Is OpenAI Really Worth $300B? ElevenLabs’ Series C & YC’s Funding Focus | E2079

26:56And I'm a little squeamish by nature about bones and tendons and stuff. I like my wings boneless. You know, that's where I stand. So there you go. Okay. On the chicken coop front though, I love the niche idea. Let's talk about what YC wants to fund because sadly chicken coops are not on there. I wanted to bucket these for us so that way I could break down Jason, you know, like this topic, that topic. There was one topic and it was AI essentially. So where YC is focused is that kind of three categories, AI infrastructure, AI applied, and then hardware, which I'll talk about at the end. So very briefly for everybody, a request for startup from YC used to come out once a year.

27:29They've increased that cadence, basically saying things are changing quickly. As we've all seen with DeepSeek, things can change overnight. So no beef with me there. Makes sense. What's on the list? Okay. They would like to someone to build an AI app store with a focus on security so you can keep your data private. Just imagine instead of the app store that has candy crush and so forth on it, an enormous number of AI specific apps that are consumer friendly. That seems obvious to me. Questions about platforms. We'll see. I like it, Jason. Yeah. An AI app store is an obvious idea. And there were, there was a version of this, which was prompt engineering stores.

28:02There were people two years ago, we talked about it here on this podcast, selling prompts. Of course, that was a short lived business because now with reasoning in things like deep research and O1 and some of the new products by Gemini and the products by ChatGPT and OpenAI, the reasoning is built in. So if you don't know how to do a prompt you can just ask it to make a prompt or it makes a prompt as we saw when we use deep research on the program countless times so i do wonder if this is going to actually exist outside of gemini and other places because they would if it's going to exist it's going to exist in those places and there is one inside of chat gpt they called it plugins at some point yeah they made a store for that it was terrible no one used it and it kind of faded away i that That was a rare aborted product launch from OpenAI.

28:48People forgot about that, actually. It's kind of disappeared. Sometimes it takes two or three swings at the bat. So, you know, that was using like integrations into Kayak, let's say, or, you know, into Redfin or other, you know, Zillow. And it was kind of kludgy. That, as I predicted, would be built into Gemini. When you do Gemini now and you ask it about a flight, it gives you Google Flights data inside of it. So it's pretty obvious Google is going to win a lot of those. and it'll probably, the AI app store will probably be the app store. Yeah. So, you know, if you open Google Play, it'll probably be like the AI agents will just live there.

29:23Just like games or ringtones and some of those other, remember there was a ringtone section in, you know, the app stores for a while because they were trying to make that into a thing where you could - Oh, that was a moment in music business when ringtones were a serious percentage of total music revenue. People forget, I mean, if you're young, you weren't there for it, But people used to have clips from their favorite songs play, and that was considered both normal and not rude. Finding great developers can feel impossible, right? It's like one of the hardest things you can do. You've got to juggle all the technical challenges, looming deadlines.

29:56And of course, you've got to be on budget, right? You're a startup or maybe you're a big company. You have a big company budget. You know, there's a CFO watching it or you're a startup and you have limited resources. In both cases, you've got to hire top tier developers. That shouldn't slow you down. It should be speeding you up. And that's where Scalable Path comes in. And they help you build your dream team and scale your business with confidence. Scalable Path has a custom vetting process that ensures every developer is the perfect fit for your stack. They create personalized technical tests tailored to your needs.

30:24So devs, when they join your team, they're going to hit the ground running. And they've got a global network that connects you to over 40 ,000 top tier developers with expertise in all the important languages and tools you use, Python, JavaScript, AI, machine learning. You get the idea. And the best part, their talent is affordable. They focus on regions like Latin America, where you can get a world-class developer without the Silicon Valley price tag. And since 2010, Scalable Path has completed over 300 client projects, perfecting a process that eliminates the endless interviews that you're going to have to sort through.

30:57And, you know, sometimes people snow you and you get somebody who's great on paper, and then they don't deliver. You can trust the talent you get from Scalable Path to deliver exceptional results on time and on budget. Scale Smarter with Developers Who Deliver by using Scalable Path. Go to scalablepath.com slash twist to get started today. They're going to give you 20 % off your first month. You might not need it. It's already a great price, but they want to make sure that they know you came from this week in startups. So please use our URL and use that code to get that 20 % off. I just want to make sure that they know we sent you.

31:27Scalablepath.com slash twist. And it was 99 cents. So it was kind of fun. Like I put the Imperial March on my first iPhone and I use a Star Trek to this day when i get messages it's a star trek um what is the communicator called on star trek not not a phaser um it does like that you know kind of anyway that was you know my phone ring used to be the imperial march um which got quite annoying forever around me so that's where i think that will live and then these compliance and ai audit tools those are already happening so yeah compliance and audit whether it's kpmg or whatever service it is those audits cost a lot of money And I think they want to replace people.

