Next in Tech | Ep. 206: NVIDIA quarterly earnings

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global Market Intelligence podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry and Company Data at S&P Global Market Intelligence. And today we're going to be discussing NVIDIA and some perspectives around a company that's had an outsized presence in this crazy thing called generative AI.

Uh, there's a lot that's going on in this podcast episode is actually going to drop just day before the next quarterly earnings call. And with me to discuss it is Melissa Otto, the head of technology, media, and telecom industry research at Visible Alpha, who's part of SP Global Market Intelligence, Melissa, welcome to the podcast.

Melissa Otto: It's great to be here. Thanks, Eric.

Eric Hanselman: Well, it's great to have you on, especially when we think about a lot of the activity that's been going on. And of course, NVIDIA has been front and center in all of the excitement that is all things generative AI. But actually, before we get started, I like to start off with a little background for our listeners in terms of what VisibleAlpha does, what the perspectives are, the draw for, and what's unique about some of the perspectives you put together.

Melissa Otto: Absolutely. VisibleAlpha, I think we're really a premium offering of fundamental investment research. We really started as a consortium of investment banks, all providing research models, forecasts, and Our technology synthesizes that to provide deep granular consensus for broker estimates data. And that is exciting because it essentially provides a view of what is priced into the market.

What the consensus view is illustrates what Investors can expect for, you know, a company to deliver from a, you know, a very granular level. Now, if we think about a company like NVIDIA, for example. One of the most important things coming into this quarter is how the company is going to guide and the guidance for its new product, Blackwell, is critical.

And what Visible Alpha does is we have a number of sources from the different investment banks that have forecasted already. Blackwell is projected to do in terms of revenues. And so, we take that, call it together, have a consensus number that gives investors a view of, okay, this is maybe what we should expect in terms of what NVIDIA may be able to deliver going forward into the next fiscal year.

And. NVIDIA has a terrific track record of beating expectations in terms of those consensus estimates, so we've seen that really for the past eight, nine quarters, it has had a solid track record of basically providing a positive surprise, and so looking at Blackwell and trying to understand From a market perspective, where it's going is an important dimension to the investment story.

So, looking at the data and what Visible Alpha provides, we can see that there is quite a bit of debate in the market when we look at their, the Blackwell new B series chips. And

Eric Hanselman: because this is something where if we think about what's been going on in the markets of late, it's not that there hasn't been.

I don't know, a certainly difference of opinion, what's been going on and I guess maybe going way back, this is one of these challenges that we always face as we get a little further out in terms of the front edges of technology, which is that this is in many cases, you know, somewhat uncharted territory, the market perspectives vary pretty wildly.

But being able to have a view of really what those takes are across a broad enough consensus, hopefully starts to allow us to take a look at maybe what all the various factors are that are in play in actually building this consensus. That's the thing I love about the data, which is actually looking, being able to take a look at each of these particular takes.

And contrast and compare is terms of where these are going, because as you're pointing out, they've got projections on what Blackwell is going to do and what the company in general is going to do. It's been interesting looking at the various sectors in which NVIDIA plays, because we focus on the Gen AI piece.

There are a lot of different aspects because they're big in gaming. They've got a big automotive component. There are a lot of different things that are in play. And granted the data center business where the Blackwell end of things going is a big part of that, but yeah, this is something where we're it's as the old Danish proverb goes, right?

Making predictions is particular, particularly hard, especially when you're making about the future. It's a complex equation.

Melissa Otto: Indeed. I mean, if we just look at the numbers for this year, we're going to get Q4 numbers next week and the company has guided to several billion dollars from Blackwell coming into Q4.

So, consensus right now is looking for 3. 7 billion in revenues. But what's amazing is that looking forward to next year, the consensus number is looking for about 75 billion. And there is a significant range of estimates around that anywhere from around 2 billion, which assumes that it's going to take longer than anticipated for Blackwell to materialize into a meaningful revenue generator.

