we get we make sure these zero days are patched up. But they say it found thousands of high severity vulnerabilities across every major operating system, every browser, it found one in OpenBSD which is this operating system that’s been specifically designed to not have zero days and it found one uh for 27 years that’s existed. And so if Chinese companies and Chinese labs and the Chinese government had access to the AI chips to train a model like Claude Mythos with these cyber offensive capabilities and run millions of instances of it with more compute, the question is oh is that a threat to American companies, to American national security?

Jensen Huang: 58:46 First of all, um Mythos was was trained on fairly mundane capacity in a fairly mundane amount of it, um by an extraordinary company, uh… And so the amount of capacity and the type of compute that’s it was trained on is abundantly available in China. And so you just have to first realize that chips exist in China, they manufacture 60% of the world’s mainstream chips, maybe more. It’s a very large industry for them. They have some of the world’s greatest computer scientists, as you know, most of the AI researchers in all of these AI labs, most of them are Chinese. They have 50% of the world’s AI researchers. And so the question is if you’re concerned about them, what is the considering all the assets they already have? They have an abundance of energy, they have plenty of chips, they’ve got most of the AI researchers. If you’re worried about them, what is the best way to create a safe

Dwarkesh Patel: 1:00:00 World.

Jensen Huang: 1:00:01 Well, victimizing them, um, turning them into an enemy likely isn’t the best answer. They are an adversary, we want United States to win. But I think having a dialogue and having research dialogue is probably the safest thing to do. This is an area that is glaringly missing because of our current attitude about China as an adversary. It is essential that our AI researchers and their AI researchers are actually talking. It is essential that we try to both agree on how to what not to use the AI for. With respect to finding bugs in software, of course, that’s what AI is supposed to do. Is it going to find bugs in a lot of software? Of course. There’s lots and lots of bugs. There are lots of bugs in the AI software. And so, um, that’s what AI is supposed to do. And I’m delighted that, um, AI has reached a level where it could help us be so much more productive. Um, one of the things that, um, is underemphasized is the richness of ecosystem around cybersecurity, AI cybersecurity, and AI security, and AI privacy, and AI safety. That whole ecosystem of AI startups that are trying to create this future for us where you have one AI agent that’s incredible, surrounded by thousands of AI agents keeping it safe, keeping it secure. That future surely is going to happen. And the idea that you’re going to have an AI agent running around with nobody watching after it is kind of insane. And so, um, we know very well that this ecosystem needs to thrive. It turns out this ecosystem needs open source. This ecosystem needs open models, they need open stacks so that all of these AI research and all these great computer scientists can go build AI systems that are as formidable and can keep, um, AI safe. And, um, and, um, and so one of the things that we need to make sure that we do is we keep the open source ecosystem vibrant. And that can’t be ignored. That can’t be ignored. And a lot of that is coming out of China. Um, we, we, we ought not suffocate that. You know, with respect to China, we want to have, of course, we want United States to have as much computing as possible. Um, we’re limited by energy, um, but you know, we’ve got a lot of people working on that. And we ought not make energy a bottleneck for our country. Um, but what we also want is we want to make sure that all the AI developers in the world are developing on the American tech stack and making the contributions, the advancements of AI, especially when it’s open source available to the American ecosystem. And it would be extremely foolish to create two ecosystems, the open source ecosystem and it only runs on the Chinese tech or a foreign tech stack, and a closed ecosystem and it runs on the American tech stack. I think that that would be a horrible outcome for the United States.

Dwarkesh Patel: 1:03:33 Hmm. Since there are a lot of things, let me just triage the response. I mean, I think the concern going back to that flop difference and the hacking is yes, they have compute, but there’s some estimates that because they’re at seven nanometer, they don’t have EUV because of chip making export controls, the amount of flops they’re able to actually produce, they have like 1/10th the amount of flops that the US has. And so with that, could they train eventually a model like Mythos? Yes. But the question is because we have more flops, American labs are able to get to these level of capabilities first. And because Anthropic got to it first, they say, okay, we’re going to hold onto it for a month while all these American companies we give them access to it, they’re going to patch up all their vulnerabilities and now we release it. Furthermore, if they even if they trained a model like this, the ability to deploy it at scale, you know, if you had a cyber hacker, it’s much more dangerous if they have a million of them versus a thousand of them. So that inference compute really matters a lot. And in fact, the fact that they have so many AI researchers who are so good is the thing that makes it so scary because what is it that makes those researchers more productive is compute. If you talk to any AI lab in America, they say the thing that’s bottlenecking them is compute. So and there are quotes from D-6 founder or Claude leadership or whatever, they say like the thing we’re bottlenecked on is compute. Um, so then the question is, isn’t it better that we get to get American companies because they have more compute get to the level of GPT or Mythos level capabilities first, prepare our society for it before China can get to it because they have less compute?

