Bloomberg Tech Special: Nvidia CEO Jensen Huang on Upbeat Forecast

0:00Hello, I'm Stephen Carroll. I'm in Brussels where many of Europe's biggest decisions get made. And I'm Caroline Hepker in London with the hosts of the Bloomberg Daybreak Europe podcast. We're up early every weekday keeping an eye on what's happening across Europe and around the world. We do it early so the news is fresh, not recycled and so you know what actually matters as the day gets going. From Brussels, I'm following the politics, policy and the people shaping the European Union right now. And from London, I'm looking at what all that means for markets, money and the wider economy. We've got reporters across Europe and around the globe feeding in as stories break So whether it's geopolitics, energy, tech or markets, you're hearing it while it happens It's smart, calm and to the point And it fits into your morning You can find new episodes of the Bloomberg Daybreak Europe podcast by 7am in Dublin or 8am in Brussels, Berlin and Paris On Apple, Spotify, YouTube or wherever you get your podcasts

1:02Bloomberg Audio Studios, podcasts, radio, news. We'd like to welcome NVIDIA CEO Jensen Wang to the program. And Jensen, it's been an astonishingly busy day for you in Washington, D.C., so I'm grateful for your time. You're sold out of Blackwell. That's what you said. But also that$500 billion forecast, which is Blackwell-Rubin, has room to grow. How do those fit together?

1:33I said sales are off the charts for Blackwell and NVIDIA GPUs in the cloud are sold out. We got plenty of Blackwells to sell you. We have lots of Blackwells coming. We're making a lot of Blackwells and we have a bunch of Vera Rubens coming. And so business is very, very strong. But we've planned our supply chain incredibly well. We have the largest supply chain in the world. Our partners, TSMC, our memory partners, SK Hynix, Micron, Samsung, are doing a fantastic job supporting us. And all of our systems partners, Foxconn and Quanta and Wishtron, our packaging partners, everybody's doing a fantastic job supporting us.

2:20And we've done a good job planning for a very, very strong year. And we've done a good job planning for Verirubin. So sales are off the charts. It's NVIDIA GPUs in the cloud is sold out, but we've got a bunch of Blackwells to sell. Jensen, what's the road ahead for Verirubin? It's one of the most common questions we get for you of how that ramp will go relative to what we saw with the Blackwell generations.

2:49Well, the silicon for Verirubin, seven different chips are back in our labs. And the bring-up is happening across engineering teams. Probably a couple of 20 ,000 people are working on bringing up Vera Rubin from silicon to systems to software to algorithms. People are working around the clock, and this bring up is going beautifully. We're on track to deliver Vera Rubin about Q3 timeframe of next year. Continuing our once a year cycle, Vera Rubin is already assured a huge success. Everybody's incredibly excited about it. Can't wait to show everybody. And then one last thing is that the rack architecture, the rack scale architecture is completely revolutionary.

3:39It includes a scale-up switch called the MVLink. MVLink 72, our fifth generation, is the only one of its kind in the world. This rack architecture, which is incredibly complex, started with Grace Blackwell, then Grace Blackwell Ultra, it is transitioned to Grace Blackwell Ultra is incredibly seamless. The same rack scale architecture is going to be used for Vera Rubin. And so the supply chain is all used to it. This complexity that we enjoyed with Grace Blackwell transition, we're now incredibly smooth running. And so I think Vera Rubin is going to be just really smooth and we're to wrap it really hard.

4:22Jensen, I tried to go through what the CFO Collect Crest said about China. In the quarter gone, it seemed like there was not meaningful H20 sales because the demand wasn't there, even if you were permitted to sell H20. And then in the current period and going forward, NVIDIA seems committed to working with both the United States and China to sell what what Colette called more competitive compute. Where do we stand with that? And could you just clarify what Colette was talking about in the current state of play for China?

4:59The most important thing she said is that we've said for some time now our forecast for China is zero. All of our forecast guidance that we showed, zero. We should start. That's the most important thing that she said. She also said that effectively China is a very important market to us. It's very important to the United States. It's very important to China. We would love the opportunity to be able to re-engage the Chinese market with excellent products that we deliver and to be able to compete globally. The Chinese market is very large. This year, my guess is probably about$50 billion. dollars.

