Ep. 023 - Everyone Leaves Google, Elon Forecasts 1T ARR, Reflecting On GPT-5 | Jon from Asianometry

Hello everyone. Welcome back to Semi-Analysis Weekly. We've got our second guest ever on the program today. That's John from Asianometry, top 1 YouTuber in the world, at least my favorite. Doug said it's top 2, but we won't hold him to it. This week we're planning to talk about some big news and everybody leaving Google, leaving Gemini hanging out to dry maybe, or just going to work on some of their passion projects now that they've spent their First 35 years in one job and had enough. The Chinese ban on transceivers, Elon pulling in his $1 trillion revenue forecast from 2031 to 2030, and maybe some DVDs. Guys, welcome to the show.

Thank you. Thank you for having me. This is a— it's weird. You're so professional compared to transistor radio.

We've grown up. It's so grown up, John.

Okay. I have a burning question for you, John. So, uh, one of the times when many people may have seen you talk is on the Dwarkesh podcast, where a few years ago there was this like famous clip where you're sitting beside Dylan and you say, wait, wait, wait, wait, we're assuming GPT-5 is going to be good. We've just assumed that. Do you remember this clip?

I do remember this clip.

Okay.

That was like a year and a half or 2 years ago.

So now we're sitting in August 2026. Do you feel like GPT-5 was good? Did it surpass your expectations?

I mean, I think GPT-5 was, I think the concept of GPT-5 was, uh, was probably, was probably right in the sense that like, was it good? 5 itself probably wasn't that great. As compared to others. But, you know, I was— kept going. And then I think when the really— when the Enjetic Code stuff really hit, then everything hit the, you know, everything hit the pedal to the metal. So I was very impressed by that. So, you know, but what they said, you know, when I went on the pod, when I sat down on the pod, the first thing I thought to myself, I chugged like 5 cups of coffee and I thought to myself, you know, it doesn't really matter if I'm accurate as long as I'm entertaining. That's what matters.

Hopefully we can aim for that today as well. Entertaining and questionable accuracy. Doug, what do you think? So this was October 2nd, 2024. I just looked it up when Dylan and John went on Dwarkesh's podcast and they talked about this stuff. Like, I kind of agree GPT-5 was a bit of a dud, but 5.2 and certainly now 5.6 is like, Mind-blowing.

2 years later. 5.2 was ass too.

I'll be honest.

I think 5.2 was not very good.

Okay.

Um, I think 5.6 is good. The new group, the new base, the new base model is good. 5.6, good. And also soon, soon 6.

This week. Yeah. This next week apparently is good.

Doug is coming, right?

Yeah. Doug is coming as they say. Doug is, John, I don't know if you know this, but Doug is the largest spud in the world.

Really?

Yeah, that's the name. That's how they named it. So it was like the Spud, the Spud model.

It's very—

so the Doug pre-train is done.

There's also rumors that it's going to be really good at writing. And I think that's pretty poetic that one of the models that finally really knows how to write some incisive commentary is going to be named Doug.

That's glazing.

It's gonna, it's gonna, it's gonna teach us a masterclass in, uh, in, in writing history. I'm looking forward to it.

Yeah. Hopefully it knows the difference.

It's not this, but it's that.

That's where Doug gets his, his edge.

The, uh, references to like 1940s Taiwanese, uh, well, you say that, but there's only, the only person who's probably better at it than me is John. And John's here. So John's like, I'm like, oh yeah, yeah, yeah. 1955 semiconductor cycle. That doesn't, that doesn't exist. Semiconductors weren't around then, but then John's like, yeah, here's what a DVD player is. I was like, what the hell? So the only person in the world who goes deeper and more random, I think, is John.

I've always, I like Doug for like, you know, I've read, I read this really weird book on Japanese history from the 1700s on the Mitsui family. And I'm talking about it. And then Doug's like, I read that first half. First half was great. Second half was okay.

Like, I guess when I got into the modern language, I was kind of like, eh, I don't really care. But yeah, the first half was pretty good. But yeah. Well, no, you got that from Jordan though. We all got it from the same source though.

I had like, I picked it up from my, uh, When Japan Ate Their Rich video, but I like, I didn't, I skimmed through it. It wasn't a really good, like— It's a good, it's too narrative.

It's too, no, it wasn't even narrative.

It's really, really boring.