And then we have very optimistic estimates around 137 billion. So, you can see there's a significant amount of debate out in the market, which contributes to The stock's volatility and to the way that it can potentially perform next year. So, the commentary that the company comes out and gives in this quarterly earnings call will be really critical around Blackwell.

Eric Hanselman: And it is something where, you know, Nvidia, while a darling right now, has had many highs and lows throughout its history. And I don't think anybody is starting to say that this meteoric rise is going to roll off all too dramatically, but we did see a significant. Wobble in the markets around DeepSeek.

And for those that for whatever reason, hadn't caught a lot of the furor about DeepSeek, this is an optimized model that got released as you were looking back about a month ago and had a really significant impact on the markets. As the markets were surprised by the idea that in fact, you might not need quite the high level of masses of computational power to do generative AI.

And that was really something that clearly took the markets by surprise. And had fairly broad effect in terms of what it was, what was really doing in terms of both market pricing and video took some big hits. This was really quite a, I guess, a disruption from a market perspective. And, but I guess it's still up for debate as to whether or not the technology disruption is at quite the same level.

Melissa Otto: Absolutely. DeepSeek was really fascinating for a couple of reasons. One is that the training costs for the H800 GPU hours in U. S. dollar terms. was so much lower than we've seen from other models for training costs. So, they, essentially for pre training the company noted that it was 5. 3 million dollars, then the context extension was about two hundred and 40, 000 and then post training was about 10, 000 and then net that gave us about 5.

6 million. And I think those, what was exciting about that is that the GPU hours is what makes developing AI applications expensive. And it's one of the reasons that many. Of the applications are not really coming to production because the cost is prohibitive. And if we can leverage a model or an approach that reduces the amount of.

GPU hours, the compute, it reduces the cost in a way that can enable these AI applications to come to production because they can be done more efficiently. They can be a much more cost-effective way of training. Then I think there's. An opportunity in the market for more of these applications to come into the mix and, you know, really transform the way we do things the way we think about am I right now we just don't have broad adoption and one of the reasons.

For that is that we just don't have many AI applications that are really transforming workflows and enabling people to do things differently. Now, if we think about if we turn the clock back and think about a world before we had smartphones, which is in many ways hard to imagine, but you know that those applications on a smartphone were very transformative to workflows into the way that us.

Live our daily lives and we need to see a similar type of transformation happening with generative AI so that we can start to. Identify areas of the market that could catalyze broad adoption. And that's I think that broad adoption is what will be really critical to the cycle.

Eric Hanselman: Well, yeah, this is one of those things where if we look at our data on the technology side, budget and data quality.

Wind up being neck and neck for the leading reasons for AI project failure. And to your point, you wind up with this issue of this, is this holding it back? Now, the other thing we also get is getting our data center team spends a lot of time thinking about energy impacts and a lot of the concerns about gender AI.

The other thing in this whole DeepSeek thing that came out, and I'm sure you heard a lot of this, was something that getting out one's statistics textbook from way back when, you know, Joven's paradox. This idea that back in the early stages of the industrial revolution, the idea was that because now you could burn coal more efficiently, we'd use less coal.

And applying that to the AI question, which is the, this issue of, okay, now if DeepSeek, in fact, is going to reduce the total amount of computational work we have to do for generative AI models. In fact, we're going to use less computational power. Well, Chauvin's paradox is that we got much more efficient at burning coal.

We now have furnaces that work better. And in fact, coal use went up and. Because simply because now it was easier and cheaper to use it. In fact, now we can use more of it rather than using less. In fact, you know, use accelerated, are we going to get to that same sort of thing with generative AIs? But that's a question that is still yet to be answered in all this.

But certainly, the markets had a thought about where that was heading.

Melissa Otto: If we think about some of the claims that NVIDIA has stated about Blackwell, Blackwell is supposed to use about one 25th of the energy. It's supposed to bring down costs significantly. It's one of the reasons that they have cited that demand for Blackwell is significant.

So, if Blackwell is something that could dramatically improve the efficiency and optimization of a data center. That may be the answer to your coal question. You know that.

Eric Hanselman: Well, but yeah, I'll, I'll go back to, you know, we had Moore's law for, you know, back before all of the, the GPU stuff, Moore's law, which was this idea that, Hey, we're constantly going to be making these devices more efficient.

And Blackwell just seems to be the latest instantiation, maybe not at the same rates. You know, Moore's law was this whole idea of 18 months, computational power was going to double, but it seems like we kind of forgot that optimization is supposed to be part of this process, you know, smaller, faster, cheaper, right?

But when we've got a computational vacuum that is generally AI, Hey, maybe there isn't enough capacity to feed this beast, you know, all together. I guess I look at where the markets are and some of the things that are going on. We do have, you know, NVIDIA is not the only manufacturer and we do have, you know, you've got AMD out there, you've got a lot of specialized semiconductor manufacturers, you know, the Marvell's, the world, you got all sorts of folks who are participating in this attempt to try and feed the demand for this thing.

But it seems like so many winds up focusing on NVIDIA and where they are. And I guess when you're that big a part of the market, all eyes are trained on you.

Melissa Otto: I mean, NVIDIA. is so much more than just a chip manufacturer. They are a solution provider, and I think that's a really important component to this whole story.

They are an integral part of transforming a data center from general computing to accelerated computing. through their CUDA platform. CUDA is really the secret sauce that a lot of developers work with and use in order to drive that. The accelerated computing is what lays the groundwork and provides the foundation for generative AI to even happen.

So, that compute is what that power that enables the generative AI applications to come to life and really happen. I don't think that the way that the data centers were structured in the past, it wasn't really possible to. Have the generative AI applications come to life efficiently and in a way that could be transformative to workflows and enable that productivity efficiency that we're all hoping is going to transform the world and our lives.

Eric Hanselman: Well, and you have hit on what is the foundation in all of this, which is the ecosystem piece. That's so much of it's not just the GPU itself. You have to have the ecosystem around it to allow the development community to write applications to this. It's the CUDA framework for all the pluses and minuses.

It is that foundational element. And we haven't gotten to a point at which we do in a lot of technology transitions of having an abstraction that gets us up far enough up above. And, and NVIDIA has done such a comprehensive job of building out that infrastructure of not only the CUDA underpinnings that are all part of this, but now to a point of actually building Blackwell systems.

And they've gone well beyond just being a semiconductor vendor to now taking that step into actually being a systems and dare we say it platform vendor that's making this next step. And. Being able to get to that same level for any of these other market participants. There's a fair amount of work they've got to do to be able to win hearts and minds onto their platforms to start to display some of this business.

But yeah, that's such a key part of this whole equation.

Melissa Otto: Absolutely. And it's one that the investment community is very optimistic about. Just to put numbers to that, the optimism. In expectations has really dramatically increased since January of last year. If we think about what the market is expect was expecting for 2026 back last year, they were looking for 95 billion in data center revenue.

Now they're looking for 184 billion. The expectation is nearly doubled. Similarly, for fiscal year 2027, the market was looking for 112 billion. Now they're looking for 225 billion. So, we've seen just in those two years in terms of the outlook. An increase in expectations by 200 billion, the opportunity clearly is significant.

Eric Hanselman: And we see that validated in a lot of our enterprise sentiment data that we do on the VAR51 research side, which is that the expectations in terms of spend, we've now got three years’ worth of this data of generative AI specifically. But in AI and machine learning over what I think we're now seven years in or whatever, that this has seen this rapid acceleration just in the last year.

And if, you know, enterprises are identifying, I think the number was somewhere in the mid 30 percent above and beyond what they had originally expected to spend on AI infrastructure. So, enterprise demand is there. We see it in terms of the device projections coming out of NVIDIA. This thing is moving pretty substantially.

Melissa Otto: Absolutely. And I mean, based on visible alpha consensus, looking ahead for this fiscal year, the CapEx spending is not slowing down from the hyperscalers. So, when we think about like. Google Cloud, AWS, Microsoft Azure, the CapEx numbers are expected to exceed 330 billion in aggregate this year, which is about 5X what it was four or five years ago.

Um,

Eric Hanselman: It's eye watering because we've been talking about it with our gang here in earlier editions of the podcast. It's if you take a look at that capital spend, although it does seem like the markets are at least giving that level of capital spending. Bit of a cautious look just in terms of when we're starting to see returns on those sorts of investments.

But I guess, Hey, maybe, uh, clearly the hyperscalers see this as, these are the table stakes to be able to at least get to that level of capability and how they're moving forward. So,

Melissa Otto: yeah, that's the million-dollar question for sure is when do we see an ROI, a return on investment from that high level of CapEx growth?

I think there's a lot of. Concern in the market that we're not seeing the same level of revenue growth from that capex growth. So, when do we start to actually see those investments pay off? When does the rubber hit the road? And there's some concern in the market that they may not deliver the level of growth that the investment has cost.

So, I'm not sure where it's going to shake out. I just simply. Look at the numbers and try to understand what the data is telling me, but it'll be interesting to see, you know, when, or if that return on investment actually comes to fruition.

Eric Hanselman: Well, it's most and Sarah, who was doing some of the analysis for the strategic investments for the various hyperscalers and NVIDIA for that matter was has a wonderful chart.

Of this complex set of interactions that all of the various strategic investments in terms of all of the various market players, it's basically wandering the casino, placing bets on every single table. And I guess there's that question of, well, Hey, you know, some of these are going to pay off, some of these are not.

But clearly what they're spending in capital in their own infrastructure, that stuff that is going, hopefully, it goes directly to that bottom line from a capital perspective, but hopefully that's something where the infrastructure actually starts to yield revenue that starts to move them forward. But.

Again, still an open question and ones that I guess we'll have to let the marks decide in the longer term.

Melissa Otto: Well, in this past quarter from Microsoft Azure and Google cloud, both Alphabet and Microsoft sold off because the cloud revenue came in line with expectations. It did not surprise to the upside and the market view that is a bit of a disappointment because of the level of CapEx spending that has been going on.

And now we've just seen that taken up a notch. So that's where the increased concern is coming from. One of the interesting dimensions to the whole generative AI story is around the output. And one of the critical dimensions to the tech stack that happens at the very top of the stack for these AI applications is the need for domain expertise.

So, it can't simply be a generic technology template. It has to be something that An expert is curating and working with developers on ensuring that the output is actually aligned to the persona and to the person that's using it so that the quality of the data, the quality of the output is something that can really add value.

That is not an easy task. So, I think while Blackwell may help get the costs down and improve the compute and the power, companies and organizations really need to identify How they can bring domain expertise into the AI application to make it really useful.

Eric Hanselman: That's one of those issues just in terms of maturity around generative AI use.

We see in our data; we see enterprises struggling to even fully define effective use cases. And being able to go measure output, they struggle with what they're using as performance indicators and actually understanding how useful some of these outputs are. So, yeah, a lot of hurdles that the average enterprise has to get over in order to really put general AI to work and maybe a little bit of rethinking.

And maybe we see that demand curve roll off a little bit because of this, but we'll see where that's headed.

Melissa Otto: Absolutely.

Eric Hanselman: And see what the impact is on NVIDIA. This has been great, Melissa. Uh, thanks very much for all of your perspectives and, uh, all of, uh, the insights, uh, I'll point our listeners to some of the visible alpha data on this, but I guess we'll see what the, uh, NVIDIA says and how the street reacts to it as this whole thing goes forward.

Melissa Otto: Thanks so much, Eric. It's been great to discuss NVIDIA with you today.

Eric Hanselman: Thanks for being on. We are at time, and that is it for this episode of Next in Tech. Thanks to our audience for staying with us, and thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing and events teams, and our agency partner, the 199.

I hope you'll join us for our next episode, where we're going to be looking at what's going to be taking place at Mobile World Congress. That annual gathering of the telecom industry that will be taking place in Barcelona. I hope you'll join us then because there is always something Next in Tech.