Dylan Patel on GPT-5’s Router Moment, GPUs vs TPUs, Monetization

Dylan Patel 0:00 NVIDIA's gonna have better networking than you. They're gonna have better s-- uh, HBM. They're gonna have better process node. They're gonna come to market faster. They're gonna be able to ramp faster. They're gonna have better negotiations with whether it's TSMC or SK Hynix in the memory and silicon side, or all the rack people, or, like, copper cables. Everything they're gonna have better cost efficiency. So you can't just, like, do the same thing as NVIDIA. You have to really leap forward in some other way. You have to be, like, five X better.

Erik Torenberg 0:22 [on-hold music] Dylan, welcome to the podcast.

Dylan Patel 0:27 Thank you for having me.

Erik Torenberg 0:28 We've been trying to get you for a while. You're a busy man. Uh, but, uh, but it worked out. Um, Guido, why-why don't you introduce why we're so excited to have, have Dylan on the podcast and what we're excited to discuss.

Guido Appenzeller 0:37 Look, I mean, I-I-I think, Dylan, you've done exceptional job in, in covering what's happening in the AI hardware space, AI semi space, and, and now more and more data center space as well. And, you know, uh, just looking at it, currently the most valuable company on the planet is an AI semi company, right? Uh, the, the, I think biggest IPO so far in AI was an AI cloud company. This is currently where it's happening, right? In any gold rush in the early days is the pick and sh- picks and shovels that make money, and I think this is the stage that we're in. So super excited to have you here today.

Dylan Patel 1:07 Awesome. Thank you. Uh, w- happy to talk about my favorite topics.

Dylan Patel 1:11 [chuckles]

Erik Torenberg 1:11 Amazing. Well, maybe let's start with GPT-5. You know, we just had the, um, some of the researchers from who, who, uh, Christina and Isabelle on here last week. What, um... y-you, you said it was, uh, disappointing. Uh, what why don't you sh-share your reactions or what capabilities you were, you were hoping to see or overall plans.

Dylan Patel 1:26 I think, I think it depends on what tier of user you are, right? Um, and so, like, right, if you're just using GPT-5 and before you were, you know, two-- twenty dollars or two hundred dollars a month subscriber, um, you no longer have access to 4.5, which i-in my opinion, is still a better pre-trained model for certain things. Um, or you no longer have access to, um, o3, which would think for thirty seconds on average, maybe. Right? Whereas GPT-5, even when you're using thinking, uh, only thinks for, like, five to ten seconds on average, right? Um, which is an interesting sort of phenomenon, right? Um, but basically, like, GPT-5 is not spending more compute per se. The model did get a little bit better, um, on a vanilla basis, right? 4o to 5 is, is actually quite a bit better. But, um, when you think about, you know, what, what is this curve of intelligence, right? It's like the more compute you spend, the better the model gets. Um, and that's whether it's a bigger model, uh, which GPT-5 isn't, right? You can, you can see it's not a bigger model. Um, it's roughly the same size, um, and, you know, or you think more, right? Uh, but again, like, this is something that OpenAI is mo-- first thinking models, you know, the first few, uh, generations of o1, o3 would think for a long time, um, and waste a lot of tokens, uh, if you will. Uh, and when you look at, like, for example, um, you know, Anthropic's thinking models, they actually think even when you put them in thinking mode, they think a lot less, right? To get to the same results or better results, right? As OpenAI was. And so OpenAI, I think, like, optimized a lot of, like, well, if I ask... Like I think, I think the silliest one w- I had asked was, like, I asked o3 once, like, "Is pork red meat or white meat?" And it thought for like forty-eight seconds. I was like, "What are you doing?" Like, "This should just, like, this should l- this should just, like, tell me the answer." [chuckles] Um, and, and, and so, like, you know, the, the, the nice thing is that GPT-5 will think a lot less even if you select thinking manually. But more importantly, they have the sort of auto functionality, the router, which lets them decide whether or not, you know, "Hey, do I route to the regular model? Do I route to maybe mini," if you're at our rate limits? "Or do I route to thinking?" Right? Um, and, "How much do I think?" Uh, but in general, the thinking model will think less. So, so there's less compute going into, uh, a power user's average query than before.

Guido Appenzeller 3:46 But is it even more interesting? OpenAI cannot control how much compute it wants to allocate to you, right? If, if, if we're in a high load situation, maybe tune the router a little bit so it's less, right? Maybe, maybe we skip. I have no idea what they're doing, um, you know, behind the curtain. But there's this meme out there at the moment that basically all they did, which is, which is a meme, right? It's not true. But, but all they did is, you know, take o3 plus a couple of smaller models, put a router in front, and, you know, offer that at a, a lower blended price, essentially, right? Is-- I think there's a little bit of that. Like, cost suddenly matters and, and they figured out a way how they can steer that.

Dylan Patel 4:19 I think, yeah. The, I mean, and they talked about how they've been able to dramatically increase their infrastructure capacity because I myself was just regularly using o3 or 4.5, right? And now I'm forced to use Auto, which sometimes gives me, you know, the o3 equivalent thinking model, but sometimes gives me, you know, just the regular base model-

Guido Appenzeller 4:36 That's what they are

Dylan Patel 4:36 ... uh, which, which sucks. But, like, um, I think for the free user, it's actually quite interesting, right? Uh, the free user was not getting thinking models pretty much ever, or not using them. Um, or in many cases, they just open the website and ask their query, and now sometimes their query gets routed there. So sometimes they get a way better model. But now sometimes, you know, the, the OpenAI can gracefully degra-degrade them-

Erik Torenberg 4:58 Exactly

Dylan Patel 4:59 ... uh, if they need to, right? And it's like, I think, I think the router points to the future of OpenAI from a business, right? Like, you can look at sort of the model companies, right? Anthropic is fully focused on B2B, right? Um, API code, et cetera, right? Where- or, or Claude code, whatever it is, right? OpenAI, yes, they have that business, Codex and, and API business, but really their majority of their revenue is consumer, right? Um, and it's consumer subscriptions. But they have no way to upsell, like, you know, how to make money off of all the free users, right? Uh, in any other application, consumer app, the free user still pays via ads. Um, but this is not compatible with AI, right? Like, it's a helpful assistant. You can't just make the r- users' result worse by injecting ads. Banner ads don't really work, uh, in AI either. So it's like, how do you now monetize them? And I think, I think they're getting-- with the router, they're getting really close to figuring out how to monetize that user, right? Um, with the new CEO of Applications, if you saw her product that she launched at Shopify, uh, I think it was Shopify, was an agent for shopping.Right? And now this, like, immediately clicks. Like, oh, if the user asks a low-value query like, "Hey, why is the sky blue?" Just route them to Mini, right? Like, you know, the model can answer perfectly fine, and that is a chunk of queries, right? And but if they ask, like, "What's the best DUI lawyer near me?" Right? All of a sudden this is like, you know, you're in jail, you have one shot. You know, you're like, "Screw it. Let me ask ChatGPT what the best U- DUI lawyer is," and now all of a sudden... You know, the model's not capable of it today, but soon enough it'll be able to, you know, contact all the lawyers in the area and, um, you know, figure out what their results are and maybe search their, like, court filings and whatever, right? And, and, and book the best lawyer for you or, or an airplane ticket.

Guido Appenzeller 6:42 They negotiate a cut, [laughs] you know, as part of that.

Dylan Patel 6:45 Yeah, of c- of course they're gonna take a cut, right?

Guido Appenzeller 6:46 Yeah.

Dylan Patel 6:46 But this is a, this is a much better way of, like, monetizing the free user is, like... You know, it, it's like Etsy, 10% of their traffic now comes from chat, and OpenAI makes nothing off of that. But they very... They really, really will soon, right? Um, and partially that's because Amazon blocks chat but, like, you know, there's a way to make money from shopping decisions, whether it's booking flights or looking for items. Um, and those you now say, "Free user, I don't care. I'm gonna send you to my best model. I'm gonna send you to agents. I'm gonna spend ungodly amounts of compute on you because I can make money off of this. But if it's a query that's like, 'Help me with my homework,' I'll send you, like, a decent model, right? I don't, I don't need to spend money on you." Um, and so this is how I think, like, OpenAI can finally make money off of the free user, and I think that's the biggest, like, thing about the router, right? Like-

Guido Appenzeller 7:34 Th- th- this is super interesting. I think this is the first time that we've seen that there's a, a, a launch of a new model where to some degree cost is the headline item, right? I mean, so, so far it was always like, who has the smartest model? Who has the highest MMLU score? Now we have suddenly people who use models for coding for eight hours a day and, you know, surprise that, you know, if you take a, a large context window and the best model, creates, you know, thousands of dollars of cost a month. So cost matters. And, and so to some degree, so where are you on the Pareto frontier between cost and performance is the new benchmark for model competitive and no longer cost alone. Is, is, is that what we're seeing here or?

Dylan Patel 8:07 I mean, I think definitely, right? Like, OpenAI said they doubled their rate limits for, uh, you know, cer- for big amounts of, uh, users. They've, they've dramatically increased the number of tokens they're serving from this launch-

Guido Appenzeller 8:18 Mm

Dylan Patel 8:18 ... which effectively says this is an economic release. Um, and I think, like-

Guido Appenzeller 8:23 And, and probably also means the tokens are now cheaper, right? Otherwise-

Dylan Patel 8:25 Yeah, yeah. For sure, for sure.

Guido Appenzeller 8:26 Yeah.

Dylan Patel 8:26 I, I think the funniest thing is, like, you know, this whole cost thing you mention is like we've seen this in the code space, right? Like, Cursor had to pull away the unlimited, um, Claude Code. You know, initially they had, like, this super expensive plan, and it had, like, unlimited rates, and then they... or only, like, a weekly rate limit. Now they have, like, hour-based rate limits. Um, and I saw the craziest, like, thread on Twitter where this guy said he changed his sleep schedule, right? Uh, modeled after, like, how sailors in the bay-

Guido Appenzeller 8:53 [laughs]

Dylan Patel 8:53 Like, you know, like, if you're sailing you can't sleep, right? Like, s- solo sailing, uh, they'll take, like, power naps, uh, when they get to the right spot so that they can, like, still be safe.

