Bloomberg Audio Studios, podcasts, radio news.
This is the Bloomberg Daybreak Asia podcast. I'm Brian Curtis along with Doug Krisner. Join us each day for the stories making news and moving markets in the Asia Pacific. You can subscribe to the show anywhere you get your podcasts and always on Bloomberg Radio, the Bloomberg Terminal, and the Bloomberg Business App.
Let's bring in Bloomberg Intelligence Senior tech industry analyst man Deep Singh joining me here in the Bloomberg Interactive Brokers studio at Bloomberg Headquarters in New York. So you heard the conversation that Ed Ludlow had with CEO Jensen Wang. What did you think? What's your reaction?
Results were great, it's still a very solid growth story. But when it comes to convincing investors that there are more catalysts for upside relative to where expectations are, I think a lot of it right now is the on how the industry actually deploys generative AI. That's what Jensen highlighted that. Look, GENAI is transformational. It's changing how compute is done and everyone will use it and we are seeing that in their numbers. The data center growth is
just unprecedented. Question for investors is will we see more positive revisions? Will estimates go up by five ten percent tomorrow? And the answer is no, simply because right now they talked about Blackwell getting delayed to four Q. Obviously they are very positive on the ramp, but everything is more like pushed out by a quarter and they I mean, Jensen was very positive with ed you know in the interview about how he expects twenty twenty five to shake out,
So he's his enthusiasm is intact. It's just when it comes to estimates going up, probably investors didn't see enough.
There was a lot of conversation around the hyperscalers. These are the big cloud computing companies that have been investing in this technology, and I think the big question for the street is whether or not these companies are going to begin to demonstrate a return on investment the ROA on AI. What do we know about that part of the story.
I think with Hyperscalers, specifically, their cloud business will ramp up. And that's what he said, like he's not Nvidia is not competing with the hyperscalers. They don't intend to become a hyperscaler vendor like a Microsoft, Amazon, or Google. They're deploying within Microsoft and Google, and I think that's where probably you will see a lot of ramp up from the hyperscalers. But that doesn't mean they will continue to
raise Capex for twenty twenty six and beyond. Ultimately, the question for Nvidia investors is is this CAPEX wave sustainable? How long of a tail it has and is there a cliff three years out?
Well, you mentioned the Blackwell design, the new chip, the GPU that in video will roll out. Apparently there's some production issues right now. Is this more so on the design side, that's the sense that I'm getting, rather than production itself, right.
Yeah, I mean, look, the complexity of the chips is growing. He talked about power, how Blackwell will use the same amount of power but be able to compute more than the current Hopper architecture. So clearly they're doing a lot more on the content side of the chip and it
requires sophisticated production, which TSMC is doing. But again he came across very confident that they will ramp up four Q and then if going into twenty twenty five, Blackwell will be multi billions in revenue, so that conviction was there in what he said during the interview. It's more about, you know, will it be enough to drive more upside in terms of the incremental buyers CAPEX going up. I mean, he talked about there are more companies that are trading models.
Is just not clear enough, like who are these beyond the hyperscaler sovereigns he mentioned sovereigns he did, and those are low double digit and billion dollars. But again when you compare it to the hyperscalers, it's not very clear if the magnitude of those buyers will be the same as the hyperscalers.
End.
So you mentioned the production side, that's TSMC, the company that is the foundry for these Nvidia chips. Do we have an indication that TSMC is going to be able to keep up with demand as well?
Yes.
In fact, he was very clear that supply is growing, and it's growing in a way where TSMC is able to accommodate in Vidia becoming a larger buyer of the manufacturing capacity that TSMC has. So all the things that he mentioned of the on the call pointed to a steady ramp up more supply of chips but what it
didn't suggest is where are more upside catalysts. I mean, everyone knows about the new architecture, everyone knows about the ramp up and supply, So it's more a question of is the upside going to come from networking or software or cloud. He said they don't intend to become a cloud player. So the software business, they said they're going to end at a two billion dollar rund rate, So that upside for consensus revisions probably that is what is missing.
One of the things that I didn't hear any discussion around was the high bandwidth memory chips. And you need those to kind of work with the GPUs, don't you when you're creating these cloud infrastructure units. Yes, and we're talking about companies like sk Heinez, like Samsung. Do we know that these companies are going to be able to keep up at the same level that Nvidia is producing chips?
Yeah, And actually that's a great point. So when you look at the gross margins for Nvidia, they seem to have peaked last quarter. So last quarter gross margins were seventy eight percent. And remember this business right now has no seasonality. I mean, it's growing, you know, triple digits so obviously you can't factor in any seasonality. This quarter, the margins were seventy five percent. They said they can
maintain that mid seventy percent gross margin. But the reason why gross margin may trend down going forward is because they have to buy more HBM chips.
They have to.
Buy more advance packaging capacity from TSMC. So that's where the bill of materials does add up for Nvidia and that will start to impact their gross margins.
Great conversation, Man, Deep, Thank you so much. Bloomberg Intelligence senior tech analyst Mandeep Singh joining us here as we look at the price action in shares of Nvidia in late trading down nearly seven percent. I'm Doug Krisner, and this is Bloomberg. We welcome our guest, Mario Morales, Group VP of Enabling Technologies and Semiconductors at IDC. He's joining us from Silicon Valley, right down the road from San Jose. Mario,
good of you to make time. What was your reaction to what you heard tonight from in Nvidia.
Well, I think I'm a lot more positive than what I've been seeing after the market closed for in video. I think when you look at what they basically shared with us is that they're still growing pretty phenomenally in the data center. I think the other markets that they serve are now returning back to more normalized growth. So I think those are all positives. I think that the market sentiment was a little bit down just because the expectations have been so high for the company.
What do you make of the fact that the company is having problems with this new Blackwell chip design? Is that a concern to you?
Not really, because I look at the position and they have such a dominant position right now. Even if they're late by a quarter or two, it still will not change the pecking order in terms of the incumbency that
they continue to enjoy within the data center market. I think what you're looking at nowadays is that these chips, whether it's the Black Blackwell platform or any other that's in the market, they're very complex chips, and they're basically now putting all of these chiplets together in one very intricate package, and so you're going to have some yield issues initially when you ramp some of these new technologies.
I know that about forty five percent of Nvidia's total business comes from the so called hyperscaleers, the companies that invest in cloud computing names like Alphabet. Amazon is in there,
along with Microsoft, IBM, you could even add Oracle. I think if it's forty five percent of the business that has been really restricted to companies that have already built out their AI infrastructure, isn't it reasonable to think that growth is going to begin tapering off because those companies really at this point aren't in a position to spend more. They have enough.
You know, when you look at the cloud service provider space and how they procure silicon, it tends to be lumpy. But the investment in AI has been so strong that I think it can continue on. Maybe not necessarily only by the cloud service providers, but I think the next wave that's going to come will be from telcos and enterprise. These are all industries that are very large that are just beginning to experiment with AI and want to really turn it on in a big way across their enterprises.
So I see more growth ahead. I think the cloud service corriders might take a pause maybe over the coming quarters, but I still believe that fundamentally on a longer term basis, we're just still you know, maybe in the second inning of what we're seeing today in terms of the amount of investment that we're seeing in AI.
So Nvidia designs that they are manufactured at a foundry that's operated by TSMC. That's essentially what TSMC does, they're the manufacturer. Are you concerned that TSMC may not be able to keep up with demand.
Yeah, I mean that's continued to be the theme right for over a year now. I think last year we saw a correction in the semic metor space, but when you look at TSMC, they tend to buck the trend.
They're a very.
Strong player in the founder space. The leader they control over sixty percent of the market. Where they're investing heavily now is in advanced packaging. So when you hear about costs, all of that means is that that's where they're seeing the big investments that they need to make in order to serve companies like Nvidia and Broadcom and AMD and others that are now beginning to use more advanced packaging in order to continue to drive integration.
So the GPUs that are designed by Nvidio manufactured by TSMC, they have to be paired with the high bandwidth memory chips do they not, and those devices or chips are created by companies like Samsung and s k Heinex in South Korea. How are those companies How have they been able to keep up with the demand. Should we be concerned about maybe a little weakness on that side.
Well, we're not seeing the weakness. In fact, since probably September of last year, the memory companies began to change their operations in order to get back to profitability, but demand returned, especially from the data center, and today we're expecting that both DRAM and NAND will grow significantly from a revenue basis, close to seventy three percent in terms
of revenue growth, so very phenomenal growth. A lot of that led not just by high bandwidth memory, but also DDR five that's being used in the flagship smartphones and the upcoming aipcs that we'll see later during the holiday season. So I think in general that the demand for memory will still remain robut this year and likely into twenty twenty five.
One of the things that is simply stunning when you look at any area of high technology is the rate of change, how quickly technology evolves. Give me a sense of what we're seeing now with not only a company like in Nvidia, but broadly when it comes to companies that are involved in artificial intelligence. Is some of this equipment going to be obsolete a year from now? Do you think?
Well?
You definitely. When you look at the SEMAC, nintor space, it's always followed Moore's law, which means that that performance or power changes every eighteen to twenty four months. Now you add AI to this, and when you start looking more at the large language models, they're actually changing every three to four months. So we've seen an acceleration of change that's happening. And you look at companies like in Nvidia or AMD, they're trying to go and keep up
with that change. And I think that's going to be the big challenge for these companies because they've been operating with More's law in mind. But these but the market
for AI is moving so much faster. So now AMD and Nvidia and Intel now have to basically introduce products every single year now, and you're going to have some hiccups because of that, because the cadence is now changing in order to keep up with this demand and this appetite that the world has for AI, especially around infrastructure.
What is your sense of the competition that in Vidia may face in the future. Right now, there is clearly very little. The company is essentially a monopolist when it comes to these GPUs. But can you imagine a world where Nvidia begins to look over at shoulder a bit?
I think so. I think over the coming year, you're definitely going to see AAMD stepping up their game. They you know, back in computext in June they announced an annual cadence for their own accelerators with the M three hundred beginning to fray, and then you got the three twenty five and three fifty. So you're going to see AMD now introducing an annual cadence of accelerators in order
to try to catch up to Nvidia. But I think in Nvidia's own customers continue to chip away little by little in terms of the instances and services that they're turning on with their own silicon, And so when I look at the competition, I think it's going to come more so from the customers. Initially, so a company like Google and Microsoft and AWS all developing their own chips and using some of those ships with their infrastructure, and
then on a longer term basis. Broadcom is still a significant player here that not a lot of people talk about. But when you think about a lot of the designs of the APIs that are today serving the cloud service providers, a lot of those acs are designed in collaboration with Brock them, so I would call them the number two player.
And then AMD coming in being able to leverage their CPU entrenchmen, especially in the data center, and hopefully that will help accelerate their own entrance into the GPU market for training and infancy.
Mario very quickly in about thirty seconds. A big semiconductor industry a conference next week in Taiwan. What do you think the big issues are for the semiconductors supply chain in Asia right now?
I think that we're still seeing a slower recovery in the spending and I'm talking about capital spending of the major companies. Even though TSMC has increased their numbers, the rest of the industry is still a lot more cautious around spending. So that's going to be one of the key things. How much of the tools are still making their way into China. That is also becoming something that
governments are watching more closely. And then how long can we continue to sustain the growth that is needed to believe it there billion dollars.
Mario Morales, this is Bloomberg. We bring in our guest, Dan Newman. He is the CEO of the Futuram Group, joining us from Austin, Texas. Dan, pleasure to have you on the program. I hope you had some time to go through what we heard from Nvidia tonight. The stock, as I mentioned, was down more than seven percent in the late US trading. Do you think this is an overreaction?
Yeah, thanks for having me, and I have spent some time and actually listened to a really compelling interview that was done Bloomberg with Ed Ludlow. I think the overall reaction is a little bit of what I would have expected based on them just doing enough. And as crazy as this sounds, just doing enough was beating raising and being really solid and consistent in what the company is delivering.
The problem is, over the last several quarters, we've come to expect these massive beats, huge revenue upside, these guides that look phenomenal, and this sort of endless upgrade cycle and this just was good. And you know what I said coming into this, it has to be great or there's only.
Downside the price to perfection story. I think we can agree on that, right. But are you troubled at all by the word that in Nvidia is having problems with that new Blackwell chip design.
I'm not really troubled because it seems he has the situation under control. He talked about the update to the mask, which was going to improve the yield, which in the in the near term means probably an uplift from Hopper. These Hopper chips are still very significant, and they're still selling in volume. Amazon and Microsoft are still building out significant infrastructure with them, so is x and Tesla and others.
I think where this is going to head, though, is when does that ramp, that one year cycle continue to drive these triple digit growth rates. And we're starting to
see that slow down. And I think whether we see this ship into the fourth quarter or even if it falls into the first quarter, what we end up having happen is we've got an entire world that's kind of watching this stock and saying triple digit revenue growth, triple digit earnings growth, and now we're starting to see that quarter on quarter you're on youar it's decelerating and then bigger questions about AI digestion and when does this start
to really find its way into companies like Salesforce, as Numbers and others that reported today that didn't get as much attention.
Yeah, you're right about that. It's a good point too. But let's get back to Nvidia and the approved fifty billion dollars in additional stock buyback. Is that a good way to kind of repurpose or redeploy capital.
I think if the company is looking at the situation as hey, there may be some some hesitancy, there may be some sellers. We saw some selling pressure into the numbers today. The company wants to be armed, to be able to lower the flow and to be able to take advantage of a lower share price as it expects that this Blackwell and then followed by Rubin will continue to drive this acceleration. You know, we did hear from Mark Luckerberg center of a Chai, you know, Satia Nadella.
There was really no indication that they plan to slow their capex spending. And that means good things for Nvidia because despite the fact that all of them are making investments or have been making investments in somewhat competitive offerings. In the current moment, these companies are building the bridges and roads and tunnels with Nvidio, and if they want to be competitive, they don't want their Blockbuster moment, they don't want to have their BlackBerry moments. So I think
they are going to keep investing. And I think that bude really well for Justin Wong in video.
So I hear what you're saying when it comes to the cloud computing companies, those so called hyperscalers talking about their intention to continue with the current level of cap x. But won't that largely be determined by whether or they or not they get a significant return on their investment. The so called ROI on AHI has to be the key, right.
You would think so. But if you listen to Mark Zuckerberg's comments, he was pretty clear that he was willing to bet even if that ingestion period, even if that consumption period is longer than we expect. And we know that he's done that before with Reality Labs, and the market sort of pushed back at him until he had to pivot on his as capex investments on that side.
But AI is not virtual reality. It is very clear that this is going to change every industry from autonomous driving to drug discovery to these large language models that are going to change software and the way we sell, the way we buy, the way we watch. And so when you put all these things together, I think these companies are willing. They have great balance sheets, most of them are strong cash positions. They're willing to make the bet. But I agree with you, I think we need to
see more measurable returns downstream from these software companies. Who else is making money besides Nvidia and these cloud companies have not been really clear. I think Azure amy Hood gave the best numbers so far, but you're talking to single digit percent about Azure revenue.
TSMC is obviously the foundry that produces these chips. I mean, are you confident that TSMC can keep up with a level of production required.
Well, TSMC is continuing to talk to how it will expand its capabilities. We've heard about some of the you know, the availability of its three danimeter being really sold out through twenty twenty five. We of course they're hearing about coosts, you know, the packaging technology, and that was part of what was going on with some of the potential delays. This new cost l technology and TSMC has the market.
They have a huge percentage of this AI movement, and of course a lot of Nvidia's business, and they're going to have to invest substantially. TSMC is a clear winner here, and you know, while we are waiting on some of these others, the sks and microns, and they're building the HBM side this partner ecosystem, and Jenson said this on the interview with that ed it was very clear that this supply partnership is monumental in terms of the company's
ability to continue to hit its growth. TSMC has given us no reason to doubt its ability. But this technology keeps getting more complex and that puts a lot of stress on TSMC to deliver well.
You mentioned the high band with memory chip manufacturers sk Heinix, Samsung in South Korea. Are they kind of the weak link in this chain, which is to say, if they don't keep up with the level of output that in Nvidia is striving for that the overall move in the build out of some of these AI infrastructure models, I mean that it's held back a bit.
I think this continues to be the reason why whether Blackwell's a little bit delayed or not, they're going to produce as much of the hopper as possible. The overall situation as it pertains to memory is they're also constrained there. So you heard me talk about three nanometer constrained, coosts constrained,
and of course HBM constrained. So right now you are literally seeing these five or so major hyperscalers plus global hyperscalers and Tier two's taking up the vast majority of the demand for all of these major parts of the supply. And you know, Jensen talked about it being around forty five percent coming from the three or four biggest, but they are taking most of the demand and until we
can expand capacity. So it's on Micron, which is investing, it's on sk and Samsung, all of which are investing. But it can't happen fast enough. This does take.
Time, given the export controls that the Biden administration has put on some of this advanced technology from Nvidia to the Chinese market. Is it fair to say that the US is far out in front of China when it comes to AI or are you concerned possibly about Chinese advancement in this area and maybe at some point it becoming a greater rivalry.
I am always concerned about China's willingness, will power, capacity and capabilities to build and deliver even in the most difficult situation. Of course, the US has been incredibly innovative, are fabulous leaders, not just in Nvidia, AMD, of course, historically, Intel, Qualcom, these companies have all been tremendously important to this innovation wave. China is doing what it can to follow suit. It's been somewhat successful in mobile devices, in PCs. We've seen
it with Honor and Huawei. We are hearing that it is having some level of success building competitive products to the tuned down in Vidio chips, the B twenties h twenties, that they could be able to compete in this particular space. I never would rule them out. I think that would
be a tremendously large mistake. Having said that, in Vidia has a substantial lead beyond everyone else right now, and then others like AMD and like what the hyperscalers are building is very advanced, and I do think China has significant challenges, but I wouldn't doubt their ability or resolve to find a way.
Well, leave it there, Dan, thank you so much. Always a pleasure Dan Newman, he is the CEO of the Futurum Group, joining us from Austin, Texas here on Daybreak Asia.
This is the Bloomberg Daybreak Asia podcast, bringing you the stories making news and moving markets in the Asia Pacific. Visit the Bloomberg Podcast channel on YouTube to get more episodes of this and other shows from Bloomberg. Subscribe to the podcast on Apple, Spotify, or anywhere else you listen, and always on Bloomberg Radio, the Bloomberg Terminal, and the Bloomberg Business App.