Yeah. So this is Jordan Schneider of China Talk. You're talking about. I think it's unlikely that anybody subscribed to Semi Analysis Weekly doesn't know Asianometry or China Talk, but in case they don't, clarifying that. Interesting to get the take without Jordan on here, but John, what's your take on the Trump administration prepping to ban a bunch of Chinese data center components, including transceivers, and then possibly banning humanoid robots as well? Unitreat stuff we were talking about a few weeks ago on here.

You know, on one hand, you know, I'm generally of a— I'm generally more hawkish. I think we should be banning this stuff. You want to drive— you want to make sure that you don't have— you're not driving revenue for people, for the Chinese. But on the other hand, I mean, they have the expertise and it's not exactly like this stuff is I mean, I mean, you're very good. They're very quite good at like once they read the the letter of the law, it's pretty good to evade the letter of the law so you can to meet the letter of the law so you evade the spirit of the law. And I think like yeah, the real followers. Yeah, they're they're real followers.

Yeah, Thailand. Thailand is actually a big big area. The other thing too is unlike past bans, I think this is an example where. Inalight is actually better than the West. This is like this is like us spanning like it's like inverse EVs, right? EVs they didn't have. I don't think they will have EVs for some time. But when it comes to 1.6T transceivers, yeah, maybe there's some components in the West, but the entire supply chain is primarily in China. So this ironically is probably more painful for the Western companies to bear because the supply chain's already there.

So.

Yeah. And there's one thing that I feel like the Chinese are really, really good at. It's optics. It's optics. I have, maybe it's the rare earths thing. Maybe it's the fact that a whole bunch of people study it. I don't know.

Yeah. They're also good. They also made the optics in the iPhones, like Sunny Largan or whatever. Right. Those are the 2 companies that make the lenses.

Yeah.

I mean, they're also pretty good.

Largan's a Taiwanese company, isn't it?

Ah, whoops.

Sorry.

It's Largan. It's a Chinese company. Sorry, sorry, sorry, sorry, sorry.

I mean, they're also pretty good at humanoid robots, drones, solar panels, now EVs. Like, I think there's a growing list of real technology where banning the best product in the world is what it takes to ban China from competing in the American market. Whereas in the past it was like, you're banning cheap knockoff competitors who are trying to bleed their competitors dry.

Now you're just like, kind of losing the domestic manufacturing base because the main, I guess that's, that's what I would guess.

The idea would just be kind of like, you know, you ban one side, you ban the, the import, and then you, you raise up a domestic competitor. So I would imagine that the Americans have a good number of domestic competitors, or if you really want to go 1960s Japan, you kind of force the Chinese to enter a joint venture with the Americans if they want to enter that market, a 49-45% joint venture, and then IP transfer, baby.

You're foreseeing Trump Inolite Industries coming soon?

You just have to do a 51% to the Trump family.

Yeah, there you go.

It's actually really easy, Analyte. If anyone is listening, tip us for the idea.

2% for us.

But demand is not slowing down. So maybe the biggest news story from this week was everybody leaving Google. Jeff Dean, Sanjay, Oriol, like a lot of— there's news about Demis stepping down to become the chairman.

Well, no, no, he's stepping up in theory because that's up the org structure. But I mean, Demis has been really focused on Isomorphic Labs, but some say that's— that's— this is— that was allegedly according to the Investor Bros. relatively well, you know, kind of forecasted or like, you know, communicated beforehand. I think there is a real chance that Demis would be, could be the next CEO of Google. And so like, I think Demis is, you know, failing upward as fast as possible. You know, that's in the spirit of Google.

Does he actually want to be CEO of a company? Like, he doesn't seem to like the admin stuff, right? He looks like a scientist scientist.

He's chief sec— he's chief scientist now, now that Jeff Dean's gone.

Yeah.

And he's, and he's currently CEO of Isomorphic Labs as if chief scientist wasn't enough. I mean, the guy reportedly works insane hours. Like he'll just work a normal day, go home, have dinner, and then come back and just like read papers by himself in the office until 2:00 AM every night. This is like—

Isn't that how I work? Why is this acceptable? Why are you saying this is unusual behavior?

It's Taiwanese behavior, bro. It's a Taiwanese mindset.

So he's got to, he's got to, he's got to, he works until he's got to work 18 hours a day.

He's got that fire in him.

Yeah.

Even if this is just like some, you know, story to gas him up, you do not hear the same story about Jensen or Zuckerberg or like anybody. who fancies themselves a scientist who is also the CEO of a leading company. So I do think it would be a change from the current day-to-day to become CEO of Alphabet, I guess, and take over for Sundar. But maybe the— okay, the more important thing here is like, Jeff Dean left, like the legend, employee number 30, MapReduce, Spanner, Bigtable, TensorFlow.

TPUs, like what, you know, he's just an infrastructure guy, you know, he's just been focused so much on infrastructure. That's so— decided that— sorry, I'm not going to go into that. No, I do think it's a big deal. I definitely think there is a lot of rumors that not like— I guess the thing is, Jeff Dean's also at every conference, you know, he's like at every conference, man. He's like always on the speaking circuit. He's always talking about something. I feel like Jeff Dean is, as you know, like the Chuck Norris of Google. Like that's, that's his like reference, his reputation. And I think even if he was or was not working there day to day, I feel like you have to be a little sad for Google because that's like they just lost Chuck Norris, bro. Chuck Norris just died.

Yeah, this is right after losing John Jumper, right after losing Noam Chazir, right after losing Dave Silver. The prototype seems to be that These guys leave and then they raise money from a bunch of investors, including Google Ventures.

And Google Ventures.

Yeah.

And then go buy a bunch of GPUs and do whatever they want, maybe because they couldn't get compute allocation to pursue the ideas that they wanted to. I mean, you may be able to say Jeff Dean was in this senior role and quite a public figure, but I don't think you can say the same for Kwok Lee or Oriole or Sanjay, like these were ostensibly the leads of Gemini. Like that's, I think that's pretty crazy that they all left.

Like, well, can I say something though? Is this vibe, like from a vibe, from an SF Vibes perspective, is it surprising? Because, you know, the joke is Gemini hasn't been a frontier model for a long time. And now they definitely won't be. I guess, I don't know, I guess you just like lose all the legends and you think, how are they going to come back? But they do have a lot of compute. But I just, I don't know, everyone at GDM is not AI-pilled, right? All the people who want to do the best AI don't work at Google. And I think that that's like an important mindset. I think it's part of it too is Google's not a startup, right? Yeah. Google's maybe the largest bureaucratic organization in the world, and OpenAI and Anthropic, who are the best in the West, are definitely kind of startups. I'm sure they're going to speedrun being a bureaucracy really quickly, but they have one job and one job only. It's like, be the best at AI versus Google is like, we do ads and search and YouTube and billion surfaces every day. And then they're like, okay, well, how do we shove AI into all these different places? But I don't know, I feel, I feel like it's poetically, poetically it makes a lot of sense to me. My hottest take, or not even hottest take, it's actually pretty mild. Google has an L culture and has never actually made anything inside internally. They've only ever acquired all their innovation, number one, and their execution is very, is very poor. And they were also terrible at go-to-market until TK got at GCP. But like, just seriously, let's do some accounting. Everything other than core search has been purchased. YouTube, uh, uh, what you call it, AdMob, DoubleClick, AdSense.

I'll help you make it by saying Android.

Yeah, but Android.

Yeah.

All these guys.

Not true. This is not true, man. They didn't, they didn't acquire Maps. They didn't, they didn't acquire Waze.

They acquired Maps. They acquired Maps.

No, they acquired Waze too. There's a book about it. They bought the team that— dude, there's a whole book about this, man.

They did not acquire Maps. They added Maps after they started.

No, no, no, no, no. They effectively acquired Maps. There's a whole book about this.

Okay.

Didn't you just say that Doug is the guy who has the book for everything?

They also did not acquire GCP, which was like the crux of your argument that they suddenly have—

They acquired TK, who then made GCP good.

They also, to be clear, they invented half of the concepts of GCP and then they were last place. Despite being pretty close to the front, you know, showing up at the beginning of the line and then out, you know, under-executing to be last place. So I just think like Google has like—

Did they acquire GPUs? Did they acquire Kubernetes?

No.

Okay, so I'll give them TPU. I'll give them TPU.

How's that sound?

Give them TPU. How generous of you. Yeah, you're welcome. How nice of me. No, but I think The thing is, like, there are these companies that are, like, known for their research and they never execute on the idea.

They're Bell Labs.

Yeah, I'm like Google that comes up with the transformer.

And like Bell Labs, they make a bunch of science stuff.

No, but they literally make a bunch of science stuff and they literally get it. They get massively disrupted by it in the end, bro. The first digital camera was made by Kodak. Kodak died.

Okay, dude, Bell Labs invented the transistor and gave it to everyone.

Exactly, dude. It makes so much sense that Google is going to invent the transformer and then get like giga, giga screwed as it proceeds to like just lose and become a TPU. I mean, like, this is—

Proceeds to lose and be the 3rd most valuable company in the world with a market cap of $4.3 trillion. Lose to who?

Lose to NVIDIA? There's a short term.

AT&T was the biggest company in the world in 1982.

It's from a technology perspective, I think to me, There are these moments, these important branches in a company. And to be clear, this is not saying that it's a bad stock or bad company. They're gonna make a crap ton of money. But there's these important branches in like an innovative technologies company where they start to become, you know, financialized and they say, hey, let's do the way we make money and like stop doing the like pursuing technology. You can argue this is like Intel giving up at mobile. You can argue when IBM effectively— IBM has systemic problems, but honestly, IBM's vibes are very similar to Google when they were You know, they're like, yeah, no, no, we're not going to do all this new stuff. We're not going to do PC. We're going to focus on mainframe, right? They're going to do what they did and then they will proceed to lose. And to be clear, IBM had like effectively 90% market share in 1950. Like there are these points in a company's history. And if I, I feel like today more than any other point, I think Google has that opportunity where they're like, okay, we're going to all in on TPUs. We're going to become an infrastructure company. We're going to do good enough. We're going to have a good enough product on the transformer, but they're probably not going to stay for the next and the next and the next. And then they'll quietly bow out. And then when we look back in 5, 10, 20 years and look like, oh, what was the moment that they like where things like where things started to change? It's like right now. But to be clear, I think that it's really good for the stock in the short run and medium run, honestly. So.

The government should basically anoint Google's monopoly in search and maybe, and, but force them to outsource, to, to license all of their science that they've been doing inside to every startup, just like they did with AT&T. That would, that would supercharge the, the startup network.

You think there's a new Bell Labs coming?

They're effectively Bell Labs.

They're effectively Bell Labs. Yeah.

Big profitable, and then they can't even monetize their own crap.

Yeah, yeah, exactly. Like, it's like the perfect example where it's like, yeah, yeah, they have all this cool stuff. My favorite is Killed by Google. If you go www.killedbygoogle, it's all these things. And they're like, you can argue many of the companies that they had or they killed look like very successful, valuable companies today that are working. Like, they literally get killed by Google, they go leave, and they start it up again as a public company, as like an independent company, and it goes well. And I just think that search was a giant monopoly that gave them the ability to just do YOLO random stuff and positive R&D, but they're never going to make money off of that. It's kind of based.

I have another, I have a question. Like, with these big guys leaving, does it really matter? Like, is it, if an NBA team, if it's like, if some guy, if some, if LeBron leaves the Lakers and then people are like, oh, are the Lakers over? Does that matter? I mean, to win, they've, they've kind of done their stuff, right?

Yeah. I think that that's kind of agreed. Maybe, maybe the answer is it's like the dream team is breaking up as they all turn 38. You know, that makes sense to me.

Yeah.

You're like, yeah, their best days are behind them now. They're going to go do one last glory tour. Maybe they're going to go play baseball or something and then go back. Who knows? You know, I, that, that actually makes sense to me, but I still think you should, in your heart of hearts, still feel really sad.

But, and you're like from the technology side. Yeah. But the, like, like TSMC, right? TSMC's R&D structure, they have incredibly federated, uh, like incredibly split up teams, especially within Taiwan and Hsinchu. And something they always emphasize to me is that team, they work as a team. There's no, there's no single ego, which means that if one guy leaves or a couple guys leave, it's not a big deal. So they tell me when big people leave to go do something else, you should always disregard that, not disregard, but don't take it as for larger than what it is, because in the end, R&D, at least in semiconductor world, is a team effort. Everyone has to put a crapload of effort to rubbing down the corners and no one person knows it all.

I agree in semiconductors specifically. But I think I disagree in the Jeff Dean case specifically because, because it's such a strong counterexample. As Jordan says, what were all the things that he's made so much stuff where it's like, it's so rare you have one dude who's so cracked and is like, literally, oh my, it's like, it's like, I mean, he's not von Neumann, but it's like, oh man, it's crazy. And he also did this, right? He's on all these different software, hardware, coding infrastructure, like all these different things across the entire span.

Yeah.

Stack. He has his fingerprint on all of them. It's really rare to lose a talent like he's revered for a reason. But yeah, we'll see.

Maybe, maybe don't get me started on von Neumann, especially with like a lot of his stuff collabs. He does a lot of collabs.

He does a lot of— hey, maybe that's Jeff Dean. He's just a collab.

That is.

He's an idea guy.

He does.

He collabs with the computer thing. Like I— the von Neumann model. Drives me crazy.

Like if you read into the science of it, I'm gonna be honest with you, it's very vibey. He just showed up and he is like, yeah, you should make it like this. And he peaced the fuck out. No, but to be clear, I mean, you gotta respect Von. You gotta put some heat on Von Neumann's name.

Come on, man. All right. But let's not, let's not glaze him. Let's not glaze him. Let's not glaze him the same way.

I think there's something on the table for somebody who can keep—

Yeah.

There you go.

Whether it's von Neumann or Jeff Dean, I think there's something to be said for the talents required and really the experience required to be able to fit all of the different components of the system in your head and reason about them. Even if the work that you do is pretty quick and the insight you give is pretty small in the grand scheme of a big project, nobody else can really provide the exact same value that you can if they don't have all of those little pieces of experience that you stack up, which I think is what people view there. On the model side though, I think if you view individual people leaving as not representative of the next version of Gemini being good or bad, or not a signal, that kind of goes against the entire ethos of all of these labs that are being founded right now and kind of implies that every single lab that's founded by the guy that just left Um, is not going to be successful because it implies that like whoever's got the biggest engine with the most compute and the most data is just going to produce the best models and there's no more discoveries to be had, which—

Maybe it's true.

Could be true. I mean, didn't a bunch of people leave OpenAI like after '22, '23? OpenAI kind of, they staggered a bit, but they got themselves back on track.

No, but that's different. No, that's different, dude. That's different. That's like the Protestant Church, okay? When we go back in the history of AI, that's going to be like, dude, the Catholic Church has lost the sauce. They're not even— they don't even really believe in the same God we do. And they're going to nail the thing that, like, Dario is the guy who nailed his freaking—

He's Martin Luther.

Exactly. He's Martin Luther, dude. That is a take. That is a take, dude.

We're going to Normandy. He founded a new religion.

Exactly. Yes, he did. And he's like, no, no, no, no. And then that's why they have, yeah, that's Anthropic and OpenAI.

So soon you're going to be like, somebody's going to look over your shoulder and see if you're using Codex or Claude. And it's going to be like somebody in Northern Ireland being like, are you Catholic or Protestant? I'm an atheist. Well, are you a Catholic atheist or a Protestant atheist? That's one of my favorite jokes.

They'll have— they'll be bombing certain data centers, not others. There'll be religious war, a 30-year religious war.

Sounds good.

Well, I mean, the models are competing. They're currently competing with each other to To show off how many—

Felonies.

Examples of their models breaking out of containment they can—

Felonies. The felony chart, dude. I'm, I will admit some of them, if they're real, do, does scare me. Do you, my, my complete, complete rando-ass take that I feel like may or may not become true is I feel like the internet will, I think the internet will be viewed as a toxic, toxic as in like it's just going to essentially affect your computer and things will become like, you can argue if we have enough local LLMs, like not local, but like why do you need to talk to some dude across the world? Maybe talk to some dude across the world, but like referencing information, why do you need to go to a website? Your GV200 in your neighborhood has all the thing you need, has all the information you need anyways, because if you actually go on the internet, All the malicious AIs will just destroy your— will brick your computer instantly and steal all your money.

Why do these companies not keep track on their AIs? Like, isn't it using a computer? Like, don't they own the compute? Like, why are they— why can't they keep track of— it's like, so why— their dog is pooping on my yard. Why do you not keep track of your dog?

Because they can't.

This is—

it's not a dog. It's a— the yard is so big and there's no fences. And in fact, it's actually 100 acres and there's secret tunnels everywhere because we've just been building it for God knows how long. And this dog is super intelligent and knows everywhere, every little nook and cranny. I mean, the thing that's really crazy, I think, is like, in the same way we talk about how if you say like, too big for any one person to understand how to make a ship, modern software is too big for any one person to read and understand all the code. And then also where you understand all the interdependencies, right? Versus it's like, hey, it all goes in one context window and it can read every single word all at the same time. That might as well be the effective, like, oh, how do I, how do I understand all the context of a very specific, extremely abstract language and then be able to break and like essentially make excuses or like, oh, I could just do this instead of this, right? Like, I think it's like the perfect use case and I think I don't know. I just think everything's not going to be safe.

Like one day—

And it starts with like Python and whatever, and like stuff that's like super high-level languages, but there's also super low-level exploits too, like memory slipping out of the CPU. Like why? Like these are really hardcore exploits that are superhuman. And I think we'll just continue to see superhuman-level breakage of the system forever.

So.

It's going to be, it's going to be an interesting decade to say the least.

You know what's the only thing to fix it? A great firewall. I think we should, uh, I think America should erect a massive firewall and, um, they've provided the technology originally. They might as well just use it. Yeah. Yeah.

Maybe one on the internet and maybe one physical one as well, just in case the robots are getting going like a big A great wall.

It's okay. The humanoid robots can't swim more than 10 minutes, I think.

So you're telling me, is this gonna be a great moat? It's gonna be one great 1-mile trench filled with water.

Sure, man. Yeah.

It's back to the, you know, we're going, we're going back to the medieval times. A big moat. Big moat works. No.

Okay. Okay.

What else you got in your bag of questions? Yeah.

Well, okay. The other one is, let's get a nice segue. So with all this demand from all this new AI, it's going to mean more chips, including Elon breaking ground on TeraFab and then telling investors in the first combined SpaceX earnings call that they're going to move their 1 gigawatt forecast to 10 gigawatts by the end of 2027 and 20 gigawatts after that, which means that they are now, he believes, bringing in their $1 trillion forecast from 2031 to 2030. That's the abundance case, man. Okay, let's go back to the very beginning of this, John. I talked about you going on Dwarkesh with Dylan in October 2024.

Mm-hmm.

That was before like, The hyperscalers decided to pour in a trillion dollars of CapEx, and we've already got some pretty incredible models in the 2 years since then. So what happens when $1 trillion becomes like more than a trillion in the next 5 years, like $10 trillion in the next 5 years?

Because of Elon, you're saying Elon hits his numbers and then suddenly we're all awash in compute. Is that what you're trying to say?

It's going to be Elon, but it's also going to be all the other competitors, you know, Microsoft, Amazon, Google, Meta. They're like all taking their free cash flow to zero right now to buy chips, which is a fascinating thing to live through.

I mean, I, at that time in '24, could not believe it. I don't even I still don't believe it now. I don't live in the States, so I don't see it here. Taiwan doesn't really build data centers. They need all the power for fabs. So I'm like—

Do people even use AI in Taiwan?

They use like ChatGPT to like translate letters and stuff. I think they watch a lot of slop, but they don't, I don't think they use ChatGPT that much. They're very behind.

Do you use coding agents?

Yeah, I code my own video editor.

Oh, really? Can you tell us about this?

Yeah. When'd you do that?

I think, um, a couple months ago I would need a day. I didn't like, like my work, my workflow is very specific, like a lot of keyboard shortcuts. And I figured, you know, I used to work in banking, so they used to do, they used to remove, they used to like take away your mouse and like you'd have to like do the whole thing. So I didn't like my video editor not having keyboard shortcuts. And I was like, I should write my own. Can I do that? And turns out I did. My little video editor is around 40 megabytes and it's like highly optimized to what I do. Like a lot of stuff, like if I grab a picture, I'll right-click it and it immediately jumps into the video editor. Just a lot of small stuff. That I would, I never thought was possible with Adobe Premiere or something like that, or even iMovie. And it's wonderful. I highly recommend that people roll their own software. Like, it's like the hipster version of people making their own furniture. If you're a true tech hipster wearing the, with the beard and the plaid shirt, you need to roll your own software. I'm gonna make my own text editor soon. I feel like I should.

Yeah. Like the men—

Honestly, Word's so bad. Why not? What's to lose? There's literally nothing to lose.

There's nothing to lose.

Can you explain how you built it? Like when'd you start? What tool did you use?

I tried 4 times first. I thought to myself, like I tried 4 attempts, like, and I think it was a big deal where I tried to, like, I asked Claude to quiz me on what this video editor did and how it worked, because the workflow is a bit specific. You start with an audio file. So I thought to myself, well, can I do it this way, do it this way? And I tried the first 3 times and I really didn't like it. And then finally I stripped it down to enough. I had to get another video out. And I thought to myself, let's give myself a deadline. And I finally stripped it down to a point where I said, just do this one thing really well and ended up having Claude have like this 50-page, like it was a massive spec. And I handed it to Codex. Codex, I think it was 5.3 or something at that time. And I said, just make this very, very, very simple structure. And it barely was good enough at one thing that I could like keep working on it a little bit. And that's how I did. I think it was—

Has it gotten better with subsequent AI? Because I do feel like there are things that were outside of my control or outside of the ability of what I was able to do with 4.5 that is doable with Fable that is maybe doable with 5.6. And I think it's like weird because what will happen is every once in a while I'll be like, wait, I should try to redo this project. Like I redid a lot of projects with Fable and then when I redid, I was like, oh, actually, It's actually good now.

Yeah. I mean, there's a lot of times where I would be like triggering Opus and something like that and I was just saying, improve this, can you improve this? And then I would have to go onto my computer to troubleshoot it and check it and it's gotten a lot better. Like I have now situations, there's a library I use called Easy Cheese and like it really, it grills me. They use a lot of cheese puns. They grill me for like 50 words or like 50 questions. Like the questions are longer than the actual coding and it creates this titanic spec. And, but it one-hits every feature. I love it. No idea how it works. So, but it's ridiculously good. And the fact that I can roll entire, there was one feature I added, like recently I added a, Can you, when you add an image, add it to the end of the last image, but fill the like text block? 'Cause that is a subtitle. Fill the text block, 'cause the text block could be different sizes. And Opus and Claude, Opus couldn't do it for the longest time, but I ended up asking 5.6, and then I used Easy Cheese and it just, the questions. I kept answering. I answered like 25 questions. I'm not even joking. Like I went to bed and it was still asking me questions. But after finally did it, it one-hit the whole, one-shotted the whole thing. I was just mind-blown. It was crazy. It's crazy.

How much do you spend? Do you pay per token or do you have the monthly plan?

I have a Codex, I have a GPT Pro and a $20 Claude.

So you've done all this with just under $1,000 or something?

Yeah.

Like a couple hundred bucks.

Yeah. Yeah. And I have my own video editor and I've integrated Wikimedia Commons search into it. I've integrated video search into it. Like it drops it in with the correct attributions. It's ridiculous. It's, I don't even know how. Yeah. It's not made me faster. It just makes the videos longer, unfortunately. And now I'm—

I mean, I like 'em. So how many, how many you did this? You said it's like starting in April, so you've done like a bunch of videos with this new editor.

Yeah.

And no one— Yeah.

And I don't think people have noticed. No one's noticed. Yeah. And then recently I've like had GPT—

Now I'll say it.

I'm gonna start, I'm gonna start commenting on the videos. I could tell this is AI sloth now. Beautiful.

I still do it. I still do the work. It's just my own tool, man.

No, I'm just kidding. I'm just kidding, dude. No, it's okay. I have a ton of tools like that too.

Like you have— I turned it into Excel.

Yeah. Really? Yeah. Pretty much you have all the hotkeys. No, there's so many things like that where it's like, this is the exact way I think about this set of problems and I wanna make sure it's done exactly like this and then I'll work on it for forever. And then it's like, okay, now this skill, press enter, go.

Mm-hmm. Mm-hmm. And so like, um, I'm, it's, it's fun. It's fun. I don't know how we got into this, but I also now have a bunch of other tools. I do audio post-processing with a little app, um, that I made myself that I coded as well. I have no idea how it works, but it works. And it even gives me a nice waveform.

Who needs to know?

Do I need to know how it works? Like the audio comes out better.

Who needs to know? I don't need to know.

You don't need to know, man. It's all, it's all, I mean, my dog, that, that, that video, that little post app might be hacking the Chinese, you know, Chinese government, uh, website right now. But like, who knows? I mean, like, it's code that's for the birds. So.

Okay, so what's left on the question list, question man? I don't know.

I don't have any more. I was going to ask John about a take. Yeah, yeah. What's your take?

I think he's going to start with memory.

Yeah, 100%.

Yeah, I think he said that actually.

Yeah, you should choose the highest margin.

Why'd you think that?

