I see a large IPO on the horizon. You're surrounded by journalists. Watch your position.
Overnight success. The place 3000. Double. Right. Information. Court declaring order inbound.
Let's just roll. We are surrounded by journalists. Hold your position. Come on. Get up. Trust the experts here. Five. We are founder Mel 5.
I see multiple journalists on the horizon. Standby. UAV online. Blaze. Double blaze. Triple blaze. Double kill. Fight.
Cook is win. Team deathmatch. We are experts. Triple blades. That's just wrong. Right.
Clearing order inbound. We are surrounded by journalists. Hold your position. Strike two. Activate golden retriever mode. Market clearing order inbound. I could. I see multiple journalists on the horizon.
You're watching TBPN. Today is Tuesday, March 17th. It's St. Patrick's Day. We are live from the TBPN Ultradome, the Temple of Technology, the Fortress of Finance. Let me tell you about ramp.com. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more. All in one place. Let me also tell you about Shopify because that's why I have this green suit. Shopify is the commerce platform that grows your business, lets you sell in seconds online, in store, on mobile, on social, on marketplaces. St. Patrick's Day, we are not drinking Guinness currently. We hope that you are.
Yeah, enjoy it. Very exciting. I haven't seen a lot of St. Patrick's Day posts on the timeline. What's going on? We got to step it up. I feel like we should have planned much more, but it seems like Tyler is in the St. Patrick's Day mood and is wearing a fantastic St. Patrick's Day hat.
Let me think of my hat. How did you wind up here? That's a nice hat. Where did that come from? Did we just have that? No, I had this. Oh, you just had it. You had it handy. You just daily drive that. Yeah. Okay. Yeah. It wasn't in your car or... Yeah, I wore it to the gym this morning. Okay. Cool. Yeah, apparently got to wear green. Jord is going with a little bit of a darker green today, but we will still be having fun. Let's pull up the linear lineup, show you who we have on the show today.
Carried no interest, the anonymous poster is coming on the show at 11:45. Then Shamscar from Palunteer Technologies is coming for a massive book launch. We're very excited about that. And then we have a lightning round going through cyber security, ceramic AI. We have Gecko Robotics on the show. I don't know how familiar you are with Gecko Robotics, but it's a very, very interesting company where they crawl up oil and gas infrastructure like literally like geckos and will inspect everything. There's a lot more to the business.
So like humans in gecko outfits crawling around. No, robots. Robots. They've been doing robots for a long time. Okay. Well, Linear of course is the system for modern software development. 70% of enterprise workspaces on Linear are using agents. So, in the news, the Wall Street Journal has an article about, this is a scoop from Berber Jinn. We didn't have to break down the paywall because we get a copy of the Wall Street Journal every day. Although, I don't know if this particular piece made it into the print edition today. It might be in the print edition tomorrow.
But Berberin has a scoop. Says OpenAI's Fiji told staff last week that the company could not afford to be distracted by side quests. Main quest only. Main quest only. What is the main quest? Probably just scaling compute, but we'll get into that. So there's a whole bunch of takes back and forth. Said the company execs are actively looking at areas to deprioritize. Of course throughout last year, OpenAI launched a ton of different initiatives across consumer hardware and I think a lot of people were starting to say like okay, do you want to be fighting a battle on all these different fronts or do you want to just really nail consumer, nail enterprise, and now that is what they are signaling internally.
So, from the Wall Street Journal article, OpenAI top executives are finalizing plans for a major strategy shift to refocus the company around coding and business users, recognizing that a do-everything-all-at-once strategy has put them on the defensive. Fiji Simo, OpenAI's CEO of applications, previewed the changes to employees at an all-hands meeting, telling them that top leaders, including Sam Altman and Mark Chen, were actively looking at which areas to deprioritize. They expect to notify staff about the changes in the coming weeks. We cannot miss this moment because we are distracted by side quests. Side quests is the key term there. We really have to nail productivity in general and particularly productivity on the business front.
I was listening to a podcast with the head of ChatGPT last night and there were a bunch of things. Yeah, Nick Turley just talking about like it's such a weird product surface area because people come to it for all these different reasons. So how you prioritize things within there, like images and ChatGPT, clearly very important. That was a separate app at one point, it got rolled in to good success. Nano Banana was also sort of like a okay everyone should download this particular app because you're going to get the best frontier image model for a while. Sora.
Yeah, we had talked about this a while back and it seems to be confirmed now that Sora's functionality will just end up in the main ChatGPT app. Yeah, so last year OpenAI announced an array of new products including the video generator Sora, a web browser called Atlas, a new hardware device, and e-commerce features for ChatGPT. Some of those are like wildly different timelines for these things.
Yeah, and remember during this time, that's when I started talking about viewing OpenAI's activities like they were a hyperscaler, right? When Google launches a new product. Yeah. You don't have to assume that hey it's going to work every single time and in fact it's the exact opposite. A lot of these experiments don't end up going anywhere, it's fine.