32:07I think that these are going to be better suited like tax GPT is doing or some of the legal services. There's the one legal, um, AI company that Sequoia just did a big round in Harvey. The guide on the side is going to be what's happening. You're not going to take the legal doc, the legal certification or the CPA out of the loop with these compliance tools. They're going to do the first pass at it. And then there's going to be human in the loop. But probably fewer CPAs required to review work than to do it all de novo. So it's probably going to be a compression back to our static team size becoming compressed team size.

32:42Another place where the YC folks are looking is building AIs to replace personal staff. They point out that wealthy people have had access to private doctors and private coaches and all sorts of things. They think that's going to become AI-ified and more generally available. I think maybe. We'll see. You mean an example of what they mean by that? You're talking about like your personal trainer or your personal assistant? Yeah, personal trainer, personal doctor. Yeah. So basically, if you're wealthy enough to hire a human to do a task for you on a regular basis, either to help you do something better or to do something for you, they expect AI, maybe AI plus robots, Jason, back to your optimist point, will fill that niche.

33:20And I agree with this in time, but I don't know if we have the technology yet to replace, frankly, the human connection that I enjoy in those roles. I like the human element in someone knowing me who isn't a computer. These things already exist as personal coaching in apps. So NutriSense, a company we're invested in, has a nutritionist that you can include in your subscription when you're looking at your glucose monitor to kind of coach you through things. We have another startup, Blaze, that was using AI to help people with coaching for like, I don't know, race car driving. And they had human in a loop.

33:56fit bod and other investment owners use machine learning to look at your last work and make you a new workout so this kind of concept has always existed and you can just take whatever rich people pay for i did a tweet about this just a week or two ago that went viral a rich person uh paying for a high-end product will eventually make its way down and it will be everybody's personal driver and that's what ubers you know you everybody gets a chauffeur kind of a situation yes door dash everybody gets a personal chef airbnb everybody gets a second vacation home you can just take things that rich people experience today which might be a masseuse which might be a personal trainer and then you can abstract it use technology use ai to make it more available and that's really what fitbit is doing uh fitbod rather so if you go to fitbod uh i think it's fitbod.me is their url it just starts for fitbod you know a personal trainer knowing what you did and then knowing what equipment you have and then making you a new workout that is, you know, a hundred bucks or 200 bucks if you live in a city, if you go to a person every week, and now you can do that for$59 a year.

35:01So it already exists. I think it's a, it's a fine premise or a fine framework to think of startup ideas. For sure. The way that I would maybe just down then is use AI and the deflationary price movement of technology products to take things that are currently too expensive and make them more mass, mass available. So it's essentially the same story we've seen with technology just in an AI context. I mean, again, that's cool, but we've also done it in a mobile context before in a cloud context. So this is the next kind of platform iteration. A couple other quick things before we let everyone go.

35:33Dev tools for AI agents, I thought was a pretty cool idea. Essentially, how do you help people build more agents quickly and better? I think everyone wants an AI agent that does stuff for them. I don't think many people have them, so I like that. And then the other one that stuck out to me was software to support open source AI usage. So who's going to build the companies that will support other people using open source software? We had Weeviate on the show, Christopher Hunter Company, Vector Search, open source with services on top, great model, venture applicable. And so I think we're going to see a lot of good work there.

36:05And then Hardware Jason, just really quickly, they talked about startups to accelerate data center build out. I liked that. I liked that they didn't back away from the idea that we need still mass investment in hardware because for like 48 hours there i feel like everyone was like oh deep seek solved the hardware problem that wasn't true no it wasn't um as i tell everybody do not trust any data or information or the government of china it is a dictatorship they you know uh lie they steal ip it's just you can't trust any report out of china uh at face value whether it's like Alibaba's earnings, I wouldn't trust it.

36:45Or it's a claim about TikTok. Do not trust anything at face value. And this idea that they spent$6 million probably leaves out hundreds of thousands of illegally imported H-100s via Singapore that they're not putting into that cost. And then you compare it to the cost that maybe OpenAI and Sam Altman quotes. Well, well, Sam Altman is going to pump up the fully baked cost of doing a model. Why? He wants to scare people from creating competitive models, right? So he's trying, and he said, don't even try to out hustle us, you know, in building these models, you will not be able to compete because we have so much hardware.

37:24So he's trying to freeze the market. Therefore, he will say it's 100 million to create ChatGPT3. It's a billion to create ChatGPT4. It's going to be 10 billion. He's pumping the number up. So you can't believe him on that side, because he's going to over underestimated and then coming out of china they're going to say well it was six million well six million maybe to run it in terms of the electricity and if you divided the cost of the h100s that are illegally in china through singapore allegedly by you know a 20-year lifespan per hour you know you get the idea you can you do fun with numbers and so can i just say though that i feel like everyone i feel like they're getting hit for something that they didn't do so what i have here is the DeepSeq V3 white paper.