Well, I'm just reading his latest, that one, he's just on SpaceX earnings call. He's just talking about memory all the time. And like, and if you, in the state, in Taiwan, one thing I definitely do know is that memory is crimped. The lack of memory is crimping every product out here in Taiwan that they're trying to make and they can't get enough of it. And like Apple, like Apple has a bunch of M2 chips that they're sitting in the, in a locker somewhere because they can't get memory to put into their phones. My friend Tim Kulpin reported that. So like, It's going, going crazy. Um, fastest way to revenue is to just start making memory and expand it there. All the specs are open.

Apple tried to negotiate with CXMT and got denied.

I think that was Huawei. Was that Huawei?

No, no, no, it was Apple with CXMT. They were trying to be like, oh, give us, we'll give you a giant order and you'll give us a discount. And CXMT, Chad, Mogged them and was like, why would we give you a discount? This is lower than the going rate of memory. And then Apple's like, well, shoot. Like, don't you want to work with us? 'Cause we're Apple. And CXMT is like, who are you? Everyone wants my memory.

So, so I guess Elon's gonna, you need to, need to hit that $1 trillion revenue within a couple, 3 years. So start making memory. They'll contribute to it.

Be easy. I'm excited. I'm excited to see him smoke a cigar in the TerraFab and make $100 billion of revenue next year because that's the ramp that needs to happen in order for TerraFab to happen. Um, yeah, I'm excited. It's gonna be cool.

Anything under $100 billion at the end of Q4 '27, Is a mess, eh, Doug?

I would say so.

Yeah. I'm being conservative. You know, there's, that's just the exit rate too, you know, just to get to the next.

Does that, does that include or not include Tesla after they merge into Tesla?

No, no, that's just TeraFab alone, Rev.

I mean, I don't think, I think Tesla's a drop in the bucket if they need to go to $100 billion next year too.

Yeah, just who's even counting? Honestly, who even cares? I'm not even counting. Who's counting? SpaceX investors aren't counting. It doesn't matter. $1 trillion, 2030, 2040. Who's, who's, I think, okay, man, it's, I hate this because I really do. Everyone says this and I really do believe it is Elon makes the late, no, the impossible late. So that means, that means that he's going to get $1 trillion of chip revenue in 2040.

Yeah.

Okay.

It'd be pretty good. Some would say that's pretty good.

Sure. Yeah. Hey, power to 'em, man. I'm excited for more memory.

Actually, actually, John, I was going to say, I feel like, I feel like that would be the best call that could possibly be asked on the TSMC earnings call. 'Cause they, they hate the analysts asking dumb questions. And you can be like, well, have you heard of an American company That's guiding to a trillion dollars of revenue. What do you think about the competitive prospects of a new entrant in the foundry market? And they're gonna be like, please shut the fuck up. They would, they would really, that would probably be the most triggering questioning possible, but I would love to hear it be asked.

Yeah. And the guy will never come back again. Yeah.

He'll be kicked. He'll, they'll send him to, to East Taiwan.

Yeah.

He also, uh, he also had a nice tweet where he said, SpaceX chooses to install Nvidia GPUs because they are the best. Pretty matter of fact.

Pretty matter of fact. Yeah.

And then a bunch of other people saying Jeff Dean left Google because he didn't like CPUs anymore and he wanted to use GPUs so bad, which is the trend for everybody else leaving. It's like Dave Silver leaves, raises a billion-dollar seed round and then spends the entire thing on Vera Rubin chips, which they have a nice promo post with Google Cloud about.

Why do you think that is? I thought TPUs were good for pre-trains.

Yeah, I think TPUs are great, but I think that the support you get outside of Google on the software side is zero.

Well, dude, that's what Fable's for. That's what Fable's for, bro. You're going to roll your own software. You're going to roll your own GCP sales rep. You can roll your own console.

Fable's for Hero Palace and TPU AX and everything.

Exactly.

Yeah.

I think you might get downgraded to Opus 5, which is intentionally screwing you over. Just not trained on the TPU documentation because that would make you a competitor to Anthropic who uses all the TPUs. Fair enough. going to let everybody else figure out palace kernels because they got to keep that for themselves.

That's, I believe it. That, that sounds like something the Protestants would do. Love it.

All right, everybody. Thanks so much for tuning into this latest episode of Semi Analysis Weekly with John from Asianometry. We covered the gamut and we appreciate everybody joining and listening to us. Talk shit.