And OpenAI, like again you have this massively scaling core business. You start to say like hey let's experiment in a bunch of these different areas. Some of them are going to work, some aren't, it's okay. But then ultimately that creates a scenario where you're sort of like opening up maybe too many fronts, right, to the competition, right? You're competing with Meta in social with Sora, you're competing with Google in some ways. You're competing with even like the Microsofts of the world in other ways. And ultimately this just feels like hey, let's narrow the fronts. And I think a lot of people expected something like this for the last few months.
OTP in the chat is surfacing a very interesting old take from Ben Thompson about OpenAI should not be in the API business. And now based on the way things look, originally that was like oh wait well maybe Microsoft will be able to handle and scale the API demand, but API demand has been so huge and it's allowed Anthropic to develop such a solid business there that it's undeniable that OpenAI should also be in the API business, especially.
Yeah, the other side of that is like they clearly need to be in the harness business too, which is why Codex is accelerating. Yeah. And so the hyperscaler, like your take of like think about OpenAI as the new hyperscaler. I think that take aged extremely well, especially with recent interviews with Dylan Patel and Doresh Jensen just went on Strategy. There's a whole bunch of new data points that show how compute constrained we are, how tricky it will be, how chips will be the key bottleneck.
There was that interesting story that Tyler was recounting from Dylan Patel about how the TPU was developed by the TPU team. DeepMind didn't realize how important it was going to be, how compute constrained they were going to be. So the TPU went and sold it to Anthropic and then DeepMind went back to them and was like, 'Wait, wait, wait. We want those chips. Like we need the chips.' And so there's this hyperscaler phrase that is really, really important in that it designates not just a big consumer company or a big business. It's particularly about the ability to marshal compute at hyperscale, right?
And so I was interested to look at the history of side quests among Mag 7 companies, among trillion dollar tech companies because it's all over the place. And so it's very hard to paint with a broad brush, I think. Right now there's a huge narrative around like OpenAI was doing way too many side quests. They need to do zero side quests. And the reality is probably like, you know, we've seen this with Riley Walls joining the labs team, like there will be small projects. There will be acquisitions. There will be a continuum of bets. There's just a level of refocusing that's going on right now. Yeah. And reorganization. And this has happened at many, many...
Yeah. I think the bet with the OpenAI labs team is you have a really small group of people that are focused on creating or being quick to new product approaches, but that can be again like a two-pizza team. Yep. And that type of experimentation is going to make more sense if the rest of the company needs to refocus on enterprise and the core business.
Yeah. Tyler, is that data point you shared earlier like public about the size of that particular team? I didn't want to mention it if it was... team. Yeah. I think people know it's at least at one point it was very small. It was like six people. Also, I was going to say I think now that you have so many Neo labs doing these kind of weird moonshot research projects, I think it also adds to this thing where like maybe you don't actually need people internally doing those same things, right? You can just acquire them if they work.
Yep. Yeah. Did we... we got to get into that conspiracy theory later, but there's this idea of like all the Neos teaming up to marshal together. We got to dig into that. Anyway, history of side quests in tech. Can we paint with a broad brush here? You know, spoiler alert, I think the answer is no. But there are some fascinating stories. So, Google has a balloon internet play that's a serious company called Project Loon where they inflate balloons that go into the stratosphere. I believe they go high up and then they deliver internet. Sort of a competitor to Starlink. Some promising stuff there. They also have...
Fiber. They have the fiber play too. Fiber play. Oh yeah, they just sold that. Yeah. People are so upset about that because like dealing with your ISP is one of the roughest things ever. I've had such a bad time with various ISPs. Oftentimes I would, running a small business, running a startup, I would put it on my personal account and then I would have like six lines for like different people and then one of them wouldn't get paid and I'd get sent to collections for like that. The only time I've ever had... Yeah. Like some credit related issue.
Yeah. Was I returned the router and then even though I still had an active account, just another line with them, they were like, 'You didn't return it. We billed you.' Yeah. And somehow... Nick, are we still paying internet at the Jonathan Club or did we fix that? I sent you this. We can plan, but we haven't returned the equipment. I returned the equipment. Okay. We're definitely getting fleeced for sure. But that was the status quo and it was awesome when Google was just handling it with Google-level service.
Hard to sell ads against. Hard to sell ads against. So they divested, I guess. But that's not even close to the weirdest thing. The weirdest Google side project that I found is they make a contact lens that will tell you how drunk you are. I'm not kidding about this. Google, they have a project. It does more than that. It's supposed to do biometrics and it can do a lot of different things. But one of the things that it can measure, it can measure glucose in your tears. It can measure any sort of biomarker that's in your tears, which apparently is like a rich source of data. But one of them is blood alcohol level. So you can put in this contact lens from Google and it will tell you how drunk.
Jarvis, am I drunk? Am I good? Am I good to drive? You're absolutely not. He's like, you sir, you should take a Waymo tonight. Would you like me to connect you? See, this is all part of the plan. They're going to stop people from drunk driving. Give them Waymos. It works. So that was a weird one. They also briefly owned Boston Dynamics, but most importantly, they bought DeepMind, which was completely seen as like this wildcard side project. How does that fit in? You know, it's a bunch of researchers, and then it became like the most critical thing in Amazon.
Has taken tons of shots at journal-related delivery and home security stuff. They also own Twitch, but they never really linked the site to live shopping and closed that loop. Have you thought that? That's weird. Jasse doesn't do earnings calls. He doesn't do earnings calls on Twitch. And a lot of the looks-maxing live streamers have decamped to a different platform. Let me get some W's in the chat for this quarter, for this capex guide. Let me get some W's in the chat for the capex guide.
Of course, Apple tried to build a car then pulled out. Who knows where the Apple Vision Pro goes. Although, of course, I hope it continues. I'm hoping for another Apple Vision Pro. Apple Vision Air, just make it lighter. Same screen. That's the trick. Make it like a thousand bucks. I think it'll sell. And they have a bunch of health moonshots. They've been working on non-invasive glucose monitoring, which is there's some book about it called like the white whale of biotech or something. It's something that people have been working on for so long and they've never been able to crack it. They've worked on this. The best thing I think they have is that you can get a continuous glucose monitor. Thank you for the W's in the chat by the way. There's a lot of W's in the chat.
You can get a continuous glucose monitor that is invasive, meaning it is pricking you. It is measuring your blood and then calculating the amount of glucose in that. That's helpful for continuous glucose monitoring. There's a number of companies that do that in sort of a DTC realm, Levels. There's a number of biotech companies that offer that as like a healthcare product. And you can wirelessly connect that now to your Apple Watch, but they can't do it in the Apple Watch. There's always been the question about like, could you just shine a light through your skin and detect the glucose in the blood? Very, very difficult to do. Potentially always possible. Everyone thought, 'Oh, it doesn't break the laws of physics.' Lots of money spent with little to show for it, at least so far. Meta is probably the most egregious side quest.
They don't do side quests. They do full quests. They create side quests in VR. There's actually one of the popular games is all about doing side quests. And yeah, I mean they bought Oculus, they bought a bunch of VR studios, they rolled everything up, they renamed the company, spent tens of billions of dollars on what is starting to feel like less of a side quest, but still is so early. You know, the Meta Ray-Bans are like a success, but to the tune of millions of units, not billions of units or anything like that or hundreds of millions of units. I think the iPhone has an install base over a billion now and the ramp from here to there on Meta Ray-Bans is going to be long.
Tesla launched a premium tequila in a lightning bolt shaped bottle. So you know it comes for all different Mag 7 companies. That one's more of a stunt but truly like side quests come in all shapes and sizes. Some are just good for morale. Like it's just fun. We do side quests here all day long. TBPN simulator. That's a side quest. What was the other simulator? Jeremy Gon simulator, that was a side quest. You know, they're just fun. Some of them are good for marketing, good for attention, good for fun. Some are complete dumpster fires where you just pour money in and it just sucks all resources. You get nothing out. And then some reshape businesses entirely. DeepMind's a great example. And there's certainly others in the Mag 7 that have really, really changed the business.
Gabe said, 'Seems like the entire Boring Company is a side quest.' Yeah, it's technically a separate company, but Elon's... Yeah. King of side quests. Although I think... Yeah. I mean, it just feels like the company that gets the least amount of attention that's not on the critical path for many of the other projects. Totally. Yeah. Still a very cool idea. I mean, every time you're sitting in traffic, it's so tangible to know, oh, like if we just had more... creating a year for the minds, basically. Yes. Yes. Yes. But very, very difficult. I mean it's been what over a decade and there's really like a very limited rollout of that technology. So, but it's still cooking. I think the business is still going and they're working on it.
So some of OpenAI's teams come from very small teams with relatively tiny compute budgets. But they get a lot of attention sometimes because of the particular product category. So Sora is a great example where small team, not huge resource investment, probably a lot of inference and training cost, but relative to Codex 5.4 for Spar... I don't know. I don't actually know like how we're looking on the order of magnitude there.
I think a big part of it... super viral and what percentage of the team, the overall team's energy is going towards a specific product, right? Yeah. And then but so like the idea of experimenting quickly and then consolidating efforts even faster makes a ton of sense. Nano Banana was a big deal for Gemini. Bringing image, video, and audio generation together in a single chat flow is clearly the next step. And so some of this is not... it's probably not going to take the form of like stop doing the side quest. It's like let's fold that into the main quest because clearly the interaction pattern of ChatGPT is where people want to go. And so once we've done the experiment, it works. Put it all together.