38:07And they say that the training costs of DeepSeq V3, assuming the rental price of this GPU at that cost per hour per GPU, pre-context, post, total. Like they tried to provide a reasonable accounting here. And then I think what everyone did was they conflated this, which is for V3, which is not R1. R1 was a improvement on that with other additional features to make it better for reasoning. But I think everyone went, oh r1 cost 6 million they're liars hey yeah let's let's it's just not apples to apples comparison and so you know you have one group that's what matters yeah and you have one group saying okay well we spent a billion dollars on h1s and we use those h1s to make this llm therefore it's a billion dollars and you know you can just have all kinds of fun with numbers depending on how you depreciate the asset or the utilization of the asset you know and one dollar an hour to rent something could be$1 ,000 an hour, depending on how you frame it and the devil's in the details.

39:04Yeah. Can I say a headline though, Jason, about the China point? Because I don't mean to argue against your point that China's GDP numbers are a little bit sketchy, but I love this from the Atlantic Council. China's economic performance, new numbers, same overstatement. So hard to agree with everything about not trusting biased sources. And when it comes to investors and people like Sam Altman, they are talking their book and I always read it with that context but a good reminder to everyone out there too and i think that's probably a good segue into this uh peak valuation i had said you know i thought we were hitting peak nvidia you know sometime in the next couple of years you would have this sort of peak nvidia moment in terms of competition is coming people are going to make their own chips and then software will become more efficient as we saw with r1 using maybe more commoditized hardware less generation of hardware better software better techniques etc because constraint makes for great art if you give somebody one sheet of paper and say, write a great song, they're going to be very thoughtful.

40:00If you give them a laptop with unlimited storage, you know, the song may never arrive. And so sometimes you need, you know, a little constraint, right? Like, or you're not going to get your paycheck unless you submit this song. I always tell the story about Bob Dylan and Blood on the track. So putting all that aside, I also thought we were hitting peak evaluation for OpenAI. Yeah. Because there seems to be so much headwinds against them. However, Masayoshi-san, who loves to counter any kind of reasonable valuation is reportedly looking at putting$25 billion into OpenAI at a$300 billion free money, yeah?

40:37I think it might be post. The$300 billion number is very fuzzy. This is all from the Wall Street Journal's reporting. But I mean, just keep in mind that they were worth$157 billion before, and that's what Jason said was maybe peak. So this is effectively a doubling. And that's such a masa move to just be like, well, double it, add a zero. and you love him for it. I don't know if I would want to put my own money into OpenAI at$300 billion, but I do want to make a comparison here because I think we lose a little bit of talk of comparative scale. So the other day, Deepsea comes out, NVIDIA loses, what was it, 17 % of its valuation in one day?

41:13$600 billion largest. $600 billion, exactly. Off the top of the market cap. Off the top of the market cap. So do you think that OpenAI is only worth half of what NVIDIA shed in one day? that's what i mean it's all yeah i i think you have to make two bets here one is what's the actual traction in business of open ai is it the consumer business of people buying chat gpt for 20 bucks a month um and you know how much of the revenue is that or is it the api and people hitting the api and paying them for requests and they're in competition with aws and microsoft azure which has all the weights and put up r1 as a way to maybe uh even more than flick um you know Sam Waltman's ear, but maybe put him in his place and say, like, listen, we're agnostic.

41:58If people want to use R1, which is based on allegedly stolen training from OpenAI, we're going to actually support that. Think of, like, let's just pause there for a second of how cutthroat Satya Nadella is. Sam Waltman has claimed, and OpenAI has claimed, that R1 was stolen from OpenAI, and that they have these reports of people hitting their API and doing this sort of learning and reinforcement learning, et cetera, and training it on OpenAI, and it's stolen, yada, yada. They stole our IP. They stole our IP. And what does Microsoft do the next day? They support it on Azure. And they're like, here, go ahead.

42:30And it's cheaper than anything OpenAI offers. Well, you know what Microsoft has become over the years? An oddly strong supporter of open source. And people forget this transition. People are often still stuck in that 90s mindset of like, you know, Microsoft 1.0. But Microsoft's been messing with open source for a minute now. And they don't care if you want to run R1 or an open AI model on Azure? Pretty much, as long as you're doing it on their compute. If you're doing it on their compute and you have their credit card. So, and then the third piece is, what would the valuation be? So you have the valuation based on it being a Google competitor, right?

43:05You ask questions, you get an answer. Will people pay 20 bucks a month for that sustained over time? Or is there an advertising model that merges? Then there's the second piece, the API going up against Azure and Google Cloud and AWS and Oracle and everybody else. And then the third piece is something else, artificial general intelligence, something else emerges that is worth$300 billion. So if you were to put these in buckets, we do have some reports on their previous revenue for each of those product lines. So we know that they were supposed to have reached$3.7 billion in 2024, which is, to be clear, an enormous revenue amount and a real accomplishment and not to be diminished.