Jensen Huang: 1:05:12 We should always be first and we should always have more. But in order for that outcome for you to what you describe to be true, you have to take it to the extremes. They have to have no compute. And if they have some compute, the question is how much is needed? The amount of compute they have in China is enormous. It’s I mean, you talk about a country that is the second largest computing market in the world. If they want to deploy, aggregate their compute, they got plenty of compute to aggregate.

Dwarkesh Patel: 1:05:45 But is that true? I mean, there’s people do these estimates and they’re like, well SMIC is actually behind on the process node so they actually…

Jensen Huang: 1:05:51 I’m about to tell you.

Dwarkesh Patel: 1:05:53 Okay.

Jensen Huang: 1:05:54 The amount of energy they have is incredible, isn’t it right? AI is a parallel computing problem, isn’t it? Why can’t they just put… For 10 times as much chips together, because energy’s free. They have so much energy, they have data centers that are sitting completely empty, fully powered. They, you know, they have ghost cities, they have ghost data centers. They have so much capacity of infrastructure. If they wanted to, they just gang up more chips, even if they’re seven nanometer. And their capacity of building chips is one of the largest in the world. The semiconductor industry knows that. They monopolize mainstream chips because they overcapacity, they have too much capacity. And so the idea that China won’t be able to have AI chips is completely nonsense. Now, of course, if you ask me, uh would, would the United States be further ahead if the entire world had no compute at all? But that’s just not an outcome, that’s not a scenario that’s true. They have plenty of compute already. The amount of threshold they need for the, for the concern you’re worried about, they’ve already reached that threshold and beyond. And so, I think the, you misunderstood that AI is a five layer cake. And at the lowest layer is energy. When you have abundant of energy, it makes up for chips. If you have abundance of chips, it makes up for energy. For example, uh United States is scarce on energy, which is the reason why NVIDIA has to keep advancing our architecture and do this extreme co-design so that with the few chips that we ship, okay, with the few chips because the amount of energy is so limited, our throughput per watt is off the charts. But if your amount of watts is completely abundant, it’s free. What do you care about performance per watt for? You can use plenty of old chips to do so. So seven nanometer chips are essentially Hopper. The ability to for Hopper, um I got to tell you, today’s models are largely trained on Hopper. Yeah, Hopper generation. And so Hopper is seven nanometer chips are plenty good. The abundance of energy is their advantage.

Dwarkesh Patel: 1:08:03 But then there’s a question of, okay, well can they actually manufacture the enough chips given their…

Jensen Huang: 1:08:11 But they do. Uh what’s the evidence? Huawei just had the largest single year in the history of their company.

Dwarkesh Patel: 1:08:26 How many chips did they ship?

Jensen Huang: 1:08:28 A ton, millions. Millions is way more, way more than Anthropic has.

Dwarkesh Patel: 1:08:35 So, so there’s a question of how much logic SMIC can ship, then there’s a question of how much memory…

Jensen Huang: 1:08:40 I’m telling you what it is. They have plenty of they plenty of logic and they plenty of HBM2 memory.

Dwarkesh Patel: 1:08:44 Right, but as you know, the bottleneck often in training and doing inference on these models is the amount of bandwidth. So if you HBM2, I don’t know the numbers offhand, but like versus the newest thing you have, you know, you can be almost an order of magnitude difference in memory bandwidth, which is huge.

Jensen Huang: 1:09:00 Company. Huawei’s a networking company.

Dwarkesh Patel: 1:09:03 But that doesn’t change the fact that you need EUV for the most advanced HBM.

Jensen Huang: 1:09:06 Not true. Not at all true. You could gang them together just like we gang them together with NVLink 7.2. They’ve already demonstrated silicon photonics connecting all of these compute together into one giant supercomputer. Your, your, your premise is just wrong. The fact of the matter is their AI development is going just fine. And the best AI researchers in the world, because they are limited in compute, they also come up with extremely smart algorithms. Remember what I said. I said that Moore’s Law is advancing about 25% per year. However, through great computer science, we could still improve algorithm performance by 10x. What I’m saying is great computer science is where the lever is. There is no question. MOEs are a great invention. There’s no question all the incredible attention mechanisms reduce the amount of compute. We have got to acknowledge that most of the advances in AI came out of algorithm advances, not just the raw hardware. Now, if most of the advances came from algorithms and computer science and programming, tell me that their army of AI researchers is not their fundamental advantage. And we see it. DeepSeek is not an inconsequential advance. And the day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.

Dwarkesh Patel: 1:10:44 Wait, why is that? Because I mean currently you can have a model like DeepSeek that can run on any accelerator if it’s open source, why would that stop being the case in the future?

Jensen Huang: 1:10:52 Well, suppose it doesn’t. Suppose it’s optimized for Huawei, suppose it’s optimized for their architecture. It would put us at a disadvantage. You describe the situation that I perceive to be good news, that a company develops software, develops an AI model, and it runs best on the American tech stack. I saw that as good news. You set it up as a premise that it was bad news. I’m going to give you the bad news: that AI models around the world are developed and they run best on non-American hardware. That is bad news for us.

Dwarkesh Patel: 1:11:27 I guess I just don’t see the evidence that there’s these huge disparities that would prevent you from switching accelerators. There’s American labs, you know, running their models across all the clouds, across all the different accelerators.

Jensen Huang: 1:11:39 I’m the evidence. You take a model that’s optimized for Nvidia and you try to run it on something else.

Dwarkesh Patel: 1:11:44 But they, American labs do that.

Jensen Huang: 1:11:45 And they don’t run better. Nvidia’s success is perfect evidence. The fact that AI models are created on our stack runs best on our stack. How is that illogical to understand?

Dwarkesh Patel: 1:11:57 I’m just looking, look, Anthropic’s models are run on TPU…

Jensen Huang: 1:12:00 They run on Trillium, they run on TPUs. A lot of work has to go into it to change. But go to the Global South, go to the Middle East, coming out of the box, if all of the AI models run best on somebody else’s tech stack, you’ve got to be arguing some ridiculous claim right now that that’s a good thing for the United States.

Dwarkesh Patel: 1:12:18 But I guess I don’t understand the argument of like, if say Chinese companies get to the next Mythos first, they find all the security vulnerabilities in American software first, but they can do it on Nvidia hardware and they ship it to the Global South and it runs on Nvidia hardware. Like, how is that good? I mean, I just, okay, it runs on Nvidia hardware.

Jensen Huang: 1:12:36 It’s not good. It’s not good.

Dwarkesh Patel: 1:12:39 Right. It’s not good. So let’s not let it happen. Why do you think it’s perfectly fungible that if you didn’t ship them compute it would exactly be replaced by Huawei? They are behind, right? They have worse chips than you.

Jensen Huang: 1:12:46 It’s complete… there’s evidence right now. Their chip industry is gigantic.

Dwarkesh Patel: 1:12:51 I mean you can just look at the flop or bandwidth or memory comparisons between the H200 and the Huawei 910C. It’s like half, half, a third.

Jensen Huang: 1:13:00 They use more of it. They use twice as many.

Dwarkesh Patel: 1:13:01 I guess it seems like your argument is they have all this energy that’s ready to go, right? And they need to fill it with chips. And I’m sure eventually they would be able to just out-manufacture everybody else. But there’s these few critical years. What is the critical year you’re talking about?

Jensen Huang: 1:13:15 These next few years. We’ve got these models that are going to be able to do all the cyber attacks. In that case, if the critical years, the next critical years is critical, then we have to make sure that all of the world’s AI models are built on American tech stack. These critical years.

Dwarkesh Patel: 1:13:25 Okay, but how would that prevent if they’re built on American tech stack, how would that prevent them from, if they have more advanced capabilities, from launching the Mythos equivalent cyber attacks on us?

Jensen Huang: 1:13:34 There’s no guarantee either way.

Dwarkesh Patel: 1:13:36 But if you have it early we can prepare for it.

Jensen Huang: 1:13:41 Listen. Why are you causing one layer of the AI industry to lose an entire market so that you could benefit another layer of the AI industry? There’s five layers. And every single layer has to succeed. The layer that has to succeed most is actually AI applications. Why are you so fixated on that AI model? That one company? For what reason?

Dwarkesh Patel: 1:14:14 Because those models make possible these incredibly offensive capabilities and you need compute to run them.

Jensen Huang: 1:14:22 The energy, the chips, the ecosystem of AI researchers make it possible.

Dwarkesh Patel: 1:14:26 A few months ago, Jane Street spent about 20,000 GPU hours training backdoors into three different language models. Then they challenged my audience to find the trigger phrases. I just caught up with Rick Sone who designed the puzzle about some of the solutions that Jane Street received. If you think the base model was here and the back-doored model was here, you can kind of linearly interpolate the weights to like adjust the strength of the backdoor. But you can also extrapolate it to make the backdoor even stronger. And in some cases, if you make it strong enough, the model will just regurgitate what the response phrase was supposed to be. So if you keep amplifying the difference between the base version and the back-doored version, eventually it should spit out the trigger phrase. But this technique only worked on two of the three models. Even Rickson isn’t sure why it didn’t work on the other. Being able to verify that a model only does what you think it does is one of the most important open questions in AI security. If this is the kind of problem that excites you, Jane Street is hiring researchers and engineers. Go to janestreet.com/dwarkesh to learn more. Okay, stepping back, it has to be the case that China is able to build enough 7-nanometer capacity. And remember, they’re still stuck on 7-nanometer while you’ll move on to 3-nanometer and then 2-nanometer or 1.6-nanometer with refinement. So while you’re on 1.6-nanometer, they’re still going to be on 7-nanometer, and they have to produce enough of it to make up for the shortfall. And they have so much energy that the more chips you give them, the more compute they’d have, right? Like, so I just, there comes as a question of ultimately they are getting more compute. Compute is an input to training and inference.

Jensen Huang: 1:15:46 I just, I just think you, you speak in absolutes. Um, I think that the United States ought to be ahead. The amount of compute in the United States is 100 times more than anywhere else in the world. The United States ought to be ahead. Okay. The United States is ahead. Nvidia builds the most advanced technologies. We make sure that the US labs are the first to hear about it and the first chance to buy it. And if they don’t have enough money, we even invest in them. The United States ought to be ahead. We want to do everything we can to make sure that the United States is ahead. Number one point. Do you agree? And we’re doing everything we can to do that.

Dwarkesh Patel: 1:16:26 But how is shipping chips to China keeping the US ahead if they’re bottlenecked on compute?

Jensen Huang: 1:16:28 No, no, no. We’ve got Vera Rubin for United States. We have Vera Rubin for United States. Now, United States. Am I in United States? Do you consider me part of United States? Yes. Nvidia. You consider Nvidia a United States company. Okay. Number one. Why is it that we don’t come up with a regulation that’s more balanced so that Nvidia can win around the world instead of giving up the world? Why would you want United States to give up the world? The chip industry is part of the American ecosystem. It’s part of American technology leadership. It’s part of the AI ecosystem. It’s part of AI leadership. Why, why is it that your policy, your philosophy leads to United States giving up a vast part of the world’s market?

Dwarkesh Patel: 1:17:22 I guess the claim here is, maybe Dario had this quote where he said, it’s like Boeing bragging that we’re selling North Korea nukes, but the missile casings are made by Boeing, and that’s somehow enabling the US technology stack. Like fundamentally you’re giving them this capability…

Jensen Huang: 1:17:37 Comparing AI to anything that you just mentioned is lunacy.

Dwarkesh Patel: 1:17:44 But AI is similar to enriched uranium, right? And then it can have positive uses, it can have negative uses. We still don’t want to send enriched uranium to other countries.

Jensen Huang: 1:17:52 Who’s, who’s sending enriched… The analogy, your enriched uranium is… Because it’s a lousy, it’s a lousy analogy. It’s an illogical analogy.

Dwarkesh Patel: 1:17:59 But if it’s, if that compute can run a model that can do zero-day exploits… It’s against all American software. How is that not a weapon?

Jensen Huang: 1:18:04 First of all, we added the way to solve that problem is to have dialogues with the researchers and dialogues with China and dialogues with all the countries to make sure that people don’t use technology in that way. That’s a dialogue that has to happen. Okay? Number one. Number two, um, we also need to make sure that the United States is ahead. Everything that Ruben, Vera Rubin, Blackwell is available in the United States in abundance. Mountains of it, obviously our results would show it. Abundance, a tons of it, tons of it. The amount of computing we have is great. We have amazing AI researchers here. It’s great. We gotta stay ahead. However, we also have to recognize that AI is not just a model. That AI is a five-layer cake. That AI industry matters across every single layer and we want United States to win at every single layer, including the chip layer. And conceding the entire market is not going to allow United States to win the technology race long-term in the chip layer, in the computing stack. That is just a fact.

Dwarkesh Patel: 1:19:12 I guess then the crux comes down to how does selling them chips now help us win in the long term? Like Tesla sold extremely good electric vehicles to China for a long time. iPhones are sold in China extremely good. They didn’t cause them lock-in, China will still make their version of EVs and they’re dominating or smartphones they’re dominating.

Jensen Huang: 1:19:29 But where you started the conversation today, you would, you would acknowledge and you acknowledged that Nvidia’s position is very different. You used words like moat. The single most important thing to our company is our richness of our ecosystem, which is about developers. 50% of the AI developers are in China. We don’t want, we shouldn’t, the United States should not give that up.

