Dan Wang's Annual Letter, Meta Acquires Manus, Nvidia's $20B Groq Deal | Diet TBPN

0:29It is so good to be back. that you can win a race has a defect like a definite ending there's a finish line whoever crosses at first wins and that's where he prefers to say that the us and china need to win win the ai future and this echoes what gurley has said too true obviously at the time he was coming out in defense of manis and benchmarks investment in manis totally he's like i don't know what the ai race is i mean he looks great now that manis is an you know an american company it's a meta property so So, you know, and I mean, the Manus team, of course, it was from China, but quickly moved to Singapore and now Manlo Park, presumably.

1:07No, I think they're going to stay there. But still, I mean, you have to imagine that a lot of talent migrates and it's like a fully American controlled asset, essentially, now that it's controlled by America and Meta and Mark Zuckerberg. Let's take it over to Tyler. Let's check in with Tyler. Oh, yeah. What are we checking in with Tyler? The chat misses Tyler. Oh, they just missed Tyler. How are you doing? And if you missed it, so we signed a massive contract extension with Tyler. He's not technically a college dropout yet, but he may as well be. It's a gap year. It's a gap year. It's a gap decade.

1:40It's a gap century.

1:44Tyler's parents, if you guys are watching, I'm just joking around. It's the age of gap years. We're in the age of gap years. We're in the age of research, and we're also in the age of gap years. And so we thank Tyler for sticking around, hanging out with us. We're going to have a lot of fun. this year. So the Dan Wong piece, we should read through some of it. He is reflecting on the AI future, what it takes to win the AI future. We've seen this amazing, just, you know, this remarkable march of research. The models are getting better. All the benchmarks are getting, you know, completely dominated.

2:16Then we're seeing the massive hyperscaler buildouts. We were reflecting on like, what does re-industrialization mean? What does American dynamism mean? What does this idea of like America needs to build big things? We need to be able to build big things, new bridges, we need to build high-speed rail. And we've failed at a lot of that obviously, but we've succeeded at building multi-billion dollar data centers. Podcasts? No, I mean yes, podcast, yes. I mean truly, truly media has been a cultural export of America for a very long time. Yeah. So Dan's saying we're very good at making data centers haven't been so good on the energy set my big question was like we've been talking a big game about energy being a bottleneck for ai for a long time it wasn't truly the bottleneck with bottleneck was chips and then data centers and then just moving the energy around but it feels like the last year we were just kind of shuffling energy assets around the board and that's why we saw energy prices go up where if just supply and demand if we've been building a ton of energy infrastructure well supply would have gone up and actual prices would have fallen or at least stayed flat because all the net new data centers would just be using the new energy infrastructure.

3:27But that's not exactly what happened. So from 2008 to 2021, America's annual growth in energy production was 0.1 % annually. Really, really bad. Not good. But it is getting better. The EIA's December 2025, short-term energy outlook projects generation growth of 2.4 % in 2025 and 1.7 % in 2026. That's like, you know, what, 20 times higher? That's great. But also that doesn't feel like, oh, fast takeoff, we're going to be doing 10 % jumps, we're going to be really, really ramping up here. So it feels like there's something that still needs to change. And then simultaneously, you have this, you know, massive political backlash to rising energy prices.

4:12And I think most importantly like we've identified some of the characters that drive AI you know we know Dario we know Demis we know Sam right we also have identified a lot of the people who are you know running the Neo clouds or doing you know build out of new data centers we don't really know the characters who's the Elon Musk of energy is something I keep coming back to there's no one who's really become like the main energy guy or gal, right? Yeah, I would say like Chase at Caruso is potentially a contender, but unclear yet if he's a Joe Rogan CEO. Other crazy stats. So while we're growing at like 2.4%, 1.7%, China is consistently putting up 6 % growth.

4:57China now accounts for one third of global electricity consumption and contributed 54 % of global demand growth in 2024. So more than half of the growth in demand came from China. There are two fields in which China is substantially behind the West, semiconductors and aviation. So there is one that I think he missed. And maybe you put it in aviation, but I think space travel, like they're way behind on rocketry. SpaceX is like the Boeing, Airbus, they don't have that equivalent. And maybe you put SpaceX and blue origin in the same category as Boeing and Airbus. I particularly don't. I think of them as separate technologies, and I think of this as potentially like the third category that China's way behind in.

5:40Obviously, we were tracking the performance benchmarks of the models. Then we were tracking the diffusion. What's the actual revenue? Is it sticky? There was a lot of risk of like, oh, this company comes up with some AI app. They ramp to 100 million. Will it stick around, or will they just gets swept by the wayside and people will go back to doing it the old way, or they'll use something else, or they'll pay way less. I don't think that that is a public perception. I would say the most popular topic. Well, beat the bubble will pop in 2025 allegations. At the start of the year, people were predicting that OpenAI would not do an up round, for example.

6:24And that just didn't happen. It was like all up rounds for basically all the major players. Like if you were long AI broadly in 2025, you did very well. In the other section of the Dan Wong piece, he talks about competition. And I thought this was just an interesting element to read. He talks about Sputnik moments. One might have expected the US to have roused itself after this bout of the trade war. But there have been too many declarations of Sputnik moments without commensurate action. I hadn't thought about that, but at one point we were memeing Sputnik moments for Sputnik moments. Remember?

7:01Do you remember this? Yeah, that was super. That must have been like a year ago. Yeah, it was like a year ago or something. But you know when it becomes like a meme that we're like, you know, everything is a Sputnik moment. Like it really is overused. Yeah, so Barack Obama declared a Sputnik with China's high-speed rail. Mark Warner repeated with Huawei's 5G. Mark Andreessen called it with DeepSeek. The more that people use the term, the less likely that society spurs itself into taking it seriously. It's a good point, right? Because I think the U.S. continues to systematically underrate China's industrial progress for several reasons.

7:33First, too many Western elites retain hope that China's efforts will run out of fuel by its own accord. Industrial progress will be weighed down by demographic drag, the growing debt load, or maybe even a political collapse. I won't rule these out, but I don't think they are likely to break China's humming tech engine. Yeah, it was interesting because you could say this was a 2025 letter, but this felt like the best possible summary of the geopolitical economic dynamic between China and the US. Anyways, go read it. But he called Europe cooked and chopped. Let's talk about Manus. This all started because Bill Gurley was an investor in Manus.

8:12And I think Bill Gurley has the best voice in venture. I agree. The best voice in venture, hands down. Who's got a better voice than Bill Gurley? It's amazing. He's got pipes. So this all started with Alex Wang over at Meta, formerly Scale AI. He says, excited to announce that Manus AI has joined Meta to help us build. Jack Randall in the X chat. Did anyone check on Dellian? Oh, yeah. That's funny because Jack worked with Dellian. Yeah, I mean, it really is. hey delian might be taking a victory lap because if miami comes back because of this wealth tax thing delia totally also that uh and patrick hollison yeah was bulbous oh yeah so did delia capitulate did they capitulate no no he gets to play musical chairs he's like what i've never said anything about manis i i don't know about that it's fine it's fine investment it makes sense dan wonk says the most read essay from silicon valley this year was ai 2027 the five authors who come from the AI safety world, outline a scenario in which superintelligence wakes up in 2027.

9:18A decade later, it decides to annihilate humanity with biological weapons. My favorite detail in the report is that humanity would persist in a genetically modified form after the AI reconstructs creatures that are, quote, to humans what corgis are to wolves. It's hard to know what to make of this document. He's talking about AI 2027. He says, because the authors keep tucking important context into footnotes, repeatedly saying that they do not endorse a prediction. He talks about humor within tech and the CCP. He gives a line from Sam Altman that is, Sam says, I think that AI will probably most likely sort of lead to the end of the world, but in the meantime, there will be great companies created with serious machine learning.

10:07I mean, that is really funny. This is the key problem for the tech industry in 2026. You need to paint a more, like the founders need to paint a more optimistic vision for AI, because it's not working right now. Because people see their energy bill going up, or they even hear about the idea of their energy bill going up, and then they see some of the slop. The slop is getting better, right? You look at some of the videos coming out of this whole Venezuela debacle, right? People in all types of roles are worried about job loss. And, you know, they see the investment going into AI and they're just scared, right?

10:47And so when you do quotes like this, whether they're serious or not, saying, I think that AI will probably lead to the end of the world, people are starting to ask, like, wait, why are we automating all of this in the first place? Can we just stop, right? Like, I don't actually I don't want my job automated. I'll keep doing it. The industry needs to figure out how to paint a more optimism, you know, because people, you know, maybe in 10 years reminiscing about a time when young people didn't just have shelter and an income provided for them. They had to fend for themselves. Right. And so in the 20th century, it was like, you know, it was easy mode being a politician or a CEO.

11:23Yeah. Because you could go into a room of business executives and said, OK, you currently spend like if you were running like an ai company yeah before the internet yeah before everything was recorded you could go and say like hey you currently spend a billion dollars a year on payroll yeah i'm going to be able to reduce that by 80 yeah there's gonna you're gonna be able to uh conduct lots of layoffs and then you could go on tv and say like ai is going to create an abundant you know future and that and the the clips wouldn't kind of cross everywhere right or where politicians could go talk to a labor union here and then go talk to business leaders here.

11:59And you just didn't have anything. I mean, we see this on our show where people come on and we're all having a conversation. And we're all like, oh, yeah, this makes sense. We all have the context. And then a clip goes out and people are quote-treating it out of context and being like, oh, I hate this. I hate what they're saying. And we're like, well, we weren't trying to get them to say something controversial, but they did outside the context of our show. Before you hate that you say it, go back and watch every episode of TVPN of 2025 just to get context. Before you criticize us. get some context.

12:24I don't know if the AGI-pilled people are so AGI-pilled that they think we're going to defeat the speed of light. Every physicist I've ever talked to has always said, no, that one holds forever. That's not like, oh, quantum computing, it can work, but it's a couple decades away, or fusion, or fusion, all these other things are somewhat engineering problems. I think the speed of light is going to be around forever. I don't know. Just need a faster horse. It's the Year of the fire horse, by the way. It is? Yeah. What does that mean? Wait, I thought they only - It's 2026, Chinese year of the fire horse.

12:56Wait, are they adding superlatives to the animals? I thought it was just horse, snake, dragon, and then they rotated through those. It's the year of the fire. We upgraded animals. It's amazing. It's the fire horse. It's amazing. So RAMP Labs modeled it out. They did a little analysis on the meta acquisition of Manus AI. They said the estimated price is$4 to$6 billion based on AI M &A comps. It's the fastest to 100 million ARR in history. Just eight months. Wow. And benchmark likely 8 to 12X in under a year. People were kind of putting the deal between$2 and$3 billion. I guess let's get in. Let's get it.

13:34What was, I guess, what was your immediate reaction? The thing that I was excited about is this is Zuck buying a product that people love. Well, that's a debate. But I like your take. But Mark Zuckerberg has done a number of talent acquisitions. Yeah, that's what I'm saying. So he just went through a talent acquisition era. Yes. They're the biggest investor in scale, right? But they didn't acquire the product, right? Yeah, yeah. Zuck has been saying, you know, beating the personal super intelligence drum. Yeah. It's been very unclear what that means. Yeah. Manus has been building high-quality agents.

14:12They're sort of time and time again I talk to people that are just very excited about Manus. They're like, just use Manus. It's the year of the agent. Use it. You'll see the future. And so just given Zuck's history from a product acquisition standpoint, I get excited because this can potentially give us some clarity around what they're like. One version of personal super intelligence is you have AI agents that can go out on the Internet and do things for you. They can go. I think in the context of meta, I think about like shopping. You see something on Instagram that you like, and you could trigger an agent effectively, like, go find this product.

14:52Like, you see a car that you like, go find this product. The elusive GT3 RS that you can't find. Yeah, basically, like, you see a make any image or video shoppable. And in fact, I would probably argue that LLMs in the chat form where you have a bunch of information with a knowledge cut off baked into some big model, a LLAMA 4, and it's vended through the Instagram search box that you chat with. It can never fully satisfy the vision of personal super intelligence, a personal assistant, something that actually can take actions for you. It feels like it's incredible. It's amazing. Like, obviously, for knowledge retrieval, it's great.

15:39But it was never going to fulfill the real vision there. So this does feel like, I don't know, I would hope that they bake this in in a really interesting way. I expect that Manus-type workflows will be heavily integrated into a, hey, when you say, hey, meta, and you're wearing glasses, like, hey, build me a slide deck to help me prep for my final later. Yeah. And I can imagine effectively the same product experience that Manus has today will be integrated into Meta AI. Yeah, there's a lot of areas where it feels very natural for Google to expand into, it feels very natural for Microsoft to expand into, Apple to expand into.

16:21Like Apple has a spreadsheet app, right? They have a word processor. They, like, Meta doesn't really have those. They did Meta at work for a while, but this whole idea of like go to open Instagram to do your homework feels crazy to me. And the same thing with open Instagram to build a slide deck, that feels crazy to me. Yeah, but everything that they're doing in AI, they're piping through Meta AI, the new standalone app. So maybe they want to just - To me, I look at that as like experimental area that you can just put anything into it. I was told by someone close to the action that apparently like the meta AI trough, the...

17:07The vibes? Yeah, meta vibes. Actually, the usage is actually insane. Really? No way. Trough? But I mean, it's terrible in the App Store, right? It cannot be ranked. Even Sora is not ranking well. One thing with Manus, this is maybe the first acquisition in the billions for a company with an Isle of Man domain. They're Manus.im. I am. You don't see a lot of multi-billion dollar.im acquisitions. Isle of Man. Tyler, is that where you're from? Isle of Chad. Did you get into looks maxing over the break? I've been bone smashing. It looks like you got into looks maxing, doesn't it? doesn't his midface ratio look different i gotta fix my uh recessed maxilla i feel like his maxilla is looking different it's looking better he doesn't have the filter on wait tyler go go like this do the the double jaw surgery yeah definitely they definitely have the filter can someone explain why meta bought manis what's the one pager for zuck other than we have an f ton of money so why not they know really really well how to build good agents and maybe that's enough nick dobo says His best tool set on the market.

18:19They bought a Swiss army knife to hand any AI model. Wide research, virtual browsers, VMs, code interpreter, PowerPoint slides, app builder connectors. So yeah, they're just good at product. I left meta because I made a bet that models were going to be commoditized and the value would be in the products on top of the models. But Metamate and Gen AI were highly politicized, sucking up all the oxygen in the room. As always, I was right. He's taking the picture. Hilarious line. As always, I was right. Ex-meta people are getting spicy. Did you see the Yann LeCun Financial Times deep dive? We should dig into that.

18:51But he took a bunch of shots at a lot of different people, basically just saying, like, you can't tell a researcher what to do, especially not me. He was very salty over Alex Wang coming in and becoming his boss, effectively. You got leveled. Well, yeah, it is an odd... I mean, it's a bold choice to put Yann LeCun on the bench. He's a, I think, touring award winner. He's like one of the greatest AI researchers in history. Where did he land on the Metis list, Tyler? Tyler, do you have Jan on the Metis list? Yeah, he actually wasn't on there. He wasn't on there? I think we were trying to rage bait him or something.

19:24No way. Yeah, he wasn't on there. Wow, we are so bad. So there were two major blockbuster deals before the end of the year. Unclear, I mean, were both sort of these zombie acquisitions or were any of them just clean, normal deals? I guess we don't fully know. We know that the Grok Nvidia deal was very much a sort of ghost ship type deal where you take the team, take some of the IP license, pay out the investors. $20 billion deal. What should we do with that, John? Should we hit that app-loving gong? We should hit that app-loving gong. We have a new gong, everyone. Look at this gong. It's bigger than ever.

20:01It's bigger than ever. We actually, it's so big. It's so loud. We have to be kind of careful with it. I mean, it's a John-sized gong. So John is 6 '8", if you're just tuning in for the first time. It's a massive gong. And it is a massive gong for a massive company. Fantastic company. There's been a bunch of chatter, says Dan Premack, about how Grok employees made out in the NVIDIA deal. He made some calls to find out. And in short, they did very, very well, even if not fully vested. We did hear that if you were there a very short amount of time and you were not senior at all, You might have had a bad run, but we'll have to dig into that a little bit more.

20:40But it seems like, in general, everyone sort of got paid. It sounds like Chamath did very well on this. He was a very early believer. Oh, I mean, Chamath haters were in shamble. We need to check on that. We need to check on that. If one of your friends doesn't like Chamath, call them. If you haven't heard from them, I would be worried. Are you doing okay? Let's just go hang out. Let's go play some golf or something. Take them out. Touch grass. Yep. Touch grass. Because it's going to be rough. for a while. The whole process took less than two weeks and was personally driven by Jensen Wong. No other bidders.

21:12Wong wired money early and wanted the deal to close before the new year. What a chat. What a chat. Send the$20 billion wire. Sir, we haven't fully executed the docs. I'm good for my$20 billion. Why is NVIDIA buying Grok? Is it a response to the TPU? We'll have to get some of the semi-analysis folks on the show to dig into it. They've been on this super crazy roller coaster. I think not a lot of people expected them to wind up at 20 billion. There's two sides to it. One is like when if you read the actual social capital memo that Chamath put together, it makes so much sense to invest at the valuation he did.

21:48With a lot of the custom silicon, the custom chips, you sort of like point your bow and then you release the arrow and then you just like wait like five or ten years because like tape out, just tape out, just like going from the design of the chip to actually getting the first one made takes like 18 months or something like that. Like it takes years and then you're sort of locked into the specific spec and you're building an ecosystem of developers that can run on your chips. And so like there's this, you make a bet on a certain architecture and there's other firms right now that are betting on like we're going super long the transformer architecture.

22:21We're going super long cerebrus like wafer scale. We want to put the whole model on one chip instead of a rack of a whole bunch of or we want to be really memory constrained or have a ton of memory or something like that. So you make your decisions about what trade-offs you're making, and then you kind of just wait. And you hope that people come up with a use for what you made. Is this offense or defense for Jensen? I mean, I think he's like almost, even though he's not literally the richest person in the world, he's running the biggest company, so he has the most capital, like fire around at places.

22:53And I think to put this into perspective, this is like us saying, investing in a nice camera, right? Yeah, yeah, yeah. Like it's not the kind of like, we take like, Jensen personally led this process. It took him two weeks. He wired early. You would do the same thing for a nice camera. Holy effing ass. Not swearing really hurts the quality. You can read it on the screen. But he's very sensational. So it's a screenshot from Reddit. it has 36 million views and 207 ,000 likes. It's a massive post. It says, I am a developer for a major food delivery app. The priority fee and driver benefit fee go 100 % to the company.

23:35The driver sees 0 % of it,$0 of it. So here's where it gets conspiratorial. It says, I am posting this from a library Wi-Fi on a burner laptop because I am technically under a massive NDA. I don't care anymore. I put in my two weeks yesterday and honestly I hope they sue me. I've been sitting on this for about eight months just watching the code push. I hope they sue me but I'm on a library Wi-Fi with a burner. We're gonna dig into all the different aspects of this but this is crazy. Some people in the chat already read this but we're gonna go through it. So just watching the code getting pushed to production I can't sleep at night knowing I helped to build this machine.

24:16You guys always suspect the algorithms are rigged against you, but the reality is actually so much more depressing than the conspiracy theories. I'm a back-end engineer. I sit in the weekly sprint planning meetings where product managers, PMs, thanks, discuss how to squeeze another 0.4 % margin out of, quote, human assets. That's literally what they call drivers in the database schema. They talk about these people like they are resource nodes in a video game, not fathers and mothers trying to pay rent. First off, the Priority delivery is a total scam. It was pitched to us as a psychological value add.

24:49Like I said in the title, when you pay that extra 299, it changes a Boolean flag in the order JSON, but the dispatch logic literally ignores it. It does nothing to speed you up. We actually ran an A-B test last year where we didn't speed up the priority orders. We just purposefully delayed non-priority orders by five to 10 minutes to make the priority ones feel faster by comparison. management loved the results. We generated millions in pure profit just by making the standard service worse, not by making the premium service better. Even in this scenario, it's like maybe unethical, but like this is actually a bull case for paying$299.

25:26If the default service is actually slower and$299 gets me five to 10 minutes faster. This made me think they should add a feature that's whenever you get around, I'll take it whenever you get around to it. And so it's just like, it's just like really, really, I'm feeling lucky. I mean, no, it's just like the lowest cost version. Oh, yeah, yeah. You might not get the pizza for three hours. Well, that was like Lyft Line. Do you remember that? Or Uber Pool, those? Amazon does this for packages too. Oh, yeah. You get a credit if you like take it. Yeah, yeah. You get like a Kindle store credit or something on Prime.

25:56It's kind of interesting. But anyway, it goes on that to much darker places. He says, but the thing that makes me sick, the main reason I'm quitting is the desperation score. We have a hidden metric driver for drivers that tracks how desperate they are for cash based on their acceptance behavior. If a driver usually logs on at 10 p.m. and accepts every garbage$3 order instantly without hesitation, the algo tags them as high desperation. Once they are tagged, