00:00Joining us to break it all down Tony Wong portfolio manager of the T. Rowe Price Science and Technology Fund
00:05and across the firm in different guises you know T. Rowe a big investor of NVIDIA's.
00:11OK let's start ninety one billion dollars plus or minus two percent in the second quarter. It was like almost
00:16a whisper number for the sell side and the buy side but not enough.
00:22You know and why is that. Hey look I think that you know the growth at this scale is just
00:29unprecedented. I don't think the market's ever seen anything like it.
00:32And I think you think about traditional semi investors you know the playbook is to sell this type of growth
00:39right because it's not sustainable.
00:41You know the margins are peaked. But I think if you really look at what's going on at the driver
00:46of the driver like what's going on the end of man.
00:49I think it's like really phenomenal in terms of agentic AI really taking off the time of work or like
00:57time of task is expanding instead of being one shot or the you know the agent working for a few
01:03minutes.
01:04I think we're going to go towards like months of work and that's requires like a lot of compute to
01:10have an agent go off and like complete a task and think and be persistent.
01:14And then this other thing is I think that scaling laws continue to hold like the frontier models are getting
01:18better.
01:19The more compute you throw at it the better they get. And you actually save money by being on the
01:23frontier because it doesn't go through so many rabbit holes in terms of trying to complete your task.
01:29So I think if you look at like the end of man it's actually very encouraging and I think we're
01:33climbing a wall of worry here as a result.
01:36So the thing that I was trying to digest throughout the call with the data points that Jensen Wang and
01:41Colette Crest put forward is evidence of that right.
01:44The scaling laws and the dollar per token and they go to this one trillion dollar figure which I think
01:49they clarified is Blackwell Rubin and it's for calendar year 2025 to 2027.
01:55And it's basically a backlog. Right. But the main overarching point is that they would argue that NVIDIA's own growth
02:02is trending ahead of even the hyperscale CapEx growth.
02:07How do you model for that. Like is that the math that you would do. Yeah. One of the things
02:12that was really encouraging I think of the quarter is that I actually broke out like hyperscale versus the enterprise
02:18and sovereign.
02:19And so you know I think the bear case is that you know NVIDIA can outgrow the hyperscale CapEx numbers
02:25but there's all this new TAM that's emerging that's not hyperscale.
02:29And I think they continue to do well in hyperscale but there's also new areas of agentic enterprise adoption.
02:35I think you look at talk to financial services firms. They're adopting a lot of a lot of this end
02:40demand that NVIDIA is producing.
02:42And so I think that plus robotics that's on the on the to come in the future. I think that
02:47you know allows them to see you know not just not be bed bracketed in this like hyperscale kind of
02:53compute you know paradigm here.
02:57Tony slightly unusual situation but I spoke to Jensen at some length on Monday prior to earnings but actually a
03:04lot of what he said about the demand supply equation was very prescient salient.
03:09Just listen to what he said we have the largest supply chain in the world our partners have done a
03:15great job securing supply for us and so all of the pieces go together the silicon photonics is lined up
03:22everything is all lined up it's just that the demand is much greater than the overall capacity of the world.
03:30If you are a student of technology history finding yourself in a situation where demand is vastly outpacing your ability
03:37to supply even in a world where supply chains are doubling or quadrupling per annum which is what he went
03:42on to say that's been an enviable position but right now it's not moving the needle for this stock.
03:50Yeah. Yeah. I mean I think that you know the market is looking at relative earnings growth and where the
03:56bottlenecks are. So capital is flowing there.
03:59But at some point you know you look at the multiple for Nvidia it's like really attractive. And in addition
04:05like there's there's a point where you know for the growth it is like tremendous looking at a peg
04:10ratio the expectation. So if the growth is durable you know I think you can see multiple expansion actually from
04:17these levels and that's pretty exciting for for holders.
04:20And in addition it is the platform that a lot of this AI inferencing is going to be built on.
04:26I think inferencing is going to be much bigger than training and training is also still growing because scaling loss.
04:32In addition I would say that like at this valuation you know they are doing a capital return program. And
04:38so I covered Apple and Nvidia for T-Row and I was super lucky to get that.
04:43I saw like you know what Apple did with the return program. You know it didn't like you know re
04:48-rate the stock from day one but over time consistent capital return at that valuation really expand the multiple.
04:54I think with Nvidia it could be a similar story where it becomes less cyclical more durable as a result.
05:01So I think it's you know they're doing the right
05:03things here with the disclosure as well as like you know with the capital return program. I think it's just
05:08a matter of time here.
05:09I think right now Nvidia is trading at like 22 times forward 12 month earnings right. I think I'll double
05:16check it. Something you just raised is really
05:18interesting which is basically use of capital. So the frustration with Apple is like do something. And maybe that that
05:26story is changing with
05:27Apple in the handover to John Ternus. But Nvidia has gone out there and invested in the ecosystem in small
05:36increments. Two billion here two
05:37billion dollars there. How does Tony Wong read that strategy of investing. I think it's really smart. I mean I
05:45think that when you're at the
05:46frontier of technology you have to build the ecosystem and bring up the supply chain bring up the partners. And
05:53you know
05:54Nvidia has been really smart in terms of using their free cash flow to extend that. It also builds their
05:58ecosystem as well as
06:00furthers the you know the technology frontier. So I think it's it makes a lot of sense. And you look
06:05at the deals they've done in the
06:06private markets they've been pretty good. So like great for shareholders great for building the ecosystem and they have excess
06:12cash
06:12still. So I think that just goes back to like they are like this single architecture platform and there's tremendous
06:19leverage on
06:20it. You see where the margins have gone. And so that's been their whole whole you know bad from like
06:24day one I think is that
06:26they're an ecosystem company. And I think you're seeing it play through here as they build it out continuously. We
06:32actually have a VC on later in
06:33the program whose portfolio companies you know are so tied to Nvidia and we'll have that conversation. Just to close
06:38the loop by
06:39the way Nvidia does trade 22 times forward earnings but historically it's like nearer to 34. Jensen has this equation.
06:46More compute equals
06:48more tokens equals more revenues. Do you see those revenues coming out the other side for all of Nvidia's customers?
06:55Absolutely. Yeah definitely. I mean you look at like the cloud demand. You look at pricing for GPUs. I mean
07:02these are like old GPUs.
07:03Pricing is holding up phenomenally well. And you know you look at like the enterprise demand inflection that's happening
07:12and you see it. You know we see in our own company we're developing AI solutions and it's eating up
07:17a bunch of tokens
07:18where you know we see the promise. Obviously there's experimentation. I think you've seen that like come through.
07:24Just real quick like on the socials that's what people just go nuts over. Right. H100 H100 pricing right now.
07:31Like are you tracking that? Oh yeah. Yeah. I mean I think you think about the bear case you know
07:3724 months ago it was that like
07:39oh like the obsolescence of Nvidia GPUs is like three or four years. Well first of all the warranty on
07:44these
07:44things are three or four years. And then also you know you have that depreciation. And so you're able to
07:49like
07:50adjust the pricing you know. But even so like the demand's been so good that you know we're seeing more
07:57value coming out of the
07:58older GPUs. And even A100s if you were using them. Even A100s. If you're out there running workloads on A100s
08:04give me a call. I want to know.
Comments