Dylan Patel and Dorcash said the TAM for GPT 5.4 for was north of a hundred billion dollars. Which is crazy for a bunch of reasons. I mean it's just a huge market that was created in just a few years. The enterprise opportunity in general is crystal clear for anyone involved. At the same time, you still do need to experiment to make sure you are early to create the next breakout product experience. And so with that backdrop, OpenAI labs and being more efficient with the shots on goal makes a lot of sense. I still think that the overall narrative around AI, just AI broadly this year will be about the main quest which I see as like compute scaling.
And so raise the money, do the deals, grow the capacity. I was thinking back to Travis on the show saying if you're doing something and it's easy, it's not valuable. The key is if money matters, which I think we would say it does, especially in certain categories. He was probably thinking of ride sharing where there's a capital war and AI compute where there's also a capital war. You need to be the best in the world at it. Yeah. And I think overall there's real competition now. Yeah. It's intense, right? You have OpenAI, Anthropic at the frontier pushing very, very hard. But the other side of this for OpenAI is like all the... yes there was a ton of stuff that was announced last year. There was hardware, Sora, etc. But Sam was running around doing all these different mega deals for the compute side that now serve the main quest. And so the positioning is actually great.
Yeah. Let me tell you about Labelbox. RL environments, voice, robotics, evals, and expert human data. Labelbox is the data factory behind the world's leading AI teams. And let me also tell you about Gemini 3.1 Pro. With a more capable baseline, it's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view or bringing creative projects to life. And speaking of Google, Google Capital Bloke, the longtime Google bull, says, 'So Anthropic and OpenAI are going to just give the consumer market to Google.' And I don't know how... I mean, yes, there... like OpenAI reportedly planning to shift the strategy to refocus around business users and vibe coders. There's a lot of stuff going on there.
Listening to some of the data around the retention curve of the various products, like I don't think that they're giving up on consumer at all. That seems like sort of an odd read. It will be interesting to see the next iteration of Google's consumer surface area because they have AI search overviews, AI mode, the Gemini app, it's in Gmail, but there's clearly some UI fighting going on. I just... it's so funny when you're in Chrome and you have the ability to open one Ask Gemini panel in Chrome and then you can open a second Ask Gemini panel in Gmail the app. And so both at the web layer and the browser layer, you have two chat boxes that look exactly the same and do the same thing. This feels very much like a V1 of what they will do here.
So there's certainly an opportunity for Google to reintroduce the AI features, make them more tightly aligned with what people are actually doing. And you have to imagine that a lot of the progress they're making on agent coding can then come to the Gemini surface area in consumer overall. Yeah. And we can pull up this chart that Sam showed of Codex usage. This is like... again this is like what I think is informing like the battle with Anthropic as well as like this chart is going to inform the strategy shift which is like hey we can run... we can take revenue into the hundreds of billions of dollars with the current products that we have. Let's focus on them and make the best possible products.
Dean Ball is using Codex. Is that what he's saying? Just hit a personal record for single coding agent session of a little under 10 hours. GPT 5.4x high in the Codex app. Unsurprisingly, no flashy app, just really complicated economic research prompt. At this point, most of my prompts do not stress these agents all that much. To be clear, this is 10 hours of continuous work. I have meaningfully exceeded 10 hours if we include periods when the agent was waiting for jobs to complete. So, what does this mean? Is he talking about like firing off one prompt and coming back 10 hours later or just going back and forth? Because he says agent session.
I think he's talking about firing off one 10-hour... Yeah. What? I thought he meant he's in Codex like the terminal. Yeah. And he's in there for 10 hours. Locked in. That's amazing. Oh, wow. Yeah. It is very interesting to see. I mean, Doug Olaflin was talking about how he moved a lot of his research. He wasn't building software. No, he's saying it's not a flashy app that he's building, just a really complicated economic research prompt. To be clear, this is 10 hours of continuous work. I've meaningfully exceeded 10.
So I think what he's doing is he's asking like, okay, pull together all the census data about jobs. Now, go and pull together all the economic indicators. Pull together all the inflation... Somebody asked how much time was back and forth was going on versus the agent spending time on its own. He said zero back and forth. So this is 10 hours of the agent just like cooking. Wow. That's really crazy. What is this prompt? It must have been pretty long, I imagine. That's crazy. It fit within his current rate limits on the $200 OpenAI tier in the Codex app where rate limits are currently higher than usual.
Did it outperform expectations? The prompt was primarily about the political economy of central government transfers to Indian states/UN territories. I think honestly this matched my expectations of what 5.4 can do when it's properly prompted. What did it get stuck on? It encountered a bunch of problems along the way but doesn't appear to have gotten stuck for too long on any one thing. It was just a steppy process so it just kept working. That is absolutely crazy. In other news, OpenAI is forming a joint venture with TPG, Advent International, Bane Capital, and Brookfield Asset Management.