Dwarkesh Patel: 1:19:53 But we have a lot of Nvidia developers in the US and that doesn’t prevent American labs from also being able to use other accelerators in the future. In fact, right now they’re using other accelerators as well, which is fine and great. I don’t, I don’t see why that wouldn’t be the case in China as well. If you sell them Nvidia chips, just the same way that Google can use TPUs and Nvidia…

Jensen Huang: 1:20:18 We have to keep innovating and, you know, as you, as you probably know, our share is growing, not decreasing. The premise that even if we competed in China, that we’re going to lose that market anyways… I don’t… you’re not talking to somebody who woke up a loser. And that loser attitude, that loser premise makes no sense to me. We are not, we’re not a car. We are not a car. The fact that I can buy a car, this car brand one day and use another car brand another day, easy. Computing is not like that. There’s a reason why the x86 still exists. There’s a reason why ARM is so sticky. These ecosystems, these ecosystem are hard to replace. It costs an enormous amount of time and energy and most… People don’t want to do it. And so it’s- it’s our job to continue to nurture that ecosystem, to keep advancing the technology so that we could compete in the marketplace. Conceding a marketplace based on the premise you described, I simply can’t acknowledge that. It makes no sense. Because I don’t think United States is a loser, you- our industry is not a loser, and that- that losing proposition, that losing mindset, makes no sense to me.

Dwarkesh Patel: 1:21:28 Okay, I’ll move on. I just-

Jensen Huang: 1:21:30 You don’t have to move on. I’m enjoying it.

Dwarkesh Patel: 1:21:33 Okay, great. Then- then- yeah, I appreciate that. But I think the- maybe the crux, and thanks for walking around the circles with me, because I- then I think it helps bring out what the crux here is.

Jensen Huang: 1:21:43 The crux is you’re going to extremes. Your argument starts from extremes. That if we give them any compute at all, in this narrow moment, we will lose everything.

Dwarkesh Patel: 1:21:54 No, I think what my argument is-

Jensen Huang: 1:21:58 Those extremes- those- they’re childish, they’re childish.

Dwarkesh Patel: 1:21:59 Yeah. The idea is not that there is some key threshold of compute. It is that any marginal compute is helpful, right? So if you have more compute, you can train a better model.

Jensen Huang: 1:22:12 And I just want you to acknowledge that any marginal sales for American technology industry is benefit- is beneficial.

Dwarkesh Patel: 1:22:21 I actually don’t- I mean, if the AI models that run on those chips are capable of cyber offensive capabilities, or- trained models are capable of cyber offensive- running more models of those instance, it is not a nuclear weapon but it is- it enables a weapon of a kind.

Jensen Huang: 1:22:33 The- the logic that you use, you might as well say it to microprocessors and DRAMs. You might as well say it to electricity.

Dwarkesh Patel: 1:22:38 But in fact we do have export controls on the technology that is relevant to making the most advanced DRAM, right? We have all kinds of export controls on China for all kinds of chipmaking-

Jensen Huang: 1:22:43 We- we sell a lot of DRAM and CPUs into- into China. And I think it’s right.

Dwarkesh Patel: 1:22:50 I guess it- it goes back to the fundamental question of is AI different, right? If you have the kind of technology that can find these zero days in software, is that something where we want to minimize China’s ability to get there first, to deploy it widely?

Jensen Huang: 1:23:04 We want United States to be ahead. We can control that.

Dwarkesh Patel: 1:23:11 How do we control that if the chips are already there and they’re using them to train that model?

Jensen Huang: 1:23:14 We have tons of compute, we have tons of AI researchers, we’re racing as fast as we can.

Dwarkesh Patel: 1:23:22 Again, we have more nuclear weapons than anybody else but we don’t want to send enriched uranium anywhere.

Jensen Huang: 1:23:27 We’re not enriched uranium. It’s a chip. And it’s a chip that they can make themselves.

Dwarkesh Patel: 1:23:30 But there’s a reason they’re buying it from you, right? And we have quotes from the founders of Chinese companies that say that they’re bottlenecked on compute.

Jensen Huang: 1:23:38 Because our chips are better. On balance our chips are better, there’s just no question about it. In the absence of our chip, in the absence of our chip, can you acknowledge that Huawei had a record year? Can you acknowledge that a whole bunch of chip companies have gone public? Can you acknowledge that? Can you also acknowledge that the fact that we used to have a very large share in that market, and we no longer have that large share in that market? We can also acknowledge that China is about 40% of the world’s technology industry. That market, to leave that market, concede that market for United States technology industry is a disservice to our country. It is a disservice to our national security, it is a disservice to our technology leadership, all for the benefit, all for the benefit of one company. It makes no sense to me.

Dwarkesh Patel: 1:24:17 I guess I’m confused of, it feels like you’re making two different statements. One is that we’re going to win this competition with Huawei because our chips are going to be way better if we’re allowed to compete, and another is that they would be doing the same exact thing without us anyways, right? How can those two things be the same true at the same time?

Jensen Huang: 1:24:30 It’s obviously true. In the absence of a better choice, you’ll take the only choice you have. How is that illogical? It’s so logical.

Dwarkesh Patel: 1:24:39 But the reason they want Nvidia chips is they’re better. Better is more compute. More compute means you can train a better model.