Guido Appenzeller 9:02 In the morning when it's not very windy. [laughs]

Dylan Patel 9:04 Well, but, like, they can't sleep uninterrupted, right? And so because Anthropic had to put rate limits that are, like, not just week based but, like, number of hours based, he's like in, like... He, like, basically sleeps, you know, multiple times a day but-

Guido Appenzeller 9:17 That's amazing

Dylan Patel 9:17 ... small chunks just so he can maximize the usage. And there's also a leaderboard on Reddit-

Guido Appenzeller 9:21 Oh, my goodness

Dylan Patel 9:22 ... where people are, like, competing to see how many tokens-

Guido Appenzeller 9:24 That is amazing

Dylan Patel 9:24 ... they're using through their subscription, and there's, like, a dude spending, like, $30,000 a month.

Guido Appenzeller 9:29 So I'm gonna find some developer in India that I can do pair programming with so I can get the day cycle, he can get the night cycle, [laughs] you know, and we both can, can maximize the, together the quota for the account. Is that the future then? [laughs] It's-

Dylan Patel 9:39 Um, I mean, but it's, it's clear, like, people are taking advantage of the negative gross margin, like, sort of-

Guido Appenzeller 9:43 Yeah

Dylan Patel 9:44 ... uh, you know, subscriptions that are offered. Um, you know, I think, I think Anthropic probably makes a positive gross margin off of my subscription. Like, I don't code enough, but, like, there's plenty of people that are definitely losing money.

Guido Appenzeller 9:55 Yeah.

Dylan Patel 9:55 Um, and so as you said, it's an economic-

Guido Appenzeller 9:58 It, it, it push more and more to, I think, just usage-based pricing, right?

Dylan Patel 10:00 Yeah.

Guido Appenzeller 10:00 I think if, if you have a, if you have an underlying commodity that you're reselling to some degree that has, that is that large a part of your cost of goods, right, you need to go to usage-based pricing.

Dylan Patel 10:10 How much, how, how much do you think the, like, um, customer capture and stickiness for these code products is? Like, um, yeah, I'm, I'm curious what you think on that, right? Like, y- you know, once you use an ID, once you integrate, you know, one of the CLI products in, like, how sticky is it or is-

Guido Appenzeller 10:28 That is a billion-

Dylan Patel 10:28 ... people just switch after? [laughs]

Guido Appenzeller 10:28 That is a billion-dollar question. That's a very conservative estimate. Look, uh, A- A- Andrew Karpathy has this, this great slide where he basically says if you're building an agentic system today, right? But fundamentally what it is is sort of this loop, right, where, where half of the loop is the model thinking, right, and, and then trying to do something. The other half is then the, the, the user verifying what did the agent do, is it the right thing, providing feedback and trying to steer it in the right direction because, you know, it can't run forever. Eventually you need to, you need to, to, to steer it back. One half of that is, is the model provider, right? They're trying to build the, the best models. The other half is really about, I think, designing the best possible UI to enable the user to give feedback, and I think there's value in that. So I think there's a certain amount of stickiness in there, right? So what are all the different tools, like, in terms of vis- like, say, take code editing, right? How can I most easily visualize what the code changes are? How can it, it most easily visualize, you know, what they impact, which files? You know, how can I for small changes get very quick feedback versus for complex ones, you know, get complex feedback? And there's now some tools that actually draw diagrams for you of what they do, right? [laughs] So I think, so I think this will be the battle, is I think there's stickiness in that, right? Uh, you know, how much exactly? This is a great question.

Dylan Patel 11:30 So in that sense, like, people should be doing subscriptions to get people locked in, right? Instead of moving to usage-based pricing.

Guido Appenzeller 11:38 Well, I think it's the customers that don't wanna do usage-based pricing-

Dylan Patel 11:40 Yeah

Guido Appenzeller 11:40 ... because it's so hard to guarantee, you know... It, it's so hard for it to get, y- it to get, it to get away from them, and you, you actually want guarantees, and you're, you're willing to, like, say... You're, you're willing to commit to pretty high spend in order to not have usage-based pricing. I think it's the model companies that want usage-based pricing.

Dylan Patel 11:57 I, I think with consumers, [coughs] it's frankly very hard to h- not have usage-based pricing just because the, the variability is so massive, right? If it's, like, you know, us coding versus somebody who does this as their full-time job, right? Th- the... You just have a factor of 20 or so difference in usage, and if that is... Th- that costs a lot of money, right?

Guido Appenzeller 12:13 Mm-hmm.

Dylan Patel 12:13 The instance is just one. I think for enterprises we could, we could see, like, more flat-fee pricing because you can average it out more.

Guido Appenzeller 12:19 You have a, a developer that's using it all day, you kinda know in a general sense.

Host 12:23 Like, h-h-how many hours a day they're programming and, and what that sort of looks like.

Guido Appenzeller 12:26 Yeah.

Dylan Patel 12:26 The, the vibe quotas are harder.

Host 12:28 [laughs] Yeah.

Guido Appenzeller 12:30 Before we leave OpenAI, I, I wanna ask a broad question, which is if Sam Altman was, was sitting here and saying, "Hey, Dylan, I'll listen to anything you, you tell me to do, any advice you have, as long as it makes OpenAI more valuable," what would you, what would you tell him?

Dylan Patel 12:41 I would say, like, immediately launch a, uh, a method for you to input your credit card into ChatGPT and agree that for anything it, like, agentically does for you, it'll take X cut, and then launch that product because, uh, where, where it does shopping, right? 'Cause, like, we know-- everyone knows that, like, Anthropic and OpenAI and all the other labs are buying RL environments of Amazon, and of, like, Shopify, and of Etsy, and of all the different ways to shop on the internet. Of, of airline websites, right? Now, just like, "Hey, integrate my calendar. I wanna, I wanna fly to there on Thursday. Make sure I don't miss a meeting." Cool. Booked, right? Like, do, do that integration, like, super well. Know my preferences on wha-wha-whether I like aisle or window. All this stuff, right? And just take a take rate. Like, I think, like, this will make them so much money the moment they launch it, and I, I think they're working on it already, but, like, I'd like to hear how he thinks about it, uh, 'cause he's shifted his tone massively on, like, ads over the last six months, right? He used to be like, "No way," and now he's like, "Uh, maybe," you know? Uh, there's a way to do it without, you know, harming the user, and I think this is, this is how you monetize the free user, right? So I think, uh, that's probably what I'd tell him/ask him about. Like a whole line of questions-

Guido Appenzeller 13:55 Yeah

Dylan Patel 13:55 ... around this.

Guido Appenzeller 13:56 I, I wanna shift to NVIDIA. Uh, NVIDIA's having a monster year. They're up almo-almost seventy percent. W-what are the possible paths fro-from here? How do you see, how do you see it playing out?

Dylan Patel 14:05 Um, depends, like, how, how pilled you are on, like, the continued growth, but, like, I think, I think you guys have a good vantage point, and we have a good vantage point of how fast revenue's growing for a lot of these companies-

Host 14:16 Mm-hmm

Dylan Patel 14:16 ... especially the code companies,

Dylan Patel 14:18 but, but even many other applications. Um, I think we can clearly see, uh, the demand side is, is accelerating, right? Um, and, and then if you look at the training side, I think the, the race is on. Um, you know, Meta's, Meta's upping hugely. Uh, Google's upping hugely. Um, you know, if you, if you just look at, like, again, just OpenAI and Anthropic and the compute that they have and are getting this year, uh, from Google and Amazon for Anthropic and from Microsoft, CoreWeave, Oracle, uh, for OpenAI, thirty percent of the chips are going to them, just those two companies. Um, but that's actually, like, okay, well, like, seventy percent of the stuff, like, who's making off of it? Well, a one third of it is, like, ads, right? Um, whether it be ByteDance or, uh, Meta or many of the other people who are doing ads. So then it's still like, okay, well, where are the rest of these one third of the chips coming from? Well, they're, like, mostly uneconomic providers who I don't think it's, like, an obvious bet that they're going to keep growing, um, you know, keep raising bigger and bigger rounds. So what happens there? Um, I think, I think with the... You know, we, we talked about, like, coding, right, like earlier. Actually, the Qwen Coder 3 model is actually super cheap if you're running it on-prem or if you're running it in the cloud with all these inference libraries. Um, and so, like, there's stuff like that as well. So I think the question is, like, how much does it keep growing? Um, because clearly, you know, I think the first third is definitely skyrocketing, right, of, of OpenAI, Anthropic, you know, lab spend. The second third of, like, ads is gonna grow. It's not gonna grow like crazy, but, like, I think there's definitely an inflection point that could be hit with gen A- gen AI ads. I know Meta's been experimenting with it a lot. Um, but I could totally be convinced that there's gonna be a huge inflection in, like, uh, take rate there, right, where you start showing me personalized ads. Um, like every person that's an ad is like-- looks like me, and I'll be like, "Okay, yes." Except, like, slightly better, so I, like, feel better, right?

Host 16:13 Inspiration.

Dylan Patel 16:13 And I'm like, "I wanna buy it." Yeah.

Guido Appenzeller 16:15 Um, it's, it's super interesting, the, the... I have no idea how this is gonna, gonna scale, right? But, but if you ask the question, how much could it scale, right? Like, how much value are we creating here? Is this-- Can we create enough value to actually, you know, keep growing for, for a long time? If you just take AI software development, right?

Dylan Patel 16:31 Yeah.

Guido Appenzeller 16:32 We know we can easily get about fifteen percent more productivity out of a developer.

Dylan Patel 16:35 I don't think that's right. I think it's way higher.

Guido Appenzeller 16:37 No, no, the-- W-with the, with the straight... Like, I talk to a lot of enterprises. Like a, a classical enterprise, straight up, you know, GitHub Copilot deployment, that gives you about fifteen percent. We can do much more than that.

Dylan Patel 16:45 But bro, like [chuckles] you know how bad GitHub Copilot is?

Guido Appenzeller 16:48 [laughs]

Dylan Patel 16:48 Like, like, like how did they-

Guido Appenzeller 16:50 We-

Dylan Patel 16:50 How-- Th-they-- Look at the revenue ARR chart.

Guido Appenzeller 16:52 This was the lower bar.

Dylan Patel 16:53 It's so funny. It's so funny. If you look at the revenue ARR chart and it's like, it's, like, Claude Code in like three months has surpassed them, and, like, Cursor co- you know, easily surpassed them, and then, like, even, like, companies like Replit are, like... and Windsurf/Cognition are, like, gonna pass them. Like, it's like-

Guido Appenzeller 17:08 You're preaching to the choir

Dylan Patel 17:08 ... what's going on? [laughs]

Guido Appenzeller 17:09 So, so look, let's assume we can get this to a hundred percent.

Dylan Patel 17:11 Yeah.

Guido Appenzeller 17:12 Right? So it's we can double the productivity of a developer, right? About thirty million developers worldwide, give or take.

Dylan Patel 17:17 Yeah.

Guido Appenzeller 17:17 Right? Let's say 100K value add per developer. This might be a little high worldwide. The US is low, but, but worldwide it's high. So that's three trillion dollars.

Dylan Patel 17:24 Yeah, yeah.

Guido Appenzeller 17:25 Right? So, so w-we're probably building technology here which adds three trillion vol- dollars of GDP value.

Dylan Patel 17:30 Yeah.

Guido Appenzeller 17:30 In theory, we could put that into, into GPUs [chuckles] because that's the main cost factor here, right?

Host 17:33 Just, just from a coding model, like not-

Guido Appenzeller 17:35 Just from a coding model. This is one use case, right?

Host 17:36 Ignoring every other use case.

Guido Appenzeller 17:37 So, so at least in theory, the value generation is here to keep growing, right? Now, how that translates to the industry is much more complicated.

Dylan Patel 17:42 I think, I think we've already seen, um, AI's value creation, uh, exceed... So, so, so sort of there's, like, the whole, like, like, the famous, like, oh, 300 billion problem or 200 billion problem. Now it's 600 billion problem.

Guido Appenzeller 17:55 Mm-hmm.

Dylan Patel 17:55 I'm sure Sequoia's gonna put out like the-

Guido Appenzeller 17:56 That was a funny post. Yeah. [laughs]

Dylan Patel 17:57 ... trillion-dollar problem, right, soon enough. But, like, like, there is some, like, reality in that, of course, but, you know, it ignores that, like, infrastructure spend today is accounting for five years of revenue, not, like, one, and the revenue looks like this, not like flat line. But I think, um, I think the main thing is that AI is already generating more value than the spend. It's that the value capture is broken, right? Like, I legitimately believe OpenAI is not even capturing ten percent of the value they've created in the world already, um, just by usage of chat.

Guido Appenzeller 18:28 Yeah.

Dylan Patel 18:28 Right? And I think the same applies to, you know, Anthropic and Cursor and, and whoever else you're looking at. I think, I think-The value capture is really broken. Um, even like internally, I think like what we've been able to do with like four devs w- in terms of like automation, like our, our spend on like Gemini API is absurdly low, and yet we go through every single permit and regulatory filing around every single data center with AI. And it's like-- And we, we take satellite photos of every data center, and we like-- we're able to label our dataset and then recognize what generators people are using, what like cooling towers, and the construction progress and substations. All this stuff is like automated, and it's only possible because of GenAI. But-- And we do it with n- like very few developers, and then like the value capture that I'm able to generate by selling this data, by consulting with it is so high, but the company is making it as like, like they, they get nothing out of it, right? Like, I think this like there's a value capture challenge here, uh, that far out exceeds the sort of creation, right? And, and, and as you get models like GPT-5 or open source models, like continuing to drive it down, it's like the value capture is just harder and harder and harder for these companies 'cause they're making, you know, fifty percent gross margin on inference if they're, you know, or, or less in many cases.

Guido Appenzeller 19:43 In, in so many words you're saying, "We're getting commoditized, uh, and, and therefore you can't capture the value, and thus you should temper your expectations of how much, how much, uh-

Dylan Patel 19:51 Um

Guido Appenzeller 19:51 ... you can spend in GPUs."

Dylan Patel 19:52 Well, no, I think, I think you can-- I think there's still ways to like inflect hugely on value capture, right? Um, like I mentioned, the ads are a huge value capture inflection-

Host 20:01 But that needs to happen before, before we see massive increase.

Dylan Patel 20:05 Um, no, I think, I think the other thing is like there's a lot of capital that's not been spent, right? Like the hyperscalers still can grow CapEx, uh, twenty, thirty percent next year, right, from what they're doing this year. In addition, companies like CoreWeave and Oracle, uh, because they're tapping capital markets, can raise way more than twenty to thirty percent CapEx. And then you go down the list further and it's like, oh, the largest infrastructure funds in the world like Brookfield and Blackstone, well, actually they're, they're turning all of their eyes to investing even more into-

Host 20:34 Mm

Dylan Patel 20:34 ... infrastructure, AI infra. Um, and then you're like the sovereign wealth funds of the world, like the G42s or, you know, um, the Norway one or GIC, uh, in, um, Singapore, like these people have barely started touching AI. And so I think there's a whole lot more CapEx that can come without it being necessarily like economically motivated day one. Um, I'm more so saying like economically motivated CapEx can only grow like so much, but there's so much other like, like, like where it's not clear from, you know, if you have a spreadsheet, you know, and you're basing it on real business, that you should actually spend this much. But people will because they believe, I believe, I think you believe, like infra, you know, people believe that this will be... You'll get profit out of it, but there's no like one hundred percent certain like, you know, way to argue it. Yeah.

Guido Appenzeller 21:24 H-how threatened, uh, if at all is NVIDIA by, by custom

Guido Appenzeller 21:27 silicon?

Dylan Patel 21:28 I think that's the biggest thing, right? Is, um, when we look at orders from Google and from, um, Amazon, right, especially their... and, and Meta, their custom silicon is... Not, not Microsoft, their custom silicon kinda sucks. Uh, but the other three, they're really upping their orders massively over the last year. Um, you know, Amazon is making millions of Trainium. Google's making millions of TPUs.

Host 21:52 Mm.

Dylan Patel 21:53 Um, TPUs clearly are like a hundred percent utilized, right?

Guido Appenzeller 21:56 Yeah. They're doing very well.

Dylan Patel 21:57 Um, Trainium's not there, but I think Amazon will figure out how to do that, um, and Anthropic will. Um, so, so I think, I think that's the biggest threat to NVIDIA, is that people figure out how to use custom silicon more broadly. Um, and this sort of becomes this sort of like if AI is concentrated, then custom silicon will do better. Um, and that's not even talking about like OpenAI's silicon team and stuff, right? Like if AI is really concentrated, uh, then, then they'll do better, uh, custom silicon. But if it gets dispersed broadly because there's all these open source models from China, um, and there's all these, um, open source software libraries from, you know, NVIDIA and, and China, and it makes the deployment costs like rock bottom, then potentially-

Guido Appenzeller 22:39 Hear me out here. If, if Google's TPU is, is able to compete with NVIDIA, in theory it could do it on the, on the open market. NVIDIA is worth more than Google these days. Shouldn't Google start selling their chips to everyone? I mean, in theory, they should be able to achieve a higher market cap that way, right?

Dylan Patel 22:54 I, I, um, I absolutely think so. I think Google's even discussing it, um, internally. I think it would require a big reorg of culture, um, and a big reorg of like-

Host 23:06 Mm

Dylan Patel 23:06 ... how Google Cloud works, um, and how the TPU team works, and how the Jack software team and XLA software teams work. Um, I totally think they could. Um, it would just take them like shaking themselves pretty hard to be able to do it. Um, yeah, uh, but I, I, I, I totally think Google should sell TPUs externally. Not just renting, but like physically with racks.

Guido Appenzeller 23:26 It's, it's kind of funny if a side hobby, in theory, has a higher c- uh, company value potential as your main product now. [laughs]

Host 23:33 Than your entire business. Especially as you think about the degradation of search-

Dylan Patel 23:36 So I think-

Host 23:36 ... as a core business. I mean...

Dylan Patel 23:37 Yeah, I think... But I think like if you were to ask like Sergey, right, like, "Hey, do you think selling chips and in- and racks is more valuable or cloud or, or Gemini?" Um, he'd be like, "No, no, no, no, no. Like Gemini is gonna be worth way, way, way more. It's just not yet today," right? Um, and so I think like, like today you say NVIDIA's the most... Again, it's like a whole concentration thing, right? If the world is super concentrated in terms of customers, then NVIDIA will not be the most valuable company in the world, right? Um, but if it gets dispersed more and more, um, which arguably we're starting to see with a lot of these open source models getting better and better and better, um, and with the ease of deploying them getting better, then you would see... I think you could argue NVIDIA will remain the most valuable company in the world for a long period of time. Um-

Guido Appenzeller 24:26 And hi- historically, no pun intended, software has eaten the world in most markets-

Dylan Patel 24:30 Yeah

Guido Appenzeller 24:30 ... right? I mean, like if, if you look at, uh, early networking days, Cisco was the most valuable company on the planet, right, for a while.It's no longer, right? The, the guys that build services on top like, like Google or Amazon or, or Meta eventually eclipsed that

Dylan Patel 24:43 Well, which is why NVIDIA's, like, making all these software libraries, right?

Guido Appenzeller 24:46 Mm.

Dylan Patel 24:46 Like, that's, that's-- And they're trying to commoditize inference, right? Like, um, I-- You guys don't, I think, even have an inference API provider investment, do you? Um...

Guido Appenzeller 24:56 Well, we have, we have all kinds of model providers.

Dylan Patel 24:58 Model providers.

Guido Appenzeller 24:59 But, but that's a different-- Yeah.

Dylan Patel 24:59 But I'm talking about a pure API provider investment, I think, right? Is that correct? I think I talked to, um, one of the team members, maybe, maybe Rog'co or someone about, like, why you guys don't-- didn't invest in, like, you know, like a Together or, like, a Fireworks. And sort of the argument was, like, "Well, we think infer- just serving models, uh, alone without making them will sort of be commoditized."

Guido Appenzeller 25:19 Yeah.

Dylan Patel 25:19 Right? Um-

Guido Appenzeller 25:20 We, we have some in the stable diffusion ecosystem, like with, like, uh, Faul-

Dylan Patel 25:23 With Faul, yeah, yeah

Guido Appenzeller 25:23 ... or the Gate.

Dylan Patel 25:24 Like that is-- Yeah, it's a little bit different, uh, dynamics there, I think.

Guido Appenzeller 25:27 They, they, they tend to be- make much more component models than the, the LM folks, I think.

Dylan Patel 25:30 Yeah, yeah.

Guido Appenzeller 25:31 So it's a little different.

Dylan Patel 25:32 Um, but, but, like, you guys don't have one of these, like, you know, Base10 or any of these, like, sort of like API investments-

Guido Appenzeller 25:37 Yeah

Dylan Patel 25:37 ... because you think, um... This is from someone on the infra team that you guys think it'll get commoditized because the software NVIDIA's making, because VLLM and SGLang, which is, like, open source software coming out of Berkeley, and now, you know, sort of has their own environments now. Um, and supported by many, like... This being commoditized means that, like, API providers aren't necessarily worth a ton, right? Is sort of your argument, maybe. Um, and I think that's, that's relevant to this whole thing, which is, you know, why, why, right? Like, why would you do this?

Guido Appenzeller 26:06 Shifting gears, what, what about the, the silicon startups?

Guido Appenzeller 26:09 Uh, what's, what's your take on those? I mean, there, there's, there's a ton of capital flowing in-into that, right? We've seen, I have not numbers, but probably billions, uh, being invested in, in chip startups.

Dylan Patel 26:19 Yeah, for sure. For sure. I mean, like, whether you're looking at, like, um, you know, companies like-- I think it's, like, pretty impressive that a few companies like, um, Etched and Revos, um, and a number of other companies, you know, Madax and, and others, like, have gotten the amount of funding they've had without even launching a chip, right? Um, you know, in the past, like, yes, silicon companies would make money or raise money, but they would at least launch a chip before they get a, you know, a big round. But, like, uh, Etched and Revos, like, have raised, you know, a lot of money without ever launching a chip publicly-

Guido Appenzeller 26:52 Yeah

Dylan Patel 26:52 ... which I think is... I mean, it speaks to, well, like, yes, silicon is super, uh, capital intensive if you're building a chip, especially an accelerator, which has so many moving pieces. Um, and there's, there's, like, there's, like, ten different AI accelerator companies out there, right? Like, that are new-ish in the last few years.

Guido Appenzeller 27:08 I think there's a lot more.

Dylan Patel 27:09 [laughs] Uh, that, that are like-

Guido Appenzeller 27:10 [laughs]

Dylan Patel 27:10 Yeah, yeah, yeah. That's fair. Um, and then, and then, and then there's the old guard, which continues to raise money-

Guido Appenzeller 27:16 Yeah

Dylan Patel 27:17 ... right? Like Groq and Cerebras and-

Speaker 27:18 Mm-hmm

Guido Appenzeller 27:19 Somanova

Dylan Patel 27:19 ... and Somanova and, and Tenstorrent and so on and so forth, right? Like, um, or Graphcore getting bought out by SoftBank, and SoftBank dumping money into this effort as well, right? There's, there's a lot of capital being invested to disper- dis- uh, dispel sort of NVIDIA's top dollar or top position. Um, but it becomes challenging, right? It's like, how do you beat NVIDIA, right? Like, the hyperscalers, I think, are, like, kind of lucky in that they can, they can do mostly the same thing as NVIDIA, right?

Guido Appenzeller 27:46 They're, they're captive customer, which is themselves, right?

Speaker 27:48 Yeah.

Dylan Patel 27:48 Right.

Guido Appenzeller 27:49 It's a huge asset, yeah.

Dylan Patel 27:49 And it's-- They can, they can just win on supply chain, right?

Guido Appenzeller 27:51 Yeah.

Dylan Patel 27:52 Like, I'm using cheaper providers.

Guido Appenzeller 27:53 It's a, it's a margin compression exercise, essentially, right?

Dylan Patel 27:55 Yeah, yeah. And maybe, maybe for certain workloads, like Meta for recommendation systems, they'll have a better... You know, they can specialize more. Um, but for the most part, it's like, "No, we're, we're targeting the same workloads. We can just simplify supply chain or, or in, in-house a lot of it and compress margin, and it'll be fine." Um, but in the case of, you know, these g- these other companies, it's like, well, they don't have a captive customer. So now you have to contend with, "Well, I'm using the same ecosystem, um, and either I can use some custom silicon provider who's gonna p- take a margin anyways on top, and that's gonna compress my, like, what I can sell for. Or I can try and in-house everything," but then it's like, this is really hard, right? Like, I'm gonna do all the software des- I'm gonna do all the silicon design. I'm gonna build all this different IP. I'm going to manage the supply chain on chips, on racks, on everything, right? Ends up being a huge effort in terms of team size. Um, all in the end, like, hey, I make a seventy-five percent gross margin as NVIDIA. Um, AMD sells their GPUs for fifty percent gross margin, um, and they have a hard time out-engineering NVIDIA, and they're great at engineering, right? Like, uh, they're, they... But yet they still take more silicon area, more memory to ach-achieve the same performance, and they have to sell for less, so their margin gets compressed, right?

Guido Appenzeller 29:08 That makes sense. But, but look, the... I think historically, if you look at it, typically, the, for new entrants in markets didn't win by marginally improving on something existing. That happens sometimes, but, but more likely, they, they jumped on some of, some kind of disruptive technology leap, right? Where it's like, we have a different approach, we have different technology. I- is that possible here? I mean, to, to, to some degree, um, maybe this is oversimplifying it a little bit, but I think part of the reason why the transformer model won was because it runs so incredibly great on GPUs, right?

Dylan Patel 29:35 Yeah.

Guido Appenzeller 29:35 Like a, like a recurrent neural network is similarly performant it looks like, but it, [laughs] it runs terribly on a, on a GPU. So, so did we sort of pick the model for an [laughs] architecture, and now it's, it's hard to come up with an architecture that, uh, you know, really-

Dylan Patel 29:47 Well, it, it's, it's hardware-software co-design, right?

Guido Appenzeller 29:49 Exactly.

Dylan Patel 29:49 Like, there is like-

Guido Appenzeller 29:50 Yeah, exactly. [laughs]

Dylan Patel 29:50 There's all this hype about, um, neuromorphic computing, right? Like, theoretically, it's amazing and super efficient. It's like, okay, great, like, there's no ecosystem of hardware, there's no ecosystem of software. It would take, like, you know, tens of thousands of people who are the best at AI today focusing on that to even prove out if it's worthwhile or not, right? Um, on a hardware side, on a software side, on a model side. And so, like, you look at, like, Groq, Cerebras, Somanova, um, they all, like, sort of over-indexed to the models that were leading at the time when they designed their chips.

Speaker 30:20 Mm-hmm.

Dylan Patel 30:20 And so they made certain trade-offs, right? They put a lot more memory on chip. And NVID- and NVIDIA was like, "Well, we're not gonna do that."

Guido Appenzeller 30:25 A lot faster at least, right?

Dylan Patel 30:27 Well, more... Like, like if you compare-

Guido Appenzeller 30:29 Like, every SRAM

Dylan Patel 30:29 ... the amount of memory, of SRAM on NVIDIA's chips, it's much, much lower-

Guido Appenzeller 30:32 Yes, correct

Dylan Patel 30:33 ... than-

Guido Appenzeller 30:33 They went SRAM instead of DRAM.

Dylan Patel 30:34 Right.

Guido Appenzeller 30:35 But, but then they usually have less DRAM, so there's a trade-off there as well, so there's-

Dylan Patel 30:37 Right. There's less DRAM, there's more SRAM. And because there's more SRAM on the chip, you have to have less compute on the chip, and so they ended up losingRight? Um, because the model sizes got too big and all this, right? Um, and so you have this like super weird dynamic where, uh, they bet on something that was actually better, right? Like I have no doubt that Cerebras would run certain types of models better than NVIDIA or Groq or, hey, Dojo, right? Dojo runs certain... You know, in Tesla's Dojo would run certain types of models way better than NVIDIA's chips, uh, because they're optimized to that. But then it's like, oh, well, actually even in vision tasks, you use vision transformers now. So it's like, okay, cool. Um, because model sizes grew and all these things, so it ends up being a, you know, catch 22 in that like you optimize for something. And so now, like today you have this new age of AI accelerator companies that are like, "Okay, we're gonna optimize for transformers." But the time they started designing, they're like, "Okay, transformers are dense models that are this big. What's the best... You know, the hidden dimension is eight K and your batch sizes are this big and your sequence lengths are this big, so let's just make a super large systolic array," um, so you can, you know, create the maximum efficiency. And it turns out, oh, look at DeepSeek, or, you know, go look at what the labs are doing. Actually, their, their shapes are much smaller. Actually, you need to m- do a bunch of small matrix multiplies, not massive, massive, massive, you know, singular matrix multiplies per layer. And then it ends up, you know, oh, well, that chip you're designing for that is actually not super effective for-

Host 31:58 Yeah

Dylan Patel 31:58 ... that. And so the software is evolving constantly because of what mo- because of what, um, works best on NVIDIA. And you see that with, you know, whether it be what DeepSeek's doing or Alibaba's doing, or what the labs are doing internally. Um, and you even see this like for Google, right? Like their open source Gemma models make different decisions because the shapes of a TPU are different than a GPU. Um, and those-- the GPU and the TPU are actually not that far apart, right? Like you would say, yes, they're very different, but like Blackwell and TPUs are, are very, very... They're converging on similar designs actually. Um, whereas like to beat NVIDIA, you can't just have the supply chain, you know, win, right? You don't have this captive customer. So now you need to do something, you know, that will give you five X advantage, right? In hardware efficiency for a certain type of workload, and then pray the workload doesn't shift, right? Because NVIDIA's also optimizing their architecture generation. They've added a lot of stuff to, um, make their model, their chips way better for the existing models. But it's like they're taking, you know, large steps every year, every two years towards something, whereas you have to like go way over there in left field and hope that models stay over there, right? Um, because you have to win by five X because NVIDIA's gonna have supply chain efficiency over you. They're gonna have time to market over you in terms of like a new process node or new memory or, you know, whatever, whatever technology, right? Even AMD, right? They got to two nanometer before NVIDIA. Um, they had higher density, uh, HBM. Um, they used 3D stacking. All these things on supply chain that should be better than NVIDIA and yet they still lose.

Host 33:32 Th- there's still the software angle, right? NVIDIA has fantastic software.

Dylan Patel 33:35 Yeah. And then there's software as well, right? But it's like NVIDIA's gonna have better networking than you. They're gonna have better s- uh, HBM. They're gonna have better process node. They're gonna come to market faster. They're gonna be able to ramp faster. They're gonna have better negotiations with whether it's TSMC or SK Hynix and the memory and silicon side or all the rack people or like copper cables. Everything they're gonna have better cost efficiency, so you have to be like five X better.

Host 33:55 But, but it-- to be fair, if, if somebody had a viable competitor, which would even be marginally cost competitive, my guess is many of the big consumers of GPUs would immediately shift some revenue there just to have a number two, right? Just to, just to turn that into-

Dylan Patel 34:05 Well, that, that's, that's AMD today, right? And Microsoft stopped.

Host 34:08 Somewhat, yeah.

Dylan Patel 34:09 I mean, like the, the-

Host 34:09 There's still pretty limited traction though, right?

Dylan Patel 34:12 Sure. But-

Host 34:12 Yeah

Dylan Patel 34:13 ... Meta continues to buy from them, and Microsoft did buy a bunch and then they stopped because it's like, well, yes, they're, you know, AMD is giving you all these advantages, but it ends up still not being better on a performance per watt basis, and they have a way bigger software team. They're somewhat competitive on like all these dynamics that I mentioned, right? So you can't just like do the same thing as NVIDIA. You really ha- and, and do it better, right? Or try and execute better like AMD. Like you have to really leap forward in some other way. But that's... The s- the design cycle takes so long that models will shift, right? Because they're like, "Oh, what's the next generation TPU and GPU look like? Okay, let's optimize for that."

Host 34:47 Mm.

Dylan Patel 34:47 And the research path is, you know, like great. Like yes, neuromorphic computing could be the most optimal thing for us to do, but no one's working on that because you have to advance in the tech tree you've chosen, right? If you restart the tech tree, you're gonna be like, "Well, this sucks." Um, and so like if it branches this way and you're over here, you're screwed 'cause you have to be five X better.

Host 35:08 There's a, there's, there's a mode.

Dylan Patel 35:08 Uh, because the supply chain stuff means that five X actually turns into a two and a half X and then NVIDIA can compress their margin a little bit if you're actually competitive, and then that two and a half X becomes like a fifty percent better. And then, yeah, so it's like it ends up being way too difficult to, um... And the software stuff, right? Everything like takes your five X and makes it like, oh, you're actually only fifty percent better.

Host 35:27 And d-defense supply chain for sure.

Dylan Patel 35:29 Yeah, defense supply chain, and then like they get, they get that, right? Like, so it's like... And, and Lutnick himself said we had to do the, this for rare earth minerals. And it's like interesting. Um, China-- there's like provinces in China that have like rules that say the H20 is not efficient enough to be deployed, which is like super bizarre because it's clearly the best AI chip China has. Um, Huawei's still a little bit behind. Um-

Host 35:54 Well, what's interesting is that y- y- you know, e-efficiency is just not, is, is so much less of an issue in China than here because they just have the power infrastructure to be able to support. So even if they're running less powerful chips, you know, you would imagine that it doesn't really matter because Ch-China has just such an infinite supply, infinite supply of power that, you know, they'd sort of be okay with it. So it's interesting.

Dylan Patel 36:14 Y- which is a, it's a big challenge in America, right? Like, um, there have been, there have been companies that were like... They would, they, they've like, you know, Jensen keeps saying he couldn't give away H20 in America for free, but I've literally like heard companies like say, like now say like, "Yeah, no, I mean, I wouldn't because like I only have this much power." Um, how am I going, you know, in data centers ready to go, uh, over the next year? If I bought an H20, I'd literally have less compute capacity and then I'd lose, right? Even if it was free, like it doesn't make sense. Um, whereas China doesn't care. They can build these things. Um, they have the muscle. Um-I'm, I'm curious how this all shakes out. You know, China's posturing really hard. They even, like, put out something that was like, "We're investigating to see if there's backdoors in the H20." It's like there's no backdoor in the H20, like chill. Um, G- you know, it's like, you know, GPU, GPUs are usually, like, firewalled from the public internet anyways. Like you, you step through stuff before you get to the GPU clusters. Um, so like a back wall- backdoor wouldn't even matter. Um, I don't know. I think, I think it'll be, it, it'll be interesting to see because, um, China can definitely deploy way, way, way more power to AI the moment they decide to. Um, but there's these, like... There's, like, competing interests, right? Like-

Host 37:31 Because they want Huawei to be better than NVIDIA.

Dylan Patel 37:33 Yeah. And then this is how NVIDIA argued to the administration. They're like, "If we don't do this..." Actually, I think it's like a very, like, s- powerful argument that like, like, for example, within Triton, which is a common ML library anyway, like, like ByteDance, uh, has open-sourced some stuff that plugs into this that is, like, super awesome, and there's, like, all these other libraries. It's not, it's not just models that China open-sources. It's like software for NVIDIA that Chinese companies open-source. Um, in a sense, like by NVIDIA selling GPUs is NVIDIA's argument again, like, was, like, uh, they were able to, you know, stop Huawei from building up a software ecosystem-

Host 38:08 Mm

Dylan Patel 38:08 ... and the Western ecosystem is better. Um, but then on the flip side, it's like, again, if you believe the models deliver more economic value to society than the hardware, which, which I actually think they do, it's just there's a value capture problem today, um, then you're giving China way more by giving them H20s, and soon a version of Blackwell that's cut down, like, like Trump said, right? Um, versus, versus, you know, selling them the chips, right? The economic value derived from selling them the chips is not as large as, you know, being able to somehow sell them AI services.

Host 38:38 So is, is China gatekeeping power for AI?

Dylan Patel 38:42 Um, I don't think so. I think, I think, again, like there's a lot of like... We, what we see is that, like, even with H20 being sold to China, in Ch- into China, um, and, and, and future versions of the chip, the H20e and other chips, um, we still see, like, Chinese companies like Alibaba, uh, renting GPUs outside of China because the GPUs they can get outside of China are just so much better on a dollar spend-

Host 39:08 Mm-hmm

Dylan Patel 39:08 ... per performance basis. Renting them or even, like, going through sort of like a Singaporean company that is effectively a Chinese company and building data centers, um, and, and putting chips in them. So it's like I don't think China's limiting the power per se. It's that it's, you know, you can only... If you can spend, like, your-- Chinese companies are growing their CapEx, uh, way more than US companies on a percentage basis next year. The doll- absolute dollar number is, you know, obviously the US companies are spending more still on AI. Um, the percentage basis, Chinese companies are growing more next year, and you, you still have the problem of like, well, dollars spent to AI output, uh, in tokens or in whatever is gonna be lower because these chips are worse. So power is not the gating factor. It's always capital, right? At least today, right? Um, now China can spend a lot more capital if they wanted to. Um, they're subsidizing the semiconductor industry to the tune of like a hundred fifty, two hundred billion dollars a year, uh, through SOEs, through CapEx that's not generating revenue, et cetera. So it's not like they couldn't do this to the AI ecosystem, right? Given, you know, Meta's CapEx is like sixty billion, right? Um, and Google's CapEx is like eighty billion, right? Like, they could totally spend way more than that on a single effort. They just haven't decided to. Um, and I just think for the US, our build-outs are constrained by power, right? Like Google has a ton of TPUs sitting, waiting for data centers to be powered and ready, as does Meta with GPUs, right? We posted about how, uh, Meta is now building these like effectively tents.

Guido Appenzeller 40:35 Isn't this to some degree also coupled to their unwillingness to go sell into a broader ecosystem? I mean, if they want to be confined in their own data centers and they're, you know, uh, didn't ramp data center, uh, build-out for their own hypers- for their own sort of hyperscale use cases quickly enough, right? Then yes, that constrains them, right? If, if they were on the open market, will we still be constrained?

Dylan Patel 40:54 Yeah. Yeah, for sure, 'cause, like, companies like CoreWeave, you know... W- why is CoreWeave valuable is really because they build infrastructure really fast, right? Um-

Guido Appenzeller 41:03 Yeah

Dylan Patel 41:03 ... and, and their software is nice, I think, but like a lot of their customers are bare metal, right? Just-

Guido Appenzeller 41:07 That's all

Dylan Patel 41:07 ... just replace the GPUs whenever they're broken and network the problem-

Guido Appenzeller 41:10 They, they grow more aggressively, and I, I think Jen-

Dylan Patel 41:11 And they'll, they'll go anywhere

Guido Appenzeller 41:12 ... Jensen likes them as well, to be fair.

Dylan Patel 41:13 Yeah. They'll, they'll go... Yeah, yeah, that's very important as well. But, um, they'll like go because it's, it, it, it, it unconcentrates the ecosystem, which is better for NVIDIA, right?

Guido Appenzeller 41:22 Having worked at Intel, I know exactly what, what's going through his mind. [laughs]

Dylan Patel 41:25 Yeah. So, so I think what's really important is that, like, CoreWeave doesn't care, right? They're like, "Oh, crypto data center? I will convert it to a AI data center," right? They bought a company for like ten billion dollars that's doing crypto mining, which was worth like two billion dollars like a couple years ago, and it's not because their Bitcoin mining business is growing. It's because they have powered data centers, right? Like anywhere and everywhere people are trying to build powered data centers, um, and companies like, like CoreWeave and Oracle are moving to the... Actually, today Google just did bought eight percent of a crypto mining company-

Guido Appenzeller 41:57 Mm

Dylan Patel 41:58 ... um, called TeraWulf, right? Um, and-

Host 42:01 Not because they're getting into crypto mining.

Dylan Patel 42:03 No, because they need the data centers, right?

Guido Appenzeller 42:04 Probably not. No.

Host 42:04 They want the power. [laughs]

Dylan Patel 42:05 They need the power, right? And it's like, um, all the hyperscalers have like said screw off to my sustainability pledges because they need power as fast as possible.

Guido Appenzeller 42:13 Yeah.

Dylan Patel 42:13 Right? Um, they're, you know, they're, they're doing things that are not, that, they're, they take a little bit longer to move the ship. But like even if you didn't do it in your own self-built data centers, there's still a lot of challenges in the open market. There's a deficit, right? Um, and that's, that's constraining American chip build-outs heavily. Um, yes, others could maybe do it, uh, a little bit faster, like CoreWeave or, or others, right? Um, Oracle's got a open mind as well. Um, but it's still constraining US build-outs heavily, um, e- even, even though the capital's been spent, right? The chips are, you know, sixty to eighty percent of the cost of the cluster, depending on what chips you're getting. So it's like-They've already bought the chips. They just can't put them anywhere because the data centers aren't ready. So applies to Google, applies to Microsoft, applies to Meta, applies to a lot of folks. Um-

Host 43:04 I mean, it's really hard to build infrastruc- power infrastructure-

Dylan Patel 43:06 Yeah

Host 43:06 ... in the US, so.

Dylan Patel 43:07 Power, grid interconnections, transmission, substations, uh, all of this stuff like, like electrical contractors, uh, electricians, um, in Texas, if you're willing to like be a travel electrician, it's like, it's like oil pay, right? Like it used to be that like if you're physically adept, you could go make, you know, 100 grand in West Texas. But like who the fuck wants to do that? Um, now it's like, well, you could, you could go like 200 miles away from Dallas and what's still a reasonable town, um, and, and build a data center and work on the wi-wiring within the data center and all this other stuff, uh, the transmission stuff, and your pay is up like 2X now, uh, versus what it was just a few years ago. This labor problem's a challenge too, and it's, it's... Yeah. I think in China, they don't have any of these problems, but they just haven't spent the capital yet. But capital is an issue as well, um, because of the scale of what's being spent, right? Like, like NVIDIA's revenue this year is gonna be like over $200 billion, and next year expects over 300 billion, plus Google's gonna spend like $50 billion on TPU data centers, right? And it's like, and Amazon's gonna spend tons and tons on Trainium d-data centers. It's like the scale of dollars is, is quickly growing to, um, nation state level stuff. Um, and, and what's more important is being able to decide to spend the dollars and what's cost effective. And so to some extent, China's still constrained by that, but they can smuggle chips in, they can build data centers outside of China, they can rent data centers outside of China, um, and be more... and have the most cost effective, you know, Blackwell chips or whatever, right? ByteDance is, you know, either the s- biggest or the second biggest customer of Google Cloud for a reason, right? And they're getting, you know, over... You know, they're getting many, many Blackwell from them, right? And the same with Oracle and the same with Microsoft and all these other companies are renting tons of chips to China anyways because it's more cost effective to do that than build it yourself. So it's not like China has this mentality where we only have to... Well, the government does, but the infrastructure companies don't, um, like Alibaba, Tencent, ByteDance, et cetera.

Guido Appenzeller 45:06 So what, what, what's the end game for data centers? I mean, like, like we, we need more power. We need more cooling. Will at the end be every... all, all data centers be next to a nuclear reactor or lots of solar, you know, next to a deep level, uh, uh, like, um, deep seawater, uh, that we use for cooling or something like that? Or, uh, what's-

Dylan Patel 45:23 Um-

Guido Appenzeller 45:24 Where's this heading?

Dylan Patel 45:24 I think that like cooling is... Like the physical cooling of a data

Dylan Patel 45:28 center are like... You know, there's this whole narrative about like, oh, AI uses so much power, and it's like not really. Um, you know, farming alfalfa uses like 100X the water of, of AI data centers. Uh, even by the end of the decade it'll be the same and it's like alfalfa's like worth very little. So it's like, I know. There's like, it's like cooling is like, not that... You know, people have like experimented with like, you know, undersea data centers to reduce the cooling cost, but it's like-

Guido Appenzeller 45:50 Yeah, that doesn't make sense [laughs]

Dylan Patel 45:51 It's like five, 10% savings-

Guido Appenzeller 45:52 It's-

Dylan Patel 45:52 But then like if you wanna-

Guido Appenzeller 45:53 It's easier to get the water out of the ocean than, than to put the data center into the ocean I think [laughs]

Dylan Patel 45:57 It's like if you wanna service it, like you're screwed, right? So like the same with power, it's like we talk a lot about like the power is not actually that expensive. It's just hard to build, right?

Guido Appenzeller 46:05 And hard to get to the right place, right? I think you said.

Dylan Patel 46:06 And hard to get to the right space and convert it down to the voltages and, and all this stuff that, that chips need. But like-

Host 46:13 So it's less the magnitude of power and more where it is and how it moves.

Dylan Patel 46:16 Well, the magnitude too, right? Like it's gonna be-

Guido Appenzeller 46:18 In, in terms of total worldwide energy consumption, AI data doesn't, is still a, a-

Dylan Patel 46:22 It's like-

Host 46:23 Very small. Right. Nothing.

Guido Appenzeller 46:23 It's like, yeah.

Dylan Patel 46:24 Even-

Guido Appenzeller 46:24 It's a fraction of a percent. It's not-

Dylan Patel 46:25 Yeah, yeah.

Guido Appenzeller 46:26 Yeah.

Dylan Patel 46:26 Even by the end of the decade, uh, you know, g- the US will be like 10% of our power will be AI data centers, which is still like-

Guido Appenzeller 46:33 Of electricity.

Dylan Patel 46:34 Of our electricity.

Guido Appenzeller 46:35 Yeah. And in terms of energy, that's even a smaller fraction, right?

Dylan Patel 46:37 Oh, yeah, yeah. 'Cause you think about vehicles-

Guido Appenzeller 46:38 By shifting to electric vehicles also, you can probably make a bigger swing than, than, uh, you know, with all the AI data centers we can build in the first place.

Dylan Patel 46:44 But outside, like it's like in Europe, like that number's not moving up that fast and like all these other countries. I, I think we need to build a lot more power, but it's not like some, some crazy, crazy like amount. It's just like doing it properly is the hard th-

Guido Appenzeller 47:01 Yeah, I agree.

Dylan Patel 47:01 Um, and, and, and again, like the cost of power, like you go look at like these deals people are signing, they're still signing like even though the price has skyrocketed, uh, from like a few cents a kilowatt hour for these massive, massive purchases to like 10, it's still, you know, when you think about the full TCO of a c-cluster, you know, the GPU cost, the networking-

Guido Appenzeller 47:19 It's a couple percent, yeah

Dylan Patel 47:19 ... all of that stuff far outstrips the power.

Guido Appenzeller 47:21 Yeah.

Dylan Patel 47:21 And same with cooling. Um, it's like not a-

Guido Appenzeller 47:23 But what, what percentage is power from... Like if you, if I do a-

Dylan Patel 47:26 In, in simple-

Guido Appenzeller 47:26 ... a four-year amortized, uh, uh, GPU data center, what percentage would be power?

Dylan Patel 47:30 80% of the cost of a GPU data center if you're building Blackwell is capital.

Guido Appenzeller 47:33 Yeah.

Host 47:33 Mm-hmm.

Dylan Patel 47:33 Right?

Host 47:34 Yeah.

Dylan Patel 47:34 It's the GPU purchases, it's the networking, it's the, it's the physical, um, data center, uh, conversion, power conversion equipment. Um, all of this stuff is like 80% of the cost, and then 20% is gonna be your land, and your power, and your cooling, and your cooling towers, and your backup power, and your generators, and all this stuff. It's like nothing. Which is why it doesn't matter if you spend, you know, 10% or 50% more on that, um, because at the end of the day, the expensive thing... Right? Like this is why what Elon did would seem silly, right? They spent a lot more money on, you know, generators outside the data center and these mobile chillers to cool the water down for their liquid cooling instead of like the more cost-effective option because it got the data center up three months faster. And so like that three months of additional training time is worth way, way, way more on a TCO basis, right? The performance you got out of the chips and the time to market and all this is way, way faster and therefore it was the right decision even though this part of the data center ballooned it cost, everything else is still there and you're still paying for the chips. And if they were sitting idle, it's not worth it, right?

Host 48:39 Just by like bypassing the grid, bypassing anything to do with interconnect, anything to do with public utilities.

Dylan Patel 48:44 Exactly. Exactly.

Host 48:45 Yeah.

Guido Appenzeller 48:45 What's your take on Intel? Where, where's Intel going?

Dylan Patel 48:48 I think the world-- Well, the US needs Intel.

Guido Appenzeller 48:51 National boundary [laughs] .

Dylan Patel 48:53 I think the world needs Intel. I, I think the world needs Intel because, like, Samsung is doing worse than Intel on leading-edge process development-

Guido Appenzeller 48:58 Mm-hmm

Dylan Patel 48:59 ... in my opinion, um, based on even on various customers in the industry having done test chips at, like, Intel versus Samsung. They think-- I think the industry generally agrees that Intel is further along, uh, you know, the sort of the two nanometer class process technology than Samsung is. Uh, but both are way behind TSMC. Um, and TSMC is, is a monopoly in, in some extent. Um, the number one question always people ask is, like, "Why is TSMC not making more money? Why are they only raising prices, you know, next year, you know, three to ten percent depending on what it is?" It's like TSMC is a monopoly. Like, they could raise a lot more, but they're, um, they're good Taiwanese people rather than, like, uh, dirty American capitalists. If TSMC was owned or was managed by Americans, I think most ownership is actually American, um, in terms of the stock-- It's on the New York Stock Exchange and all this. Like, you know, they would have raised prices a lot more. Um, and, and so, like, there is this, like, difficult, difficult, uh, thing to be done that like, hey, there's one island that controls all leading-edge semiconductors and not just all leading edge, like the majority of trailing edge production as well.

Guido Appenzeller 50:02 Yeah.

Dylan Patel 50:02 Um, something needs to be done. Intel is behind, but not like, not like absurdly so, right? Like, if something were to happen to Taiwan, Intel would have the most advanced technology in the world, right? Uh, it's just, it's not economic.

Guido Appenzeller 50:14 Can, can you keep Intel as one company if you want, want them to be competitive?

Dylan Patel 50:17 I think the process of splitting it would take so much executive time and so much executive effort that you would been, have been bankrupt by then, right? And that's the big challenge. Like, I think Intel should be separate, right? Um, but to properly split the company and for all the management time that's needed is, like, absurd. And instead, like, what you need is, like, you need Lip-Bu Tan, who's the CEO, who, who's CEO of Intel. Um, you know, there's a lot of drama going around about him 'cause he's, he's one of the greatest semiconductor investors ever,

Dylan Patel 50:46 right? He's invested in so many different companies first. Um, you know, he was on the board of, like, SMIC, which is China's TSMC effectively, which is, like, a big, like, drama or, like, the big- the... Some of the biggest tool companies in China. He's the first investor in them because, you know, there's a multipolar world there, and he's making good investments. Um, but like, you know, now, like, people are getting mad about that, but it's like, no, he, he, he recognizes, he, he... The companies, like, he understands the supply chain. He needs to not spend his time on splitting the company because then he never actually fixes the company, right? Intel's problem is that, like, it takes them five to six years to go from design to shipping the product, um, in some cases more. Uh, and when they tape out a chip, right, like, you know, you send the design to the fab, the fab brings back the chip. They go through fourteen revisions in wha- in some cases, whereas, like, the rest of the industry goes through, like, one to three, right, revisions if they're good, um, of like send the design in, uh, get the chip back, test it, send the design in, right, uh, for, for a public launch, and they'll launch a chip in three years.

Guido Appenzeller 51:45 So but i-i-if you look at Intel today, right, they still don't have a competitive entry on the, uh, on the AI side.

Dylan Patel 51:51 And they won't.

Guido Appenzeller 51:52 Uh, uh, right. C-can you... So what... But what does it mean for their offering? I mean, they, they, they're still doing great on, on, on CPUs. They don't have a good AI, uh, AI chip product. Is it long-term sustainable positioning, right, I mean, as, as, as a, as a standalone chip company?

Dylan Patel 52:05 I mean, IBM still makes more money every launch off of mainframes. So it's not like, it's not like X86 is dead. It's like you don't get the growth rates, but, like, you could totally run this as a very profitable enterprise. Um, and, and I think the same with PCs, right? There's, you know, there's some turmoil, there's some Arm entry, there's some AMD competition and Apple-

Guido Appenzeller 52:22 They don't care about most of them, yeah

Dylan Patel 52:23 ... but, like, I think it's a very, it can be a very profitable business if it had, like, one-third the people or half the people working on it. And so, like, Lip-Bu Tan to fix Intel needs to go into both the design company and lay off a shitload of people, but, like, keep all the good people, um, and make sure that they're designing fast and they're launching from design conception to launch is two to three years, not five to six. Um, and, and that's on the design side and make that profitable. And then on the fabs, he has to do the same thing. There's all these people, like, um, one of the heads of, of, uh, fab automation at Intel-

Guido Appenzeller 52:55 Yeah

Dylan Patel 52:55 ... um, I, I explicitly told Lip-Bu Tan because, you know, we have a couple ex-Intel people who are actually good in the company, um, that worked on the fab side and were like, they were, were like, "Who's the worst people in friends?" It's like, "Oh, this guy sucks." I explicitly told Lip-Bu Tan. He had never talked to the guy because it was like four layers down. The company has, like, absurd amounts of hierarchy. It's like four layers down. He, he goes and talks to the guy and he's out, right? It's like, like he figures out, like, who's bad, right, um, and who's good. And he has to go in and he has to like, "Hey, the vast majority of the team at Intel is the one who led the world in production and process technology for twenty years."

Guido Appenzeller 53:29 Yeah.

Dylan Patel 53:29 Right? But there's a lot of, like, built-up crap. So he has to go figure this out, right? He can't waste his time on like, oh, all this like structuring to split. Like, I w- I think it would be better if the company split. I just don't think he can spend the time to do that. Um, and if the design side of the company is, you know, you're not really gonna get into AI, you're not really gonna... You, you have to make some money there. But the fabs I think could truly become a competitor. But they're gonna go bankrupt by the time anything, you know, can happen. So they have to figure out how to get capital. So he has to figure out how to get capital. He has to figure out how to clean up all the crap-

Guido Appenzeller 54:02 Yeah

Dylan Patel 54:02 ... uh, make the, um, you know, yields go up, right? Make the product ship way faster. Like, all of these things are basic problems.

Guido Appenzeller 54:09 I think, I think the goals are completely correct. I mean, I think, I think the big challenge, just reflecting back on my time there, right? The... I think the big challenge is that right now if you look at Intel, right, they have essentially software, sort of the, the chip design making the... And then there's, you know, the core manufacturing part, right? And they have three diff- very different cultures, and it's very hard to get everything under one umbrella, right? And, and so I think that is the big challenge, right?

Dylan Patel 54:29 I think you c- you should even run-

Guido Appenzeller 54:30 But he has a lot of money

Dylan Patel 54:30 ... the company separately, right? Like, but, like, you can't physically separate them entity-wise because it's gonna take so long to sever all these things, um, because he doesn't have time, right? Like, Intel is literally gonna go bankrupt if they don't have a big cash infusion or they lay off like half the company, right? Uh, which some could argue you need to lay off like thirty percent of the company anyways. But, um, there's a lot of bad things that happen if that happens, right? Um-And they need to spend a lot more on building the next generation fab, even if they fix the fab, and they don't have money for that, right? So there's like, there's like a lot more, more important problems than like physically separating the company, even though I think long term, yes, the fab has to be separate from the chips design software, right? Like or chip design, uh, part of the company. Just like that's gonna make each company much more accountable, uh, be able to service their customers better, etc. It's just that's gonna take too long and they're gonna go bankrupt by then.

Erik Torenberg 55:21 Awesome.

Dylan Patel 55:22 I, I but I think, I think, I hope, I hope, I pray someone does something, right? Like you get a big capital infusion. I don't know. The big hyperscalers are like muscled into like, oh, okay, wait. If TSMC eventually grows their margin to seventy-five percent because they're the monopoly, um, plus they intake all this stuff like co-package optics and power delivery and all this, like all of a sudden the cost is gonna spike. So we should actually just throw five billion dollars at Intel each, right? Screw it. And, and that could actually give Intel enough of a lifeline to potentially get to something and maybe be competitive, um, that's, that's the hope.

Erik Torenberg 55:53 Can we, uh, can we finish by finishing th-this game that we started when we gave Sam Ad-- uh, Altman advice? If, if Jensen was here, what, what advice would, would you have for him?

Dylan Patel 56:03 Hmm. If Jensen was here...

Erik Torenberg 56:06 Ha.

Dylan Patel 56:06 Uh, you know, I think, I think he has a massive, massive balance sheet, right? Jensen does. He's, he's cash... Free cash flow is like, you know, ridiculous. Um, the tax cut, the new, the new Trump, you know, tax bill, um, institutes something really incredible, which is that you can depreciate all of the GPU cluster cost in year one. Um, which we put out like a note about how like the tax implications to like Meta are like ten billion dollars a year. And across each of the major hyperscalers is like massive. It's like, well, NVIDIA's gonna spend tons and tons of cash, um, or ca- uh, they're gonna spend like, you know, tens of billions of dollars of taxes. Why don't you get into the infrastructure game somehow? Um, now this is obviously gonna be like crazy because like now they're buying GP-- their own GPUs and putting them in data centers and doing stuff, and they're competing with their own customers, but they're already doing that anyways because their customers are trying to make chips. Um, but they should like accelerate the data center ecosystem with investments, right? Uh, because really we, we think we can like have very high degree of like accuracy on what they're gonna do next year in terms of revenue, because it's just the number of data center watts that are being built, right? Like this is harder thing to shift up and down, right? Now, there's a little bit of share difference between how much is TPU versus GPU, but like, it's like you have to accelerate the infrastructure and you need to spend all of this capital that you're building, right? Like, okay, do you wanna go the route of like doing buybacks and dividends? Like great, like you're a loser if you do that, right? Like you, you can make more money by reinvesting and building a bigger company that's not just chips into the ecosystem or servers into the ecosystem, but actually like controlling the infrastructure end to end somehow. Um, so I think there's something he could do there, uh, with this massive war chest. And there's a reason like NVIDIA's done some buybacks and they've done some dividends in-increasing,

Dylan Patel 57:56 but the cash on their balance sheet keeps growing, and they're gonna have north of a hundred billion dollars of cash on their balance sheet by the end of this year, I think. Um, so it's like, what are you gonna do with that? Um, I think, I think there's something moving into the infrastructure layer much more, um, that they, they could do if he really wants to be the king of the world, right? Uh, which I think he does.