This is great news to form a joint enterprise products across the firm's portfolio companies and beyond. The proposed deal is coming to Fogo de Chão. Bane Capital owns Fogo de Chão, the Brazilian steakhouse. Maybe now they can take Apple Pay. Oh yeah, we got cooked on that. That would be a good... I don't know. Maybe they deliberately don't. I wonder if Apple Pay is expensive for them and this is actually a cost consideration.
FGCMO said, 'This news came out a little bit earlier than we planned. We're excited to be building a deployment arm and we'll share more details soon. Companies have a ton of urgency to deploy AI in their organizations and we're sprinting to meet that demand. More than 1 million businesses run on OpenAI products. Codex is now at 2 million weekly active users, up nearly 4x since the start of the year. API usage jumped 20% in the week after GPT 5.4 for watch launched in frontier which launched last month to help enterprises build, deploy and manage AI co-workers that can do real work has way more demand than we can handle. That's why we launched Frontier alliances. So we leverage our ecosystem of partners to scale and that is also why we're launching a dedicated deployment arm tasked with embedding forward deployed engineers deeply inside enterprises. This project has been in the works with our investors and alliance partners since last December and we are grateful for them and their partnership. We're still early, but the speed of adoption is a clear signal of where this is headed. We're excited to not just be building these technologies, but also building many ways for companies to deploy them and get impact.'
Interesting because there was a big talk for a long time about like the next-gen private equity firm will buy businesses and like deploy AI inside them. And this feels like well maybe that works at the mid-market private equity level but you don't necessarily need a new AI-native private equity firm because Bane Capital... is what I've always said like traditional private equity is not going to just be like oh we'll figure out AI in like 2030. It's not really on a roadmap right now. They're working as hard as they can to implement it. The question is like will traditional private equity be able to basically roll out and unlock the value of AI better than some of these like AI-native more like venture-oriented roll-ups.
Okay. So 5.4 mini announced today. So different sizes of models don't count as side quests. These are main quest, right? These are main quest aligned. But what is the pitch for a mini or a nano model? Yeah. I mean, so this is just like I think someone ran some evals and the mini models are like equivalent to GBT 5 when that first came out, but it's just way cheaper, right? So if you're like sensitive, if you're running a ton of queries, but you don't need to be like the max intelligence, then you just use this. So cheaper but also faster. Yes, probably. Yes. And potentially runs on older hardware. I think that's unclear.
This is bullish for our vintage Neocloud. Yeah, the oldest GPUs. Yeah, this was Gray's idea yesterday. Yeah, we're gonna run this on 1080 Ti from my gaming. People love classic cars. Why not classic GPUs? Yeah, I mean it's like how small the model is doesn't actually matter on... Imagine running your vintage Neocloud with a wood-fired hearth. Yeah. Powering it all. Yeah. But I mean you have to imagine that what's after Blackwell? What's the one that announced today? Uh, Ver Ruben. Ruben. Like you have to imagine that there will be models that only run on Ruben and need to be sort of rearchitected to run on Hopper I would imagine.
Uh, sure. I mean I think there's like small performance boost that you can get... or it might just be slower. But you can always run any model on like any hardware. It's just like it could be way slower because you have a fleet of A100s where you could just be on like a smaller rack of Rubens. Interesting. Well that's exciting. Cali came out with the $1 billion. Before we talk about this, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And let me also tell you about Public.com. Investing for those who take it seriously. Stocks, options, bonds, crypto, treasuries, and more with great customer service. So, what did Kali launch?
Kalshi launched the $1 billion perfect bracket challenge. $1 billion for a perfect March Madness bracket. So, you can win $1 billion for entering this competition. I haven't read the fine print, but it seemed it is positioned that way. Vinnie says, 'It would be hilarious if Citadel built out a team to attempt to financially impair their competitor SIG through this.' So, Sig Parametrics from Sesuana is basically like the financial backer for this promotion. So, if Citadel can figure this out, they could leave their competitor with a $1 billion bill. Yes, obviously trying to nail the perfect bracket is functionally impossible, but never pass.
So there are nine quintillion possible brackets. So, the odds of a perfect bracket are one in nine quintillion. And I believe that they are capping... I like my odds. I like my odds. And I believe that they are capping out entries at 10 million entries. So if you math that out and then you take some insurance out like you should be able to hedge out any of the risk if you are good and you don't suffer from any skill issues you can basically with strong basketball knowledge if you have ball knowledge apparently you can get the odds down to like one in 10 billion and at that you know from nine quintillion to 10 billion we're talking...
If you have nine quintillion you can sort them by how likely they are take the top 10 million. That's, you know, maybe there's some kind of power law thing here where we actually... Yes, you're getting your odds. Get close. We get close. But I think if there's 10 million... assume that there's 10 billion reasonable brackets and out of the nine quintillion, so you're discarding like all the craziest upsets to get to that 10 billion. You already ranked it and narrowed it down by like what six orders of magnitude or something. And then you're submitting the top 10 million of those 10 billion. Like your odds of winning are still one in a thousand.
You can make it a thousand accounts... and that's a no-no. I think that they are capping the entire campaign to the first 10 million entries. So you have to be on Khi and you have... my open claw already submitted nine quintillion. The SEC prepares proposal to eliminate quarterly reporting if you liked earnings. I was thinking like is this bad for us? We like earnings days but it's not like our whole shtick. So I think we'll be okay. We will continue soldiering on. Good. Alexander says finally less transparency around financial results. Yeah. I don't know.
I don't know if this is going to be enough to make it easier to be a public company. It's certainly a burden, but the kind of shareholder lawsuit risk and all that other stuff feels like a much bigger burden. What this is just going to do is create more volatility, right? If you're only getting updates, you know, twice a year, you can just see these like massive swings cuz it's like, hey, we haven't heard from this management team in a formal capacity for 6 months. The business could have, you know, changed wildly, you know, growth rates fluctuating, etc. And so again, if you love volatility, you're probably going to love this.
So, it's good for the VIX. That's bullish for VIX investors out there. There is a world where I mean the typical public versus private debate is that when you're private you can think in multiple years depending on who your investors are maybe think in decades even whereas when you're in the public markets you need to think in quarters and so advancing that from quarterly to six months and you start putting management teams in okay what can we deliver that will show up in six months as opposed to what can we deliver that shows up in three months that's twice as long it feels like you could wind up with better run companies doing more ambitious things. I don't know. There is a bull case.
Yeah. No, I mean the other side of it is you just get more complacency. There's less of that. You're not like on the daily march, you know, you finish earnings and you're like, 'Okay, 90 days. We got to do this again.' Like there's no days off kind of thing. So, are you an advocate for monthly reporting? Daily. Daily earnings calls. Potentially hourly. Just put... I mean this is the crypto folks. They're like put on the blockchain. I want to be able to see in real time the revenues of this asset stream, how things are moving around. I want full transparency at all times as an investor. I don't know.
I think that there is a potential for a good outcome here, but we will have to keep an eye on it and see what actually happens. This is just a proposal. It's just the proposal isn't even... we are preparing for a proposal. So the proposal hasn't happened. We're preparing for the proposal and then the proposal... it's an advanced idea. No, it's a preliminary talk. It's an advanced idea. Yeah. For a talk for early talks around a proposal. Exactly. Around a potential proposal. You know, the FDA actually works that way. ANPRM. It's like advanced notice of proposed rulemaking. So they tell the industry, okay, we are thinking about making a rule. Good luck. Good luck. And then everyone's like, we're suing. Don't change the rules. Because I have set up my entire business for this particular set of rules and if you change it, it's like completely over for me because every business is set up like this to narrowly thread whatever regulation is there.
Really quickly, let me tell you about vibe.co where DC brands, B2B startups, and AI companies advertise on streaming TV. Pick channels, target audiences, and measure sales just like on Meta. And let me tell you about Cisco, critical infrastructure for the AI era. Unlock seamless real-time experiences and new value with Cisco. Nvidia announcing Nvidia DLSS5 and AI powered breakthrough in visual fidelity for games coming this fall. Video games are going to get more realistic. Prepare over the next couple decades. This is going to be... it's entirely possible this happens. Continue. Let's pull up the video.
It's funny because when I see the video... I want to see a big... when I see the video my thought was weren't video games already this realistic? But I haven't played video games besides Tetris on theatic. They... Oh, yeah. Long time. So, let's play GeForce RTX. DLSS 5. DLSS, of course, stands for deep learning, super shading, something like that. DLSS. It uses deep learning. It's used AI for a long time. The whole pitch for DL super sampling has always been go from a 720p render to 4K upres, just interpolate the pixels. This is different. You can see that it's not just becoming higher resolution. It is becoming Gen AI. Like there is a world where you render that at higher res. The lighting is changing. The shading is changing. The makeup, the structure of that person's face is changing but it's still driven by the underlying... Why are you posting the bubbles? Cuz I'm ranting.
I want to see DLSS5 do this. Oh yeah. But this is what we were talking about yesterday. We were like, what will it take to get TBPN Simulator photoreal? I think we should try and pipe it through DLSS5. Although, who knows what this is going to run on because Nvidia is not shipping new graphics cards to gamers or something like that. They're delaying the next gaming graphics card. But I think that looks better. I don't know. People are upset. There's a community note that has yet to be accepted, but the community notes are not happy. Like one of the community notes is just AI slop. And it's like yeah, like that's in the name. It's deep learning. Like what do you think the DL stands for? And also why are we on DLSS5?