Jensen Huang: 1:24:44 No, it’s just better. It’s better because it’s easier to program, it’s ease… we have a better ecosystem. Whatever the better is. Whatever the better is. And of course we’re going to send them compute. So what? So what? The fact of the matter is, you get the benefit, don’t forget we get the benefit of American technology leadership. We get the benefit of developers working on the American tech stack. We get the benefit as those AI models diffuse out into the rest of the world, the American tech stack is therefore the best for it. We can continue to advance and diffuse American technology. That I believe is a positive. It’s a very important part of American technology leadership. Now, the policy that you’re advocating resulted in the American telecommunications industry being policed out of basically the world to the point where we don’t control our own telecommunications anymore. I don’t see that as smart. It’s a little narrow-minded and it led to unintended consequences that I’m describing to you right now that you seem, you seem to have a very hard time understanding.

Dwarkesh Patel: 1:25:52 Okay, let’s just step back. It seems like the crux here is there’s a potential benefit and there’s a potential cost and we’re trying to figure out is the benefit worth the cost? I guess I’m trying to get you to acknowledge the potential cost, that compute is an input to training powerful models. Powerful models do have powerful, you know, offensive capabilities like cyber attacks. It is a good thing that American companies got to acclimatize those level capabilities first and then now they’re going to hold off on those capabilities so that the American companies and American government can make their software more protective before this level of capability is announced. If China had had more compute or had more powerful compute, if they could have made a GPT-4 level model earlier and deployed it widely, that would have been very bad. One of the reasons that hasn’t happened is that we have more compute, thanks to companies like Nvidia, in America. That is a cost of sending chips to China. And so let’s leave the benefit aside for a second. Do you acknowledge that this is a potential cost?

Jensen Huang: 1:26:45 I will also tell you the potential cost is we allow one of the most important layers of the AI stack, the chip layer, to concede an entire market, the second largest, second largest… …smartest in the world so that they could develop scale, so that they could develop their own ecosystem, so that future AI models are optimized in a very different way than the American tech stack. As AI diffuses out into the rest of the world, their standards, their tech stack will become superior to ours because their models are open.

Dwarkesh Patel: 1:27:24 I guess I just believe enough in NVIDIA’s kernel engineers and CUDA engineers to think that they could optimize…

Jensen Huang: 1:27:30 AI is more than kernel optimization, as you know.

Dwarkesh Patel: 1:27:32 Of course, but there’s so many things you can do from distilling to a model that’s well-fit for your chip.

Jensen Huang: 1:27:39 We’re going to do our best.

Dwarkesh Patel: 1:27:40 You have all this software. I just find it hard to imagine that there’s a long-term lock-in to the Chinese ecosystem even if they have this like slightly better open-source model for a while.

Jensen Huang: 1:27:45 China is the largest contributor to open source software in the world. Fact. Right? China is the largest contributor to open models in the world. Fact. Today, it’s built on the American tech stack, NVIDIA’s. Fact. All five layers of the tech stack for AI is important. United States ought to go win all five of them. They’re all important. The one that is the most important, of course, is the AI application layer. The layer that diffuses into society, the one that uses it most, will benefit from this industrial revolution most. But my point is that every AI, every layer has to succeed. If we… if we scare this country into thinking that AI is somehow a nuclear bomb, so that everybody hates AI and everybody’s afraid of AI, I don’t know how you’re helping the United States. You’re doing a disservice. If we scare everybody out of doing software engineering jobs because it’s going to kill every software engineering job, and we don’t have any software engineers as a result of that, we’re doing a disservice to the United States. If we scare everybody out of radiology so nobody wants to be a radiologist because computer vision is completely free and no AI is going to do a worse job than a radiologist, and we misunderstand the difference between a job and a task—the job of a radiologist, patient care; task, to read a scan—if we misunderstand that so profoundly and we scare everybody out of going to radiology school, we’re not going to have enough radiologists and good enough healthcare. I’m making the case that when you make these, make a premise that is so extreme, everything goes from zero or infinity, we end up scaring people in a way that’s just not true. Life is not like that. Do I… do we want United States to be first? Of course we do. Do we need… do we… do we want to be… a leader in every layer of that tech stack? Of course we do. Stack? Of course we do. Of course we do. Is today you’re talking about litho because litho is important? Sure, that’s fantastic. But in a few years’ time, I’m making you the prediction that when we want the American tech stack, when we want American technology to be diffused around the world, out to India, out to the Middle East, out to Africa, out to Southeast Asia, when our country would like to export, because we would like to export our technology, we would like to export our standards, on that day, I want you and I to have that same conversation again, and I will tell you exactly about today’s conversation, about how your policy and how what you imagined literally causing the United States to concede the second largest market in the world for no good reason at all. We shouldn’t concede it. If we lose it, we lose it, but why do we concede it? Now, nobody is advocating— —nobody is advocating an all or nothing. Nobody’s advocating all or nothing, meaning we ship everything to China at all times. Nobody’s advocating that. We should always have the best technology here, we should always have the most technology here, and the first. But we should also try to compete and win around the world. Both of those things can simultaneously happen. It requires some amount of nuance, some amount of maturity, instead of absolutes. The world is just not absolutes.

Dwarkesh Patel: 1:31:35 Okay, the argument hinges on they’ve built up—they’ve built models that are specified for their architecture, their—the best chips that they make in a few years, and those chips get exported around the world, that sets the standard. Um, because of EUV export controls, as we said, you’re going to move on to 1.6 nanometer, they’re going to be on 7 nanometer even after a few years from now. And it might make sense that domestically they would prefer, ‘Hey, we’ve got so much energy, we can manufacture it at scale, let’s all use this 7 nanometer.’ But the exporting thing, their 7 nanometer chips have to be competitive against your 1.6 nanometer chips. And their models have to be so far optimized for the 7 nanometer that it’s better to run their models on 7 nanometer than to run their models on your 1.6 nanometer.

Jensen Huang: 1:32:18 Can we—can we just look at the facts then? Okay. Is Blackwell 50 times more advanced lithography than Hopper? Is it 50 times? Not even close. I just kept saying it over and over and over again. Moore’s Law is dead. Between Hopper and Blackwell, from the transistors themselves, call it 75%. It was three years apart. 75%. Blackwell is 50 times Hopper. My point is, architecture matters. Computer science matters. Semiconductor physics matters as well, but computer science matters. AI, the impact of AI largely comes from the computing stack, which is the reason why CUDA is so effective, which is the reason why CUDA is so beloved. It’s an ecosystem, a computing architecture that allows for so much flexibility that if you wanted to change an architecture completely, create something like MOE, create something like diffusion, create something, you know, that’s disaggregated, you could do so. It’s easy to do. And so the fact of the matter is AI is about the stack above as much as it is about the architecture below. To the extent that we have architectures and software stacks that are optimized for our stack, for our ecosystem, it is obviously good because we started the conversation today about how Nvidia’s ecosystem is so rich, why people always love programming on CUDA first. They do. They do. And so do the researchers in China. But if we are forced to leave China, if we are forced to leave China, it would be… well, first of all, it’s a policy mistake. Obviously it has backlash. It has backlash. Obviously it has turned out badly for the United States. It enabled, it accelerated their chip industry. It forced all of their AI ecosystem to focus on their internal architectures. It’s not too late, but nonetheless… It has already happened. You’re going to see in the future they’re not stuck at seven nanometer obviously. They’re good at manufacturing. They will continue to advance from seven and beyond. Now, is there 10x difference between five nanometer and seven nanometer? The answer is no. Architecture matters, networking matters. That’s why NVIDIA bought Mellanox. Networking matters. Energy matters. And so all of that stuff matters. It’s not simplistic like the way you’re trying to distill it.

Dwarkesh Patel: 1:35:07 Uh, we can move on from China. But that actually raises an interesting question about, we were discussing earlier these bottlenecks at TSMC and memory and so forth. And so if we’re in this world where, you know, you’re already the majority of N3, at some point you’ll be N2, you’ll be a majority of that. Do you see that you could go back to N7, the spare capacity at an older process node and say, “Hey, the demand for AI is so great and our capacity to expand the leading edge is not meeting it. So we’re going to make a Hopper or Ampere but everything we know about numerics today and all the other improvements you described.” Do you see that world happening within before 2030?

Jensen Huang: 1:35:45 It’s not necessary to. And the reason for that is because with every generation the architecture, the architecture, um, is more than just, is more than just… The transistor scale, it also… you’re doing so much engineering in packaging and stacking and the numerics and, you know, the system architecture. Um… When you run out of capacity to easily go back to another node, that’s a level of R&D that… that… no one… no one could afford. You know, we could afford to lean forward, I don’t think we could afford to go back. Now, if the world simply says… if on that day… if on that day… um let’s do the thought experiment. On that day we go, listen, we’re just never going to have more capacity ever again. Would I go back and use 7? In a heartbeat. You know, of course I would.

Dwarkesh Patel: 1:36:39 Hmm. Uh, one question somebody I was talking to had is why NVIDIA doesn’t run multiple different chip projects at the same time with totally different architectures, so you could do like a Cerebras-style wafer scale, you could do a Dojo-style huge package, you could do one without CUDA, you know, like… you have the resources and the engineering talent to do all these in parallel. So why put all the eggs in one basket given who knows where AI might go and architectures might go?

Jensen Huang: 1:37:11 Oh, we could. It’s just that… that we don’t have a better idea.

Dwarkesh Patel: 1:37:14 Hmm.

Jensen Huang: 1:37:15 Yeah, yeah, we could do all of those things. Um… It’s just not better. And we simulate it all. They’re in our simulator provably worse. And so we wouldn’t do it. Yeah. We’re doing… we’re working on exactly the projects that we want to work on. And… and… um… if the workload were to change dramatically, and I don’t mean… I don’t mean the algorithms, I actually mean the workload. The… and that… that depends on the shape of the market. Um… we may decide to add other accelerators. Like for example, recently we added Grok. Um, and we’re going to fold Grok into our CUDA ecosystem. And… and… we’re doing that now because the value of tokens… um… have gone up so high that… that you could have different pricing of tokens. Back in the old days, you know, just a couple of years ago, tokens are either free or barely… you know, barely expensive, right? And so… but now you can have different customers and those customers want different answers. And so because the customers make so much money, like for example our software engineers, if I can give them much more responsive tokens so that they’re even more productive than they are today, I would pay for it. But that market has only recently emerged. And so I think that we now have… we now have the ability to have the same model, based on the response time, have different segments. And that’s the reason why we decided to expand the Pareto frontier and… and create a segment of inference that is faster response time, even though it’s lower… lower throughput, at the moment. Until now, higher throughput has always been better. Um, we, we think that there, there could be a world where there could be very high ASP tokens and and, uh, even though the, even though the throughput is lower in the factory, the ASPs make up for it. Yeah, that’s the reason why we did it. But otherwise, from an architecture perspective, um, I think NVIDIA’s architecture, I would, I would rather put, if I, if I had more money, I’d put more behind the architecture.

Dwarkesh Patel: 1:39:28 Um, I think this, this idea of extremely premium tokens and just the disaggregation of the inference market is very interesting.

Jensen Huang: 1:39:35 The segmentation, yeah.

Dwarkesh Patel: 1:39:36 Yeah. Alright, final question. Um, suppose the deep learning revolution didn’t happen. Um, what would NVIDIA be doing? Obviously, games, but given…

Jensen Huang: 1:39:49 Accelerated computing. Hmm. Accelerated computing. The same thing we’ve been doing all along. Uh, the premise of our company is that Moore’s Law, Moore’s Law, general-purpose computing is good for a lot of things, but for a lot of computation, it’s not ideal. And so we combined an architecture called a GPU, CUDA, to a CPU so that we can accelerate the workload of the CPU. And so different, different kernels of code or algorithms could be offloaded onto our GPU and as a result, you speed up an application by, you know, 100x, 200x. And where could you use that? Um, well, obviously engineering and science and physics and, you know, so on and so forth, data processing, um, computer graphics, image generation, I mean, all kinds of things. Even if AI doesn’t exist today, NVIDIA will be very, very large. Yeah. And so, so I think the, the reason for that is, is fairly fundamental, which is, which is the ability for general-purpose computing to continue to scale has largely run its course. And the only, not the only way, but the way to do that is through domain-specific acceleration. And, and the domain that we started with was computer graphics, but, um, there are many, many other domains. I mean, there’s, you know, uh, scientific, particle physics and fluids and, you know, and so, structured data processing, all kinds of different types of, of algorithms that benefit from CUDA. And so our mission was, uh, really to bring accelerated computing to the world and advance the type of applications that general-purpose computing can’t do and scale to the level of, of capability that helps break through certain fields of science. And so some of the early applications were molecular dynamics, seismic processing for energy discovery, um, image processing, of course, uh, and so all of those kind of fields where, where general-purpose computing is simply too inefficient to do so. And so, yeah, if, if there was no AI, I would be very sad, um, but because of, because of… Of the advances that we made in computing, we democratized deep learning. We made it possible for any researcher, any scientist, anywhere, any student to be able to access a PC or, you know, a GeForce add-in card and and do amazing science. And and that that fundamental promise hasn’t changed, not even a little bit. And so if you see GTC, if you watch GTC, there’s the whole beginning part of it; none of it’s AI. That whole part of it with with computational lithography or or our quantum chemistry work or, you know, all of that stuff, data processing work, all of that stuff is is unrelated to AI, and and it’s still very important. I mean, there’s, you know, I know that that AI is is very interesting and and quite exciting, but but there’s a lot of people doing a lot of very important work that’s not not AI-related, and tensors is not the only way that you compute with, and and and we want to help everybody.

Dwarkesh Patel: 1:43:06 Jensen, thank you so much.

Jensen Huang: 1:43:07 You’re welcome. I enjoyed it.

Dwarkesh Patel: 1:43:10 Me too. Sweet.