NVIDIA beats earnings estimates as Doug O'Laughlin flags GPU shortage and power bottleneck as the next big constraint

that. So congratulations to Riley Walls on his move to open AAI. Let me tell you about graphite code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. And without further ado, we'll bring in the birthday boy. Doug Olaflin is in the reream wedding room. Let's bring him in to the TVP Ultradome. Beautiful shirt.

Happy birthday to you. Happy birthday to you. Happy birthday to Douglas. Happy birthday to you. How you doing?

Hey, happy to be here. Uh I guess on my birthday, nothing better than Nvidia earnings, which hasn't come out. We're going to be live reacting.

Yes, we'll be live reacting

and they planned that of course years and years ago. They

on your birthday.

They knew that today would be the day.

Uh they went back in time and they told my mom to uh to to to get it done today. Yeah.

Just for me.

Yeah.

Um how you guys been?

We've been good processing the something big is happening. Then the Catrini article. There's been a lot of

sort of doom, but also there's glimmers of boom in there. There's it's mixed.

I feel like you've generally been a Catrini defender kind of saying, you know, a lot of people were just like this is bad and it should never have been published. I think your view was

there's some good stuff in there.

There's some It's a good thought exercise.

Yeah.

Yeah. Well, here's the thing is like it went super viral for a reason because it obviously resonated. Um I don't think really crappy ideas resonate that hard. Um and and I think honestly a lot of the things that I'm most concerned about I would actually align with it. Um my biggest concern is that we're going to print deflation and he kind of talks about that like hey one data center does all the knowledge work.

Um the history of how this works usually isn't quite that extreme but that was the point of it. Here's my extreme think case. What I was kind of shocked about and I feel like honestly I bet you would agree just like the level it resonated with. Um

yeah,

it it was like covered on Bloomberg almost on a daily basis. Uh so if you're like a pro whatever that does a push notification, it was like Satrini says and Satrini co-author and all this stuff. I was like, "Oh my god." Um

it really broke through.

That was Yeah, it really broke through.

Yeah. Unpack the deflation question a little bit more. So technology has been deflationary for a long time. TVs are getting cheaper is the line. You can see that if you look back at inflation, you see healthcare and education inflating while basically everything that's on like a technological moors law style curve is getting cheaper over time. The the bigger like broader broad deflation question is can we move some of those or will some of those inflationary categories move into a deflationary territory? What are you thinking? Is is education, healthcare at the top of the stack? Is what do you mean by knowledge work becoming deflationary?

So, let's just put it this way. The uh the ghost GDP article, right? Hey, uh let's just kind of

like do like an index, one index versus the other index. Uh we're going to label it at $100 for all these like knowledge work widgets. And um let's just see what the cost of that looks like over over time. year $100 with a human and maybe year it's $102 maybe they get a little bit smarter so it goes down but like okay what if the agent gets 20 times better years year two the the index of knowledge work is going to be $20 that's that's just deflation right um the entire world is effectively just like spitting out TVs that are worthless in three years and that's kind of scary and I think is a big deal I think the thing that is the most resonating with the whole piece is that clearly if if this is going to be as big as many people think it will be um you know cloud code maxi here um then what happens is that you're going to quickly disrupt so much of society that the government and like companies and everyone needs to start caring very quickly I don't think 10% unemployment is going to happen uh the world works just slower than that but I I clearly think that part of your framing is like when we had uh technology rollouts throughout the last you know few hundred years. It involved like heavy machinery that had uh that took a lot of time to diffuse in part because uh the the machinery needed to be built and shipped around the world and spun up. And then you have with AI it's like okay the internet is is the greatest distribution engine in history and as technology advances it can like be live in somebody's office or with somebody's company effectively instantly which creates and and and with like uh let's say uh the printing press it was you know or cars or or electricity it was like a much more it was forced to be a much longer roll out.

Yeah. There's just like for example the railroad piece I wrote about, dude, it took 50 years to put all those rails, right? Um, historically, let's say this AI thing would be like a historical atoms thing. It takes 50 years of installing the AI widget at your house in order for you to get AI. Um, versus now, you just press a button, you download it, and it's there. And so the pace of that disruption is just so so much quicker. And so if it looks like, you know, if if it's as big of a deal as people are concerned about, then essentially you're going to broadcast deflation everywhere all the time all at once. Um, and it's going to be almost costless to get there. And also it's just it's going to get relentlessly better. Um yeah just and then

how do you how do you how do you square that though with the fact that right now today

AI is the best at coding that is where it has the most traction within knowledge you know knowledge work and yet I have I have not gotten a call from a great engineer saying hey can you help me land a job I'm unemployed and you see this in Citadel put out a response saying that that software engineer job listings have gone up uh year-over-year. And so it's so hard for me to square these two things where we could be this in this in this uh you know doomsday scenario for white collar work.

Yeah. We're not seeing that at all

with software engineers. Now, some people might debate me and say, "Well, oh, I just graduated

school and and yeah, new grad opportunities might be lower, but nobody nobody has bought when when some of these larger uh tech company CEOs have laid off a bunch of people and said, "We did this because of AI." It's like nobody nobody's actually buying that. It's just a a nice story. Yeah, I think um that's going to be an interesting thing because I think it will be politically and broadly unpopular to say, "Hey, I just laid off 10,000 people because of AI."

Yeah.

Like we are humans at the end of the day and you know, you do care about what your neighbors think of you. And just being like, "We fired everyone because AI is better than uh whatever." You're like, "Okay, well now you got to get you might as well hire a five person security guard, right?" Um there there's a little bit of like being an [ __ ] So I think how it actually works and you're looking at um and I agree with that comment because there's this really great um tweet that's like oh it's crazy in the agentic coding world pretty much it pegs the human CPU at 100%. That's how I feel at least is I'm doing more work than ever but I'm like literally working harder than I've ever worked. The reality is I feel like there is a productivity uplift but I guess we're all in the sugar high where you're able to do so much work that you're just like going around like crushing it. So these software engineers, yeah, you don't need new grads. And essentially, the person who has the seat gets to uh, you know, print more, let's just use the printing press, right? Instead of writing, and you unemploy all the new writers, but you're sitting there just printing all day. And so it really is great for the install base of people who've been doing it, but very terrible for net new. And I think the net new is where you're going to see this, meaning like new grads, new people entering the information uh the information services. That's where I think you're going to see most of the carnage. People are not just going to be firing people out the gate. Um it's going to be like a layoff or or a firing typically occurs weeks if not months after an executive has decided, hey, this you know, we we need to make a change here just because no, it's the thing that everyone hates doing. How do you how have you been processing? It feels like the people that uh the Catrini piece resonated with the most are also the ones that have been saying that the AI is a bubble and there's too too much investment going into this and it's all going to be a zero. uh it just seems like two ideas that are difficult to

you know I feel like there's a certain uh aspect of haterade where when you're a real haterade you're going to find um whatever whatever flavor supports your thesis we find this at semi analysis man when we say something positive or we perceive as positive people take it as negative or really honestly they take it

we've we've had we've had that too

if if there's a company that people don't like and you and you say something positive about it. It's It's like people respond like extremely

emotionally to that.

Yeah. And so I think Catrini made the perfect ammo at the perfect time.

And and like honestly if you look at the stock market um I'm sorry I keep watching uh for video over and over. Um uh if you look at the stock market um I think right now it's been a really interesting year to date because like stocks are maybe like whatever 2% off alltime highs. Maybe they're all the way back or they're pretty I forget what the like year to date looks like, but underneath the hood in software, it's been very very very painful. I think there's a lot of parts of the index that are in a lot of pain right now and pretty much you're just like seeing these crazy rotations underneath the surface and so people are very like angsty. There's no other way to put it. Um if it it's thing big things are happening in the stock market. the the the idea that uh like oh you're holding a you're holding a name and it hasn't nuked 15% yet you should get out because some there's going to be some there could be a blog

partnership or blog post and and what what's the there's not a lot of ROI on

how have you been processing like the the the moes in software companies like uh I like I've I've always understood like the SAS apocalypse like that narrative of course you know we got to debate timelines and maybe these things are just shifting to value stocks as opposed to growth stocks. But, uh, the stuff with network effects, the Door Dashes, the PayPal's, the regulatory modes, like it just feels like there's a lot of companies out there that are being more broadly punished. But is there something I'm missing there or or do you think that uh it's just excitement and people rotating into other stuff that they might be learning about like memory and energy and

semis? I I don't know. Um, but I definitely think that that aspect is definitely there. Like the ones that are like really shocking to me is insurance brokers.

Um, if you've ever if you ever follow that space, like once upon a time as a hedge fun analyst, I followed AON. It's like one of the most boring spaces of all time. Uh, real estate services.

Uh, real estate services.

No, like CBRE, CBRE, um, essentially an insurance broker. You're having all these things, these network uh, businesses that are just effectively being like, yeah, in fact, it's going to be nuked. Um, I think the hard part about that is it's it's like it's complicated. I think there's a lot of advantage for number two to defect really quickly. That's probably the most valuable thing you can do. For example, we've written quite a bit about this to like our higher paid tier service stuff. Like Walmart defecting to Agentic Commerce makes a lot of sense, right? Because they're not Amazon.

Um, everyone that isn't number one should be defecting to win market share and that makes a lot of sense to me.

So like it it wouldn't be Door Dash. Door Dash would probably put up the, you know, put up the walls, do their best to like have exclusive products and then someone else will vibe code something on the margin and essentially try to win market share by

But at the same at the same time, Dra was on the show and his stance was, yeah, I don't really care as much about my ads business about making sure that I'm just everywhere. And so he was seemingly pro- agent even though

ads business is is real. uh have you been processing uh uh agentic commerce like it feels like this should like the tech is there

and but like consumer adoption takes time you know yes you can touch up a photo with gen AI not everyone does um like what would what does good look like because we're seeing stats from Shopify it's like still in like 0.01% Oh, 1% of checkouts are agentic and I would expect that to 10x or 100x, but even then you're talking about like 1% of e-commerce. Like we're still talking like really small numbers. Um like how fast do you think it ramps and and what do you think happens this year?

So I think the place to watch um and it's probably the most interesting place is China because you can argue uh the the leading e-commerce adopter has always been China. They were doing Door Dash way before it got hot in the United States. They were doing e-commerce way before it got hot in the United States. And we just saw this.

Yeah. Live streams, right? Like like dude uh Tik Tok and Doian was always um always been on the bleeding edge. Like I would argue for like global internet online consumer preference behavior. China is the leading indicator. Like America is a lagger boomer, you know, like we just don't we don't buy like we used to. Okay. The consumer is not quite um quite as young. We'll just put it that that way. Um, I think what's interesting is we just had this giant incentive out of like I believe it's uh I want to say it's Tencent Tens did this like um I I'm now I'm probably going to misspe the the the bubble tea promotion. Okay, they did like 10 million or whatever bubble tees. And so I think there there's an example that we finally have the first wave of people possibly being able to adopt it and that's going to be like all eyes on that for if adoption actually kicks off there. Like that's what happened in 2015 to kick off mobile payments actually during the CCT uh CCTV uh gala they essentially gave away like payments and and they incentivized people to use the mobile wallets and that was like the beginning.

Okay. So I think that same analogy is probably where people need to look towards for what um what could be the adoption curve for a gentic commerce.

Is that earnings? Do we hear that earnings came through? No, I'm I'm just I'm just I'm

I'm just locked in.

I'm I'm locked in, bro.

We'll have We'll have this effect.

Yeah. What? So So, uh h how do you imagine Agentic Commerce actually rolling out in China? Because there's a whole bunch of AI labs, uh you know, DeepSeek, Highf flyier, like they don't have a do they have a big consumer footprint already because it's they don't have a lineage there. So you might expect uh you know agent commerce to be driven more by the platform that already has a billion daily active users over there. Uh but how do you see like the the stack uh piling up over there?

So I think the best example of this is Alibaba which actually is an e-commerce company. They're like you know the Amazon edition.

They have Quen.

Yep.

They have a delivery app. They have like an inventory app. They have a fully vertical in fully vertically integrated stack

and they have AI on top.

Mhm.

They're going to make money if people just use their product.

Mhm.

They can get transaction fees. And so that's how I kind of think about it. I mean, in a perfect world, if Amazon was not bad at models, they would be crushing, right?

Yeah.

You'd obviously be like down on Roffus.

Don't talk down.

Just because it took down AWS once or twice or three times for hours doesn't mean they don't got it. I I actually want to talk about a conspiracy theory because I feel like this is Have you been noticing the instability of public clouds?

Yes.

Am I crazy? Am I crazy?

I don't think you're crazy.

No,

it's either there's two there's only two things it could be.

Yeah.

Um vibe coding.

People are just pushing crap on prod and they have no idea.

Probably true. Number two, CPU shortage.

I think it's a CPU shortage. Wait, the the you don't think there's like a third if we're wearing the tinfoil hat, which is the world's been very unstable and there'd be

more cyber attacks and stuff. I mean,

maybe that's actually that's a good one. Um, but I'm really surprised to see all three at the same time. And then like my favorite, probably the most public version of this is the GitHub instability. Yeah,

dude, GitHub is so unstable right now.

Interesting. Yeah, the I saw the uptime numbers. It was like below 99% which is really low considering

No, dude. It's like 90% in

90%. That's really low. I mean, people are pushing a lot of code to GitHub. So, yeah. Walk me through walk me through the CPU shortage because is that just CPUs that are being used to scale new like EC2 instances and like normal like you need a Linux instance and so you need a CPU to provision or is or is the AI boom sucking CPUs in like a GraceHopper demand like they're the the CPUs are getting dedicated to AI server clusters like where are the sup what's the shape of the shortage basically

I think the shape of the shortage is partially on RL Well, because if you want to do an RL gym, meaning like you want to have an Amazon.com for your you have to you have to literally simulate just a lot. That's one real demand driver. Um I don't know how I don't know how to quantify that, but clearly they're out of it. Number two, I think um I do think all the code have you seen the charts like I think it's f.com of like all the new apps coming online, they they require infrastructure to run. And um I had a third one. Oh, the other thing that's interesting is it's also a little bit of a of like a lapping supply chain thing. So, the last time we bought a lot of CPUs was in 2020 and 2021. Historically, you depreciate the the CPUs on a 5year cycle. And so, after 5 years, you just like literally throw them away.

It's 5 years.

So, all of the infrastructure that we purchased and we've been trying our absolute best not to purchase any new CPUs and spend all that money on GPUs. So, they hadn't been investing this entire time and now this slight demand curve comes up and that's enough to essentially sell out all the CPUs. I think that that's probably part of it as well. It's going to probably be this multi- problem thing, but it's like one of the more interesting like conspiracy theories like YouTube went down. I I don't I don't remember a time in like the last 5 years YouTube has ever gone down.

Yeah. So,

yeah. Yeah. Yeah. That is crazy. uh take us through uh the effects of the CPU shortage in the TSMC context because TSMC seems to be like a crazy battleground uh Apple Nvidia and then you'll talk to a new semiconductor company like we had MaddX on the show yesterday and he's like yeah I'm getting line time at TSMC and then you read Ben Thompson and it's like well they're not really investing in capex and it feels like uh is is TSMC a bottleneck she shortage as well. We got Nvidia earnings. What happened?

They got a revenue beat.

Okay, we will jump back to TSMC, but they beat revenue. Uh let's review and read through

67.4 billion against an estimate of 66.2 billion.

We have some breaking news. Nvidia has announced earnings. U let's take

Wow, you guys are faster than me. I'm on Bloomberg.

Nothing liveer than live. Um, here we will all read the earnings and I will tell everyone about Plaid. Plaid powers the apps you use to spend, save, borrow, and invest. Securely connecting bank accounts to move money, fight fraud, and improve lending. Now with AI

and back. Do you got anything yet, Doug?

Am I crazy? I don't have anything.

Did maybe Tyler just sent this to us. Did Did they just

refreshing Twitter? It's possible somebody's just putting up is just engagement farming

maybe

and uh trying to

they might have gotten us. They might have had us in the first half.

Well,

we shall see.

Got gone.

Anyway, let's go back to TSMC. So the nature of the TSMC bottleneck with regard to CPUs, is that an important factor or is there more fab capacity across Global Foundaries, Intel, Samsung to sort of meet that CPU shortage?

So Global Foundaries is nothing.

Um they don't have a leading edge chip. Uh so they will not have any CPUs there. Intel is actually the real

winner, I think, because essentially like TSMC is locked up. Everyone's fighting for space at TSMC. Like Nvidia has the most. Um all the co-ass is accounted for. All of N3 is accounted for. All of N2 is accounted for. Essentially, they're completely sold out. They're going to invest a lot more, but like you know, get in line. It's another year. Um so whatever is left over kind of goes to the other foundaries and that's really two companies. That's Intel and that's Samsung. I think Samsung right now probably makes way more money if they spend it all on memory. So, Intel really is the swing provider of CPUs in the world. And that's like kind of amazing. Like, I'm kind of my my belief in a a sweet karmic Intel victory is all the CPUs that can't be made at TSMC because the accelerators are being made instead go back to Intel and they just make them, but they don't design them.

Yep. Uh make them but don't design them. So, they would be a second source for Nvidia potentially as well.

Well, like it's not just Nvidia. It's not I think I think Nvidia is going to get their I think Nvidia is going to get their application.

Oh, at TSMC. Okay. So, they don't need to TSMC.

Okay. Because there was a theory about like, hey, maybe twist Nvidia's arm, get them to dual source at Intel, and that gets the flywheel going a little bit more just as like a give. I I guess but I and there's been some conversation about the MV MVL link with um with Intel specifically co-ackaged but I I I think it's going to mostly be um Vera like I think it's not going to be I I think TSMC is going to give them the capacity.

Um I think there's a whole world of chips that are not that like think about Graviton um or the Axion project at Google or all the

ARM explain that Google project I haven't heard of that. Um they there's a custom CPU.

Okay.

Yeah. So like all these

don't they have like a YouTube chip? They have like a number of custom CPUs, right?

Yeah. V uh yeah VCU I think it's a video encoder whatever unit but that's like really dimminimous suit. It just doesn't I don't think that adds up.

How important is the is the Gro cerebrus these faster model on a chip companies and does Google have an answer for that? I Okay, so to be clear, semi semi analysis is house view wasn't that it was like the most important thing and we've always we you know we we have that chart about batch right the fact that hey doing more for everyone is much more valuable clearly there seems to be a little bit of a niche market and a niche market demand and so cerebrus and LPUs definitely have some kind of demand pull and I think we will see some kind of announcement at GTC but like you know definitely wait and see that's kind of my impression um that's kind what we wrote about today in our by the way I I think it's crazy I wasn't aware that the Ver Rubin post would come out until today like literally I learned today and I was oh so I had to read it live with everyone else I think that's that's a big deal too and we have a lot of information there um about what's coming at GTC so

well well we're excited to follow it um Nvidia earnings are usually out by uh 20 minutes after the hour 420 but it appears that they are delayed so we will we will see um I mean Of course, we'll we'll have to have you back on the show to deep dive everything that's happening in the semiconductor.

This this makes it this makes it so much better. I had like the bogey and all the information and what the setup is going to be. And honestly, if we have a two minutes, I'll tell you what to expect.

Okay. Yeah, please.

I think uh everyone thinks the guide will be 75 billion. That's what everyone's really obsessed about.

Um I think the questions that are going to be answered or asked rather is about memory price inflation. Um, that's going to be a big deal given the fact that I think memory prices are like, you know, mooning right now. HPM and DRAM. I think Nvidia locked it all in. So, everyone is going to be like, "How sure are you your gross margin isn't going to go down?" And they're going to be like, "We're really sure." But they're going to be like, "Okay, how sure and what time?" And then there'll be like this random like circle that nobody can't answer. Um, and then I think one of the questions that people are going to definitely ask is about the power side. Um it seems like next year Nvidia has more chips than power and so that's going to be and I mean to be clear Nvidia has no power, right? They have customers who acquire power and you know one of their big competitors in terms of market share which is Google has chips and power and so I think that that's going to be a huge bottleneck and I think that's going to be one of the bigger questions that happens this quarter. Hey, how do you deal with the power bottleneck? Um because you you're going to sell a lot of chips, but if you can't even plug them in, this is a huge issue.

And it's the idea is is the idea that even if a a lab or a hyperscaler said, "We're actually good on chips right now. We don't have the power for for them." Nvidia would be like, "Well, you actually have to buy them if you want to maintain priority." How do you think they would play that?

I I don't know. I mean, I think the thing is it's not going to matter because the only company with a chip to sell you next year that is not accounted for is Nvidia.

And I think the the shortage in GPUs is really underappreciated. There's like um you know, you can see it publicly. I forget which research shop shows it like the availability of B200s, but our anecdotal information as well as the like the live price tracking and contract price that we're aware of, there's like no H100s for sale.

Yeah.

Like the H100s are sold out. That's like boom, you're, you know, your five-year depreciation like bear thesis that everyone was really freaking out about last year. Well, good luck. You can't even buy one today. Um, so that's probably the biggest interesting thing and I think probably the single most bullish thing you can say. Um,

love it.

You know, like a four-year-old chip effectively is completely sold out today.

Yeah.

What what's that what does that say about the next generation?

That's remarkable. Well, thank you so much for taking the time to join. Uh, bummer that they delayed on us, but next time we'll time up so that we have even more time to hang out. But

I'm I'm I'm so happy. Honestly, live reacting to Nvidia earnings. I was like,

no, no, no. We we we'll get this flow going. Uh we we this is always fun. But I mean, just great to talk about everything. And happy birthday. We'll