Oh, because this is like a decade old technology. I don't know. People are very upset. One of the community notes: AI technology has disastrous effects on the environment. Is the majority of gamers did not ask for this. Yeah. Is that... are those the most photorealistic video games that we have today? And then they're showing how they're even getting more realistic. No. Yeah. Because they don't look... they look like games that are just like kind of like animated characters that aren't trying to be photoreal.
So there's a range of games there that some of them have pretty outdated engines, pretty outdated rendering pipelines. Some of them are on the latest and greatest and look very high fidelity. There's also a whole bunch of trade-offs in game development around the scale of what you're building and how many poker chips you want to put into great graphics. If you're making some moody first-person shooter that's going to be dark and there's going to be lots of blood splattering and aliens and stuff like you could potentially make it ultra photoreal. But if you're trying to build like something like Minecraft, it's like this massive world that's interacted with a ton of different people. Like by narrowing down the scope, like the beauty of Minecraft was that the whole thing's voxel-based, which is like these little blocks. And so the computer can store like an entire world very efficiently because all it needs to say is just there's a cube there and it's green. There's a cube there and it's black. There's a cube there and it's gray. What do you think?
I was going to say Minecraft is a good example of what this can improve, right? Go to the next post. Oh yes. Pull that one up. Yes, this is what Minecraft could look like. This is what Minecraft could look like. There are drawbacks obviously because you're driving the style transfer from the particular images from the shape. You're not reconstructing the entire image. You're just applying this like rerendering on top of the underlying structure. And so you can get really weird outcomes like this. Of course you can rerender the full scene, but if you're doing that on a per frame basis... so right now basically what we're seeing with DLSS is the best AI can do in 10 milliseconds, I think, or something like that because it needs to do this basically if you're running a game at 60 fps, it needs to be able to do it 60 times per second. So you can't be waiting around for images in ChatGPT or Nano Banana to cook for a minute on each frame. It has to be real time.
But also on consumer hardware, right? Yeah. On consumer. This is not going to be like hosted in the cloud. The whole point is that it can run in your gaming PC. And by that token, this feels like it's not just four years behind. This feels like it's like one or two years behind in terms of the actual quality of the output. I'm actually pretty impressed. If this was going back to like Dolly 2 quality, I would have expected that because Dolly 2 was running on a fleet of A100s, H100s, something like... Sure. But I mean, the underlying tech is very different, right? Because you have like a very clear ground truth, right? Where you know what's supposed to be there. There's this kind of upsampling.
Well, so the data pipeline here, the actual ground truthing is a little bit harder for something like this. They probably have to use image gen to generate the high-res versions to actually generate all that because previously DLSS was really easy to train because you would just go and run the game at 4K on great hardware and then go and run the exact same... you would render it twice like from the same exact game footage. So you'd have someone play the game and on one monitor it would be 720p. On one it would be 4K and then they would record both and they would say these are exactly the same thing. Every frame is the same. Learn how to take the 720p image and create the 4K image.
Now they have to say okay well we have a 4K image. Let's run that through a Gen AI pipeline to make it like photoreal. Do that a ton probably on a fleet of GPUs that are not consumer hardware. Then you have your training data. Then you go run the DLSS pipeline and get this like style transfer algorithm that can run 60fps anyway. But I mean there have been like open source image models that have done like really good upsampling for a while now. But at this level or wait, are you talking about upresing or are you talking about like I would consider this like re-rendering using like something that feels like diffusion like this feels like because if you look at any of these examples, it's very clear that they're not just increasing the pixel count like the hair in that woman's image like the hair has a different shape after you turn DLSS on. It's not just higher resolution.
Yeah, that's true. So, like there has to be something like that they're training on. But and I think that there probably will be very small hallucinations still if you were to zoom in. Whereas that wasn't really true with the 720p to 4K pipeline. Like you would get like whatever the cutout of that image was that you were upresing would remain the same. Whereas this is actually like rerendering it basically. Well, none of this will matter when we have a billion GPUs in space. Let's head over to space. Let me tell you about Phantom Cash. Fund your wallet without exchanges or middlemen and spend with the Phantom card. And let me also tell you about Octa. Octa helps you assign every AI agent a trust. So you get the power of AI without the risk. Secure every agent. Secure any agent with Octa.
Take us to space. Jensen. Let's listen to Jensen talking about data centers in... courtesy of Nick. I'll spend very little time on this this time. However, we're going to space. We've already been out in space. Thor is radiation approved and we're in satellites. You do imaging from the satellites in the future. We'll also build data centers in space. Obviously very complicated to do. So we're working with our partners on a new computer called Vera Rubin Space One and it's going to go out to space and start data centers out in space. Now, of course, in space, there's no conduction, there's no convection, there's just radiation. And so, we have to figure out how to cool these systems out in space. But we've got...