43:43The breakdown that I've consistently seen is that OpenAI makes most of its money from subscriptions, the 20 bucks a month that I pay, the corporate account that Launch has, et cetera. Whereas Anthropic, which is smaller and is probably closer to the billion dollar run rate mark, has a lot more business on the API side. Those API calls from corporate customers. Now, I don't think that either of these starting points means that they won't converge at some point in time. But I think that this is the clutch here for OpenAI. Because if they get Project Stargate off the ground, if they raise more money from SoftBank, they will have had every single possible advantage that they could want, right?

44:19So they are supposed to go from$3.7 billion in revenue last year to$11 or$12 this year, according to some reporting about internal projections and so forth. If they get to$12 billion this year, everyone will shut up. Yeah. If they don't, we'll see. Yeah. And these prices keep coming down. So my understanding is with R1 coming out, OpenAI is also going to lower the token cost and everything like that. So what we're going to experience over the next couple of years is, I mean, the cost of running jobs on AI is just depreciating. like is it 10x a year 5x a year we're going to need another moore's law for how much it costs to do a job you know uh and there's some unit of a query and how difficult it is to do that query because now if you can start running these things locally on a couple of macbook you know macbooks or m1s you know where are people going to actually send their jobs and then it becomes about fidelity and i actually think you know ai queries we these might become so commoditized that we look at them as good of a business as storage and bandwidth, which is another way to say a commodity business.

45:30I think this is becoming modified so fast that anybody who wants to create a robot, self-driving, a SaaS app, a mobile app, pick your verticalized agent, they're going to be able to create it so cheap that it all becomes commoditized. What happens to self-driving as a concept if there's so much ai there's so much training data out there and it's so cheap that i don't know any car manufacturer can just say fire up this open source ai fire up you know uh all of this training data and we're back in the game and everybody has self-driving that's you know 99.59s at the same time then it becomes just a manufacturing issue who can manufacture the most and run the most at scale network.

46:17Yes, but that's why I think it's smart that Waymo and Tesla and WeRide are trying to do quick commercial applications of self-driving because if it does end up just being, I can get a Ford, a GM or whatever, and they're all the same, then no one really has leverage. But if I don't buy one because I've already been captured by Waymo or Tesla or someone else who just wants to sell me a robotaxi by the minute, by the hour, by the day, by the ride, that's a completely different model. So it's a race in that way to the point of commoditization of a model. Now, about the AI curve, I'm going to show a post from Dario Modi, the guy who runs Anthropic.

46:55And he did a blog post about DeepSeek V3 and the training costs. And he said that if the historical trend of the cost curve decrease is 4x per year, then we would expect a model that's 3 to 4x cheaper than 3.5 Sonnet, one of their models, about now. He argues throughout this that essentially the lower cost for training DeepSeek v3 is in keeping with the cost curves we're seeing today. The crazy thing is, if this persists for another 18 months at the current rate, it will change the world forever by bringing such high-level models to the market at such low cost. That's why I kind of get excited whenever OpenAI drops something and they're like, it's really expensive.

47:34Please don't use it too much. Like, O3 is supposed to come out today. I guarantee you O3 is going to be expensive as hell today. but in six months it's going to be three three pennies a call from three different companies i mean yes it's uh it's a i think open source is going to win the day and there's a great irony here i talked about it on all in which will come out uh today as well over this weekend it's kind of rich that open ai stole everybody's content got caught with their hand in the cookie jar you know and then they're complaining and then they flip to a closed source flip to a for-profit so they do all these things that i would consider unethical uh or highly unethical stealing other people's content without permission uh flipping from a non-profit to a for-profit then they're crying foul when somebody does their original mission which was to open source and now it just happens to be the chinese doing it and they they have stolen uh some content apparently or they have you know they don't have the same content regulations there they don't care about ip over there they just steal it as quick as possible and that's their philosophy of it and now open ai is crying foul i think open source wins sorry that's just my belief i think open is going to win this.

48:39And if I did raise 30 billion and I was Sam, you know what I would buy? I'd buy Quora. I'd buy Reddit. I'd buy Twitter. I'd buy the New York Times. I'd buy the Washington Post. I'd buy the Disney Corporation. This is where actually the value might reside. And especially if the New York Times wins its lawsuit, if the New York Times wins its lawsuit and then Disney joins the New York Times, Washington Post, Reddit, Quora, just pick anybody who is at scale data. And they all start creating a voting block and they just go to every single LLM and say, if our data is in there, we're going to just hammer you with lawsuits until you say uncle.

49:15And when that happens, that would be the equivalent of if the content industry had been as coordinated, the news industry, the magazine business, as the music industry was with Napster, with Google. Can you imagine if Google couldn't index New York Times pages or all content, CNET back in the day, all that stuff that was in there? If they had just said, yeah, no, you can't index us. And as a group, we will not let you index our content unless you pay us a license. Google would have had no choice. And Google would be giving 70 % of their advertising to the content creators or some number of them, just like the music industry does with Spotify.