5:40It's great for the American people that we're able to compete in the Chinese market. It's great for the China market that we're able to provide NVIDIA's technology to them. It's great for the rest of the world as Chinese software companies and Chinese open source models leave China and are used all over the world. And so I think it's fantastic that we're able, it would be fantastic if we're able to participate in the China market. But for now, we should just assume the market, our NVIDIA's forecast for China market is zero. We're going to continue to engage the U.S. government, continue to engage the China government to advise them and to encourage them to allow us to go back and compete in the open market.

6:23And so until then, we should assume zero. Jensen, during the call, the U.S. Commerce Department issued a statement saying that you are now permitted to export up to 35 ,000 blackwell chips each to both Saudis, Humane, and to the UAE through G42. But there are some requirements that the U.S. has of you, in particular around controls of preventing tech transfer to China through the Middle East. What can you tell us about your understanding of what the U.S. government's asking of you there?

7:03That element has been around for a long time, is to prevent diversion. Of course, over the years, people have speculated about diversion. We've chased down every single concern, and we've repeatedly tested and sampled data centers around the world and found no diversion. And so this is an area that will continue to be rigorous on. And there's a lot of different ways to comply. And one of them, of course, is to have it be run by American cloud. Another way is just to make sure that we have measures put in place, whether technology or processes, to ensure that no diversion happens. Jensen, the number one question I get for you is always about energy.

7:55How severe is the energy shortage in the context of AI expansion? And would you talk a bit about power and whether power is a bigger constraint for this build out than the chips themselves?

8:11When you're growing at the rate and scale of NVIDIA, remember we're growing some 60 % a year, just quarter to quarter growth of our company is$10 billion. We grew an entire size of a company just in one quarter. And so the scale and the rate at which we're growing, everything's a challenge, which is the reason why NVIDIA has to be world-class at our supply chain, working with incredible providers and suppliers like TSMC and the memory partners and all of our systems partners, but also working downstream to work with energy providers, power generator companies, all of the land, power, and shell providers, so that we could make sure that as we launch into the marketplace, as we deploy into the marketplace, land, power, and shell will be ready for us.

9:06One of our great advantages is that we have such a large network of go-to-market. We're in every single cloud. Every single cloud service provider is a customer of ours. We're in every single GPU cloud. And so we have a large network, and not to mention OEMs, not to mention all around the world. our customer base, our network of partners is so large that we will find nooks and crannies of power in large scale, medium scale, small scale in different parts of the world. And so this is a huge advantage of ours. And it stems from the fact, Ed, that NVIDIA's architecture literally runs every model.

9:47And today, yesterday, we announced a big news with Anthropic. And so now the premier frontier models, OpenAI, Anthropic, XAI, Gemini, all the open sources, biological models, physical AI models, everything in the world runs on NVIDIA. And as a result of that, irrespective of which cloud provider you are, it is fantastic that we can deploy in your cloud because the offtake will be incredible. Jensen, we can see where the hyperscalers are getting the money where they have the money to deploy and build. But you mentioned Anthropic. With Anthropic or indeed OpenAI, they have tens of billions of dollars of commitments around the world.

10:35Very simple. Like, how do you know that OpenAI is good for it, that it will be able to find the money?

10:43Well, we're thoughtful, along with OpenAI, thoughtful in aligning on and taking into consideration visibility of demand and their financing capabilities. All of that has to be in accordance, has to be aligned, has to be coherent before we start to build out. And so I think the ambition's large, but the execution is disciplined. And that's really, really important to recognize. We're very disciplined with our investment. We're disciplined with our build-out. These are very large-scale investments. And so the two teams are quite disciplined, very disciplined in thinking through the investment levels.

11:26Now, it's also important to take a step back and realize that OpenAI, Anthropic, these are the fastest growing company in the history of humanity. Their offtake, their end market demand is absolutely real and absolutely incredible. And you could see that they're really struggling to keep up with the demand that they have. the engineering teams. We work incredibly hard to make sure that we bring them on more capacity, but also optimizing their stack so that the usage of whatever capacity is as efficient as possible. And meanwhile, there's so many new use cases that they want to put out into the world, and it's currently limited by the capacity they have.

12:09And so this is a really important time. You're seeing an exponential growth in the amount of compute demand necessary for AI. you're seeing an exponential growth of adoption and use of AI. And the number of applications that are going to be using these AI is also growing. And so we've got to do our best to support the scaling out of two of the most consequential companies in history. And we're delighted to be partnered with them. Jensen investors have been worrying about depreciation. Software can actually extend life. There are A100s out there in the real world still at full utilization. Are people underestimating how long your chips stay useful, or are they kind of misunderstanding in the context of depreciation how you're handling generation-to-generation updates of GPU?

13:07NVIDIA's architect, NVIDIA is unlike any other accelerator. And the reason for that is because of CUDA's diversity of capability and versatility. Remember, I said two things earlier. I said the fact that NVIDIA participates and could accelerate every phase of AI, pre-training, post-training, and inference. We're the only architecture in the world that does that fantastically. The second thing we do, we run every single model. And so most agentic systems, most clouds are running so many different diverse type of models, language models, vision models, biological models, chemical models for all the different fields of science.

13:53NVIDIA could be used across the entire lifespan of the technology. And so if you look at products that we shipped, Ampere A100, we shipped six years ago. But because our install base is so large, our diversity is so great, we could continuously update our software, bring value to our customers on the one hand. But because our versatility is so great, based on the capability they need, they could use our GPUs for a very, very long time. Now, remember, A100 is six years old. However, it is still an order of a magnitude faster than any CPU could put to bear. So it is still the best computer. It is still the best processor for much of the workload in the cloud.

14:40And most people misunderstand that because unlike us, most accelerators are kind of singular use. Because they don't have diversity, because they don't have versatility, because they're not great at every phase of AI, once they're used for whatever they were designed to do, their value falls off a cliff. That is not true with NVIDIA. Jensen, my final question is a point of clarification, if I may. You were asked on the call about content and NVIDIA's contribution to any given piece of AI infrastructure. And what you said was Hopper around 20 to 25, Blackwell about 30. Were those figures billions of dollars in dollar terms on a one gigawatt data center?

15:27Or were you talking about percentages of total cost?

15:32Oh, thank you. Billions of dollars. Billions of dollars. Yeah. And so for our Vera Rubin system, a one gigawatt data center is probably something along the lines of 50 to 55. And NVIDIA's contribution is probably about 35 of that.

15:53NVIDIA CEO, Jensen Wong. Billions. Billion noted. I was asked to ask you, the question's been asked and answered. Astonishingly busy day for you in the nation's capital, Washington, D.C. That's a great question. The difference between percentages and billions is a big deal. Yeah, it's in billions. The stakes are high, but we're always grateful for your time. I know it's been a busy day. Thank you for joining us on Bloomberg Television. That's NVIDIA CEO, Jensen Wong. I'm Carol Masser. And I'm Tim Stenevek, inviting you to join us for the Bloomberg Business Week Daily Podcast. Now, every day we are bringing you reporting from the magazine that helps global leaders stay ahead.

16:29We've got insight on the people, the companies and trends that are shaping today's complex economy. That's right, Tim. We're all over global business, finance, tech news, all as it is happening in real time. And we've got complete coverage of the U.S. market close. Got to say, basically, if it impacts financial markets, if it impacts companies, If it's impacting trends and narratives that are out there, we are on it. We also have a lot of fun doing it. Bloomberg Business Week also brings you the analysis behind the headlines through conversations with our expert guests. And we are doing this all live each weekday.

16:58And then we bring you the best analysis in our daily podcast. Search for Bloomberg Business Week on YouTube, Apple, Spotify, or anywhere else you listen. Check it out on your way home from work to catch up on the conversations that you miss during the business day. And on the weekend, check it out for a complete wrap-up of your business week. That's the Bloomberg Businessweek Daily Podcast. I'm Carol Masser. And I'm Tim Stenevec. Subscribe today wherever you get your podcasts.