Lots of great engineers working on it. It was very funny listening to Dylan Patel talk about space data centers and saying that, like, yeah, even if you solve everything, it's still hard to make chips. He's just like, it still comes back down to the chip bottleneck above all else.
Yeah. I mean, this is why you see Elon doing the Terapab thing, right? Yeah. I really want to know what his plan is for that because has he put in an order for a tool or is he going to do Terara ASML? Because the supply chain is what, 10,000 companies?
I mean, historically they've gone super, super early in the supply chain, right? They've done like, you know, mining stuff and... Yeah. But tool making is a special, special industry.
Well, speaking of Dylan Patel, SemiAnalysis was featured at NVIDIA GTC. The inference king has been crowned. Nvidia won a massive belt and it looks like Jensen's holding it up. He is in fact standing in front of an LED wall or projector. But a beautiful thing to see the SemiAnalysis logo. All WWE. The entire world is WWE.
Nvidia Extreme co-design revolutionized token cost. The GB NVL 72 is the inference king with 50x higher performance per watt on inference X by SemiAnalysis. 35x lower cost. Very, very exciting and congrats to Jensen for becoming the inference king and winning the inference max award, or inference X as it's now known.
Jensen is also confirming what we see in our GPU availability data. There is an epic scramble for compute. B200 basically unavailable. Availability for GH Grace Hopper 200, H200 and A100 also collapsing. Low availability means high demand. And people were kind of going back joking about like he said like one trillion of demand or something over some period of time and people were like, oh, so he's guiding down. It's like, I guess that's where we're at.
Yeah. It's hard when you're doing cumulative revenue. Let's move over to a much smaller deal for Netflix and Warner Brothers and the final deal which went to Paramount, of course, and David Zaslav is in the Wall Street Journal. First I will tell you about Vanta, automate compliance and security. Vanta is the leading AI trust management platform.
So David Zaslav's deal pay could top $800 million after last-minute tax benefit. So, Warner Brothers chief executive David Zaslav could collect more than $800 million in severance and other payments after rival Paramount acquires the company. The sum includes cash and payments for options and restricted stock holdings as well as a newly adopted tax reimbursement for Zaslav in a securities disclosure filed late Monday.
The total doesn't include the more than 20 million he's likely to get for shares he owns outright. So, he's been investing as well. About 504 million is due to Zaslav if the deal closes, the company said, while 47 million would be triggered if he is fired or leaves under circumstances within a year of the close. 116 million in equity has already vested.
Some 335 million would only kick in if other payments trigger a 20% federal excise tax on golden parachute severance payments he receives. The ultimate value of the payout is based on tax code rules that are expected to cause it to significantly decline with the passage of time. The company said in a filing without the tax payment, the total would be about 667 million.
Companies typically try to keep severance at or below three times salary and target annual bonus to avoid or minimize the tax, and investors often criticize companies for reimbursing or grossing up executives for the tax. The company said Zaslav wouldn't have to pay the excise tax if the deal closes in 2027, which would save the company costs of the tax gross-up. So there's been a lot of back and forth.
A lot of people are asking, did he earn that much value? And the reason that I think it is warranted from a business standpoint is when we were sitting here last year talking with someone who won't be named in the media space, somebody who has done deals not at this scale but in the billions of dollars. He was sitting here telling us at this very table that he didn't think there would be any bidding process at all. And it was just going to land with the Ellisons for a pretty predictable price.
And Zaslav did his job as CEO to get the best possible price for shareholders, and it's remarkable. And that is why, you know, he was able to have outsized impact. He's getting outsized compensation. He increased the enterprise value through his deal making by tens of billions of dollars and he's getting a significant but I think warranted payment.
And if you don't like this, then you probably don't like the game of business very much either. Or you think Warner Brothers Discovery should have just used OpenClaw on a Mac Mini or something like that to negotiate the sale. Put the whole company on eBay and let everyone bid. Just do an auction.
It's actually eBay's fees are pretty high. I think eBay takes like multiple percentage. So it might take a higher fee. I think they would take a higher fee actually. It's like a hundred billion dollar deal. Don't they take like 3%?
So by that I mean maybe we're saving money using a human in this case. But of course, some of what he did, you know, between the dinners and the photos and all of these different negotiations and somehow continuing to get Paramount to make offers but then say that it wasn't their best and final. I don't know how you do that. He's clearly getting multiple offers back to back that were not the best and final.
But quickly let me tell you everyone about Lambda. Lambda is the super intelligence cloud building AI supercomputers for training and inference that scale from one GPU to hundreds of thousands. And without further ado, we have Carried No Interest, the anonymous ex-user with the most fantastic profile picture. How are you, No Interest?
I'm going to repeat Carried No Interest as many times as possible so that I don't say your real name. Carrie, I'm going to say your real name. Good to have you here, Carried. Thanks for taking the time to join the show. Why don't you tell us what's been on your mind lately?