Erik Torenberg 58:16 Uh, Sergey and Sundar?

Dylan Patel 58:18 Um, ooh. Um-

Erik Torenberg 58:20 [laughs]

Dylan Patel 58:21 I, I think, I think they should open up the Komodo on TPUs, right? Like start selling them. Um, open up the software. Open source a lot more of the XLA software, 'cause there's OpenXLA and there's XLA, but the vast majority is closed source. Um, really, really open up the Komodo on that, um, and be a lot more aggressive, right? They've-- they're still pretty not aggressive on data centers. Um, they're pretty not aggressive on a lot of elements of the company. Um, the TPU team's next gen dir- designs are pretty not aggressive, partially because a lot of the TPU team, uh, has left to go to OpenAI, the best people that I knew. Uh, it was actually really annoying. I knew like four people, uh, or five people, and they all went to OpenAI [chuckles] and it's like, fuck, like now I don't get as much ins-

Erik Torenberg 59:03 It's alright.

Dylan Patel 59:03 I met some other people, right? But it's like, you know... Um, I think they could be a lot more aggressive in many ways across the company. They don't have to be, right? But they could.

Erik Torenberg 59:10 Right.

Dylan Patel 59:11 Um, because AI, you know, like this ChatGPT take rate, the shift of search queries, the monetizable ones, especially from to, to purchasing agents, is gonna really screw Google long term if they don't, you know, get their act together. I think they've gotten their act together on DeepMind. There's still some inefficiencies, but Sergey works on, works within DeepMind a lot and they're driving hard. They're still a little bit behind, but like, I think like physical infrastructure, TPU, um, and, and how much money they could make and how much they could take the wind out of everyone else's sales-

Erik Torenberg 59:40 Yeah

Dylan Patel 59:41 ... if they start selling TPUs externally, um, and reorg around like building data centers much faster so that they do have the most compute in the world, 'cause they did. Uh, but now there's certain companies that are gonna surpass them, uh, potentially over the next few years if they don't really get their act together. So I think that's what I would say for them. Um, yeah. And also like, like learn how to ship product better, right? [laughs]

Erik Torenberg 1:00:03 [laughs] Um, Zuck.

Dylan Patel 1:00:05 Um, I think, I think Zuck... You know, it remains to be seen what goes on with superintelligence, but like they're trying to move for super fast with the data centers. Uh, you know, like screw it, we'll build tents, uh, instead of like physical data centers 'cause we only need these for five years anyways. Um, you know, the superintelligence moves, you could, you could say whatever you want, but like, you know, trying to buy like Thinky for like thirty billion or, or, um, SSI for thirty billion didn't work out, so then they spent, you know, not even that much on hiring, not thirty billion on hiring all these people. Um, so I think they-- he recognizes the urgency with the models, with the infrastructure. Um, so I really think he needs to like... You know, i-if you read his website post about like AI, like I think, you know, he sees the vision, right? There's the wearables, there's integrating AI into that. There's being your AI assistant to do all this purchasing and stuff. I think he sees the vision, but I think he also needs to focus on like actually likeReleasing that faster, um, but also, like, the products that they do outside of their core IP every time they launch something is kinda mid, right? Um, you know, uh, Meta Reality Labs is doing well, but I think they should, like, go more explicit, like have a ChatGPT competitor, have a Claude com- uh, a Cl- like Claude Code competitor. Like, just start releasing way more products, um, because they're really just focused on their individual gardens rather than, like, branching outside of it.

Erik Torenberg 1:01:27 D-do you think Apple should have that same sense of urgency? Or if Tim Cook was here, what, what would you tell him?

Dylan Patel 1:01:32 The funny thing is, like, some of their best AI people are now, like, at Superintelligence. [chuckles] Uh, they're building an AI accelerator. They're gonna... They're, they're, they have AI models, but they're just, like, way slower. Um, they did mention on the last earnings call they're gonna allocate more capital to this, but it's like, guys, Apple, like, you guys are gonna lose the boat if you do not spend, like, 50, $100 billion on infrastructure.

Erik Torenberg 1:01:51 You don't think the current Siri will, will cut it? [laughs] Sorry.

Dylan Patel 1:01:54 [laughs] Um, I think, I think, like, more and more you'll see people, like... You know, great, Apple has this walled garden, but, like, they can only do so much to protect it, right? IDFA, like, they shut down ads to or data sharing to Meta. But Meta b-made better models, and now they have way more data and way more power over the user than they ever did before. Kind of it was good that Meta kicked the crutch off of them, or, or Apple did. Uh, but the same applies to, like, AI. Like, yes, they have access to the text, and they have access to this, but, like, I think other people are gonna be able to integrate user data, um, and agents will be able to integrate all this user data, and they'll start to lose control of what the user experience is as more and more gets disintermediated by AI being the interface rather than touch, rather than, you know, touchpad and keyboard. Um, and I don't think they've truly realized what happens when the interface to computing is, is AI. Like, they, they market it, but, like, that's gonna shift computing really heavily. They have great hardware, and their hardware teams are working on awesome stuff and form factors. But, like, I just don't know if they, like, get what is actually gonna happen to the world in the next five years, uh, truly well enough, and they're not building fast enough for it.

Erik Torenberg 1:03:05 What about Microsoft to, to that end?

Dylan Patel 1:03:07 Um, Microsoft has the same problem, I think. Um, they were super aggressive in '23 and '24.

Erik Torenberg 1:03:13 Yeah.

Dylan Patel 1:03:14 Um, and then they pulled back heavily, right? Uh, now, like, OpenAI's slipping through their grasps. Uh, there's that whole thing there. Uh, they cut back on data center investments heavily.

Erik Torenberg 1:03:23 Yeah.

Dylan Patel 1:03:23 They were gonna be the largest infrastructure company in the world by, like, a factor of 2X.

Erik Torenberg 1:03:27 Yeah.

Dylan Patel 1:03:28 Uh, which would've been... You know, you could argue maybe that was, like, too much, and maybe it wouldn't have been economical. Um, but, like, they're losing grasp on OpenAI. Their internal model efforts are failing, uh, spectacularly. Like, they're on LLM arena right now, and they're pretty decent there, but it's like that's just, like, a sick authentic model. Like, um, it's a code name but, like, whatever. Like, MAI is, like, failing. Azure is, like, losing a lot of share to Oracle and CoreWeave and Google, um, and so on and so forth, right? Their internal chip pro- chip effort is by far the worst of any hyperscaler. Like, uh, they're just, like, misexecuted. Like, GitHub... How is GitHub not the highest ARR software, uh, code model?

Erik Torenberg 1:04:09 I mean, they, they, they only had the best IDE, the best source code repository, the best enterprise Salesforce, the best model company as a relationship, and they were the first to market, right? [laughs] It's like they had everything going for them.

Dylan Patel 1:04:20 And, like, there's just nothing, right? It's like-

Erik Torenberg 1:04:22 I'm-

Dylan Patel 1:04:22 ... like Co- GitHub, uh, GitHub Copilot is failing. Microsoft Copilot is, like, still crap, right? Like, um-

Erik Torenberg 1:04:28 Yeah, it's unusable.

Dylan Patel 1:04:29 It's like what, what is go- Like, you know, what is going on? Like, you need to shake the crap out of the company. Like, I think they win a lot because they have the best business-to-business relationship with so many enterprises. But, like-

Erik Torenberg 1:04:40 Best Salesforce on the, on the planet. Yeah.

Dylan Patel 1:04:42 Yeah, but, like, they end up, like, not having the actual product to sell them, which is, like, really scary, so they need to really work on product.

Erik Torenberg 1:04:48 Yeah.

Dylan Patel 1:04:48 Um, Sa-Sun-Sat-Satya's done great on sales and stuff, but, like, yeah.

Erik Torenberg 1:04:52 I-if Elon was here, what, what advice would you give him?

Dylan Patel 1:04:55 A lot of people at xAI are mad about the porn models, like porn bo- porn stuff. It's fine. Like, you're gonna make a ton of money off of this. This is how you accelerate the revenue of that company. Um, but, like, he's, he's losing a lot of talent and, and axing a lot of good projects. Uh, but Elon has a, is a magnet to amazing talent and building stuff, so I won't bet against him. But it seems like since he left the administration and focused on stuff again... But I think... I don't know. I think, I think he's focused on a lot of things, and I think, like, Robotaxi is starting to look good actually again. Like, I haven't ridden one yet, but I have some friends who've ridden one. It's, like, looks pretty decent. He could, like, not make these snap decisions which often are the reason why he's amazing, but, like, some of these snap decisions are hurting him.

Erik Torenberg 1:05:34 Yeah.

Dylan Patel 1:05:34 I'm not sure if I can give Elon that much great advice because, uh, I think, I think maybe it's just, like, focus on, like, the products again, right, more. Um-

Erik Torenberg 1:05:42 Yeah.

Dylan Patel 1:05:43 But he is working on that stuff a lot. Yeah.

Erik Torenberg 1:05:45 I think that might be a good place to, to wrap. Um-

Dylan Patel 1:05:47 Awesome.

Erik Torenberg 1:05:48 This was a great discussion.

Dylan Patel 1:05:48 Yeah.

Erik Torenberg 1:05:48 Dylan, thanks so much for joining us.

Dylan Patel 1:05:49 Thank you for having me.

Erik Torenberg 1:05:50 Thank you.

Dylan Patel 1:05:50 Yeah. [upbeat music]