49:47Spotify has to give what, 60, 70, 80 % of every dollar? 65, I think, is their current kind of gross margin rate. Yeah. Yeah. So they have to just ship that money. Even if it was 10%. of Google's search revenue that went to content providers. Dude, I would be like rolling up in a Bentley today. Like, I mean, that would be so much money for me. Do you know how many millions of words I have published over the course of my career that has just been taken? I don't feel bad for OpenAI. I also agree that open source is probably going to win. One thing I don't know and I need to look into this more is what is the possible copyright exposure of meta with Llama and their open source models?

50:23Because I don't think that just having something to be quasi open source obviates the copyright risk of using copyright information in your training. So I'm curious if we're using open AI as the example here, but even if a lot of the open source AI groups are going to have the same problem. They will have similar issues and they will get injunctions against them and it'll be a little bit more kludgy, like who do you sue and who do you get an injunction against? But there's nothing that says like some open source community couldn't get a legal letter saying hey you use star wars and created an open source star wars you know model creator and you're making star wars ip we're going to get injunction and make github or gitlab or any number of places that post it or fork it and we're just gonna send legal letters to them and and shut it down and put it in the underground which is what happened with victorrent right or other services those things just became underground services they still exist they just don't exist um in a mass way because the legal options are so much better and fluid and affordable and just more ethical i think people like to be ethical they won't be if they're getting ripped off but i pay for three music services i pay for spotify yeah i pay for apple music as part of my bundle and then because i got into high fidelity i started doing cobus uh q-u-b-o-z i think it is which is like this french high fidelity thing and so i am paying unnecessarily i guess for spotify uh because i have apple music in the family, but my family prefers the interface, but Apple Music comes for free.

51:51And then actually with my YouTube subscription, my premium, I get YouTube music. So you have four and also you have some access via Prime. Oh yeah, maybe my Prime has free music too. I think there's some music attached to Prime as well. So that means, Jason, you are literally the most important music fan in the world. Thank you for supporting the arts. I do this in a slightly different fashion. I buy vinyls and tapes and CDs from bands that I love when they have records come out. and I have a tape of Fit for an Autopsy's latest LP The Nothing That Is I don't have a tape player but it was small and it was like 10 bucks and I'm like I love you guys here's you know so I love your support I think it's cool to do that I love you know people doing that and you'll see there's two vinyls somebody gave me a gift behind me those are two Dire Straits of the original Dire Straits albums that somebody just handed me at a conference and I'm like someone knows what you like yeah and just put it on the shelf they'll never get plays they'll just be on the shelf forever but I like it Once my front living room stops being a in-house daycare for my girls, I can put my record player back in there and that'd be great.

52:52All right. Closing off the OpenAI story, a couple of notes about this. OpenAI did participate in the last$6.6 billion funding round for OpenAI. So SoftBank already has about$500 million into the company that way. It also did later last year, a$1.5 billion tender offer. So they already own about$2 billion worth of buy-in in equity in OpenAI. so if they do another 15 to 25 billion and then what happens with the conversion you know like they keep pinning these valuations to open ai open ai keeps hitting these record revenue numbers so how does the non-profit collect on that what happens to this like 10x or a thousand x yeah then we turn to open source i mean the the layers of rules that have to occur and then layering this amount of money on top of it what happens if they legally can't flip what happens to all that money does it get returned to investors i mean i would love to understand the legal um risk that thrive capital is taking uh christian over there josh kushner or masayoshi son like they're putting lp money into this potential uh house of cards and if they can't flip into a for-profit do they just get their money back but what if the money's spent like this is a large amount of risk to take for tens of billions of dollars or any billions of dollars absolutely so i don't know man i just this seems like pretty frisky to me oh it is uh it's we're talking about softbank jason the mobile pizza company and the dog walking company they don't care and that's what i like about it because everyone else is trying to like do what we do you know what's the revenue multiple masa goes i like sand do you want 25 billion dollars okay but um for fun i'm gonna read this lead from the journal because it's just it's too funny for words reading for everyone from the Wall Street Journal.

54:39In his newly built palace near Tokyo, lined by stone statues of Roman emperors and surrounded by an 18-hole golf course, Masayoshi-san was stewing. After declaring for years the imminent arrival of the artificial intelligence revolution, the chief executive of SoftBank had missed out on it. Quote, I haven't been able to do anything he thought, according to a speech he gave to SoftBank investors last year. Can I just get old like this and die? I love this guy. He's awesome. I mean, greatest investor in the history of investors. I mean, and the worst well i mean you just take the swings you know like if it takes a certain amount of um hootspa cojones to be able to make giant swings like this and not care um about past failures or zeros and then just keep momentum investing this is like the ultimate momentum investing and it's um it's great that it exists in the world to a certain extent um whether i would want to participate in it as an LP?

55:37I don't know. I mean, I've chosen not to. I do a different type of investing, but hey, more power to him. I think we have some good questions from the audience that we could potentially hit. I saw somebody was... Yeah, let's drop one. So here we are from CCCC. If Stargate goes through, isn't a big chunk of that budget for hardware? And would this be a bullish sign for companies like Grok and NVIDIA? Grok is a company that does AI inference hardware. Sure. Used to offer it as a kind of a purchasing now offer kind of via cloud service. And NVIDIA, of course, makes all the chips. Just to be clear, CC, I think Jason and I are still pretty bullish on the hardware component of AI and worrying about open AI's valuation, I don't think undercuts the overall story of the need for a lot of spend on hardware's chips and networking equipment.

56:21We're just concerned that maybe this one company with this one model and this one historical corporate structure might be having a valuation attached to it that will be difficult to earn into. But Jason, your thoughts? You know, sometimes we overbuild capacity for certain things and sometimes we underbuild it we've underbuilt electrical capacity in the united states uh and at a point in time we overbuilt fiber and we overbuilt storage and then you know creative entrepreneurs look at the plummeting cost of both of those and created youtube and then youtube was like you know what uh you could upload unlimited and there were used to be caps you could i think maybe youtube it was like you could upload a five or ten minute video yeah and then they went to an hour and then you know google photos or gmail started at whatever gig and then it went to terabytes and then it went to unlimited essentially so you can create products and services that are based upon over building and so one person might overbuild capacity and then another person might find a use for it and we saw this happen in the big data revolution we it was too expensive to store uh storage logs so i remember when we were doing email newsletters in the 90s brian alvey had set up these qmail servers for us and we had all the delivery stats And I love looking at the delivery stats, but he said, listen, you know, we have to put another hard drive in.

57:34We got to put another, you know, disk array in. And I was like, yeah, but what do we do? And there was like, oh, there's a piece of software to roll all this up and then give us the top level reports. So then we started saving the top level reports of opens and closes. And I was like, man, it'd be really good to have this granularly to see which people open the email every day, which ones open it every other day. But we had to literally have discussions about how many of the logs we wanted to store. Yeah. And the same thing with page views on Engadget. And we basically at a certain point said, don't just delete the logs and keep the top level of reports.

58:03And we paid for some software to just keep PDFs on another hard drive because it was getting too expensive. Somewhere along that journey, cloud computing came out and we didn't have to rack servers anymore. Then it became a discussion of, well, what's our storage bill? The same thing happened with this very podcast. We were spending 10 ,000 a month storing these like giant files. And then we realized, you know what? We'll just put the final edit on YouTube. It's in 4K. It's in HD. We always have it. It's always stored there. We have a backup on, you know, this other storage system. And then Libsynth and other people started making it free or close to free or unlimited.

58:34So that's what we'll see with this overbuilt. If it's overbuilt, people are going to do seemingly ridiculous things with that extra capacity. Some of them will be amazing. Some of them will be superfluous and silly. And, you know, people can afford it. So I'm not too worried about it, to be totally honest. And the only thing I am worried about is the government paying for this. We're in this whole thing of trying to control our spend. And I saw some headline go by that maybe Sam Altman and these guys were trying to get money from Trump to do the build out of these servers. Absolutely not. If there's people like Masayoshi-san out there, or sovereign wealth funds or venture capitalists, they should be paying for the build out of this, not the American public.

59:14Hard no on that. Enough with the public paying for everything. I don't want to see Sam Altman or anybody else going with their hat out asking Trump to give him$100 billion for us to win the AI race when it should be won by open source. Wow. Lot, lot to unpack in there. Um, I'll, I'll say that I, as far as I, I I've, I've read, there is no public money in Project Stargate confirmed. I know that he probably wanted some, but I don't think he's going to get it. And I think your point about SoftBank is correct. But the last thing I'll say on this is, you know, Satya over at Microsoft has said that we're building out this capacity because there's so much demand for AI inference, which is using kind of models out in the wild.

59:56And Anthropica said that they are compute constrained. And part of the OpenAI Microsoft beef has been OpenAI has wanted more compute than Microsoft was willing to give them. So to me, there is the chance that we are going to overbuild, Jason's point. And if we do, there'll be some positive benefits of that as well. I'm just not worried about overbuilding right now. This is the job of capital allocators. There are capital allocators out there who have choices of where to put their money. They can put it into municipal bonds they could put it into treasuries or they could put it into core weave or project stargate uh yeah but back to stolen ip that's like a protected series this is a really good series and a really good movie like how do you just go out with the audacity and steal the term stargate i wonder if they paid jason they stole the internet and then they chewed it and then they're spitting it back to us for a dollar i mean who cares yeah who cares if they steal somebody's IP for their movie.

1:00:46Do you think Vossette cares if Sam? No. I mean, literally, that is a beloved science fiction series. And they just, I mean, why don't they just call it Star Wars? Just steal that. Star Trek Wars. Star Trek. Yeah. I want to hit a couple of things, though, before we move towards it. First of all, we have talked about antitrust on this show ad nauseum because it's a big deal for venture, for startups. The news is that over at the DOJ, after Trump took over, there has been a suit from the DOJ to essentially block the HPE, Fuel Packard Enterprise, and Juniper deal. And here is the logic from the DOJ.

1:01:25We've excerpted this. I'll read it for folks who are on the audio part. The proposed transaction between HPE and Juniper, if allowed to proceed, would further consolidate an already highly concentrated market and leave U.S. enterprises facing two companies commanding over 70 % of the market, the post-merger HPE and market leader Cisco Systems International. This substantial lessening of competition in a critically important technology market poses the precise threat that the Clayton Act was enacted to prevent. Jason, this strikes me as an incredibly reasonable argument against a combination inside of an industry.

1:01:57However, it's also a technology deal worth billions of dollars. Market feedback was surprise at the Trump administration's move here. This is not startups. This is big companies. This is public companies. does this make you worry about the trump administration's posture towards technology combinations i'm just curious this has to be a multi-year investigation so while trump's been office for 10 days i guess 11 this is obviously probably been in the works for at least four years for whenever this deal was struck that they started looking at it so i don't know when they started this investigation so while trump has inherited it i think looking at router technology uh you know which is what these folks are in um dell has these obviously huawei which is banned in the united states and countless other people build these these are commodities you know routers sending data and packets is a commodity business and it is a race to the bottom yeah you can go on uh i talk about this unify that i love uh i'll give them a free plug here this is like an upstart Unify is this incredible IT company that I'm in love with.

1:03:06And if you go look at what they build, and you just go to the pricing section for these routers, and you just click on the dream machine here. So if you go to switching, and you just prick any of their switches, professional, professional Macs, enterprise, these things are unbelievably cheap. And they're using the same ones in homes as they're using in enterprises. And then you go over to camera security, click the next tab over camera security, then click the next tab over door access. And you look at all of these integrations that they have VoIP, the next one over, they have taken the entire stack from Cisco and commoditized it.

1:03:44And if you go anywhere and you see Unify routers, these routers for Wi-Fi cost a hundred bucks, a thousand bucks. The switches cost 500 bucks. The cost of these things is essentially free compared to when I started in the industry. Yes. The equivalent of these were, you know, when you were putting them in, you would basically say we're going to put you know four thousand dollars for each pc on people's desk to set it up and then we're going to spend two or three thousand dollars to set up the network for each person then we'll spend two or three thousand dollars for you know each person's share of the server capacity and you would build server so you'd be looking at an enterprise you had a thousand people in your company let's say you were where i work sony music is a thousand people sure you would say ten thousand dollars per person every four years we'll spend ten million dollars on it to put in this infrastructure now the same pc is 500 bucks the same router per person is 100 or 200 per person and then you know it's all done with wi-fi you can do cables whatever it's all been commoditized so this is the kind of case that makes no sense it's a commodity business it doesn't matter if it consolidates down to two or three players because anybody can undercut it and even when you look at the unify stuff when you go to ui.com they're not sponsors or anything here they're looking for things to do they have ev charging they added right they're doing digital signage they're looking for things they added premium audio to their offering so when you go to integrations they've now added essentially a sonos competitor inside their uh product now why would you have that well some number of people who use these networks want to have music in their offices or their cafes or whatever so the long tail of these things then you go on to amazon and you look at um netgear like the real cheap consumer stuff then under netgear there are commodity ones that look exactly like netgear probably made in the same factory that have some you know bozo logo on it you've never heard of so there is nothing to worry about here this is just overreach consumers have won this already and enterprises have won it it's silly that we're even having this discussion in the review mirror because this was settled 10, 20 years ago.

1:05:53Yeah. So all of that was kind of what I was thinking. But here's the part that surprised me. And here's why I brought it up, because this is not just a Biden era DOJ hangover. Over in CRN, there was a quote that says from someone who is party to this overall transaction, my first thought was this must be a holdover from the outgoing DOJ leadership. So I was shocked this assistant attorney general was appointed just recently by the new administration. so this is actually a potential data point that the Trump administration even though you and I may disagree with them on this case might have more antitrust teeth than people thought and if that's the case there will be some sad faces out there but I don't think it's 14 billion dollars right is that what this acquisition is going to be for Juniper it's nothing like what are we doing here oh god I'm not going to use the R word but this is R word silly this is silly Jason thank you we can say silly it's silly shockingly enough HPE and Juniper are both very mad about this.

1:06:48And just really quick, before we go, I want to show the chart. We mentioned it earlier about NVIDIA's revenue via Singapore, how other companies might be getting these chips. This is a chart, Jason, that shows. Okay. Well, I think you'll just understand this from the get-go. But if you track reported revenue from NVIDIA, who sells chips, to Singapore, a country that you are allowed to send chips to, their revenue from Singapore went from$1 billion in the second quarter of 2023, calendar, to about$7.7 billion in the most recently reported quarter. Where do you think those ships are going, Jason, given that Singapore is such an enormous place?

1:07:26Yeah, I mean, the population of Singapore divided into this number would be really interesting to sort of see, like, how much, you know, just have fun with numbers. I think Singapore is a tiny country in terms of population. It's 6 million people, so it's$1 ,200 per citizen of Singapore into NVIDIA spent. That, if that was the US, that would be$400 billion. Anyway, it's pretty obvious what's happening here. NVIDIA knows what's happening. Our government knows what's happening. Singapore knows what's happening. These things are getting on a slow boat to China. That's what's happening, folks. And so these export controls, if we're going to be serious about them, Singapore and NVIDIA and the government are going to have to get on the same page here.

1:08:11The perception I have about all this is this race is reminding me of nuclear proliferation, which is there are nine countries that have nuclear weapons. And the only way to stop this is to control the hardware, the metal, the materials. Yeah. it's gonna be pretty hard to do obviously but we did it with nuclear proliferation russia france germany uk united states south korea china israel they deny it india and i guess pakistan uh has some rogue stuff north korea has some rogue stuff so iran has some programs so you're looking at like nine or so people confirmed and then you know a long tail of three or four people who are uh mucking about ukraine obviously gave them up when they um you know separated from the ussr security a little footnote there don't ever give up your nukes yeah just newsflash hold on to them hold on to them never give them up you it'll not work out well for you but i think that's what we're going to be looking at here potentially is you know just slowing stuff down but that's the most you're going to be able to do and it's going to create abundance for everybody on the planet so i think the hopeful version of this is like nuclear power we could have clean energy for everybody the downside is nuclear weapons with ai we could have robots getting your chickens from your chicken coop and making you eggs and having free food and having free products and services that go down to zero what is it going to cost to make a car when there are no humans in the factory what is it going to cost to take coal out of the earth if we even needed to or do any speaking of berries when robots and ai do this perfectly we're going to see the price of food energy and products and services continue to plummet you can buy a flat panel tv now for 200 bucks like a giant one 300 bucks like open up amazon look i was shocked at how cheap our most recent tv was i'm gonna just to give people an idea of this you you pull up a 50 or 60 inch cheapest led on amazon prime right now and you sort it by price for a 50 inch or 60 inch you might be able to get a vizio for you know at walmart a vizio might cost you 200 bucks 300 bucks for a tv that 15 years ago was five thousand dollars but jason i literally this is just a blank search log out of amazon 60 inch television 60 inch absurd 60 inch the second one cost 300 the third 55 inch costs 234 240 i that's insane what does it cost to get it to your house like the shipping cost is free delivery tomorrow between eight and i mean what are we talking about here folks i mean just keep doing the math that 55 inch tv like i said 10 years ago would have been at a zero right 20 years ago at a zero you'd be talking about at ces it was 3 000 so we've seen this happen with tvs you're seeing it happen with the m4 you look at the price curve of the m4 mac mini this computer for 499 i believe is the entry price 599 is more powerful than anything else in the lineup right now without a monitor and that's mac that's like the high end you look at it in a a chromebook you can buy a chromebook for 99 bucks you can buy a chromebox for 99 bucks you can buy a stick pc for 50 bucks the cheapest um mac mini uh that runs the m4 non pro is 600 you can spend up to for m4 pro you can spend up to 1400 but that's a sure 12 core cpu 16-core GPU, 24 gigs of memory, and a 512 gigabyte SSD.

1:11:46That's for somebody doing rendering of like CGI, you know? Yeah, but think about like five years ago, you would have gotten half of that for twice as much. And that's the cost curve we talked about from Dario and Anthropic and everyone else. So this is why I remain optimistic. The world has many things going on in it that I have feelings about, but I do have pretty reasonable confidence in the fact that technology is going to make our lives better. I want to put one more thing on our docket to talk about for next week and for our notie gang to maybe start talking about, which is anti-capitalism slash hacking capitalism themes.

1:12:19I want to just sort of talk about this next week at some point. There's a trend called no buy 2025. And then there's Apple selling less iPhones, a trend we've talked about here for a couple of years and, you know, people skipping, upgrading it. But this no buy 2025 combination of a reaction to inflation with fire, financial independence, retire early, overworking, all this kind of stuff and this abundance that's coming is a very interesting trend I want to explore next week. No buy 2025. Very interesting. I have lots of thoughts about this, too. I have an essay brewing on the different perspectives on what AI is going to do to prices, employment, consumer surplus, and so forth.

1:13:00So, great. We'll get to that. I'm x.com slash Alex he's x.com slash Jason we are Twist Monday, Wednesday, Friday thisweekinsertups.com slash Docket twist100.com you're handsome and lovely everybody goodbye there we go alright see you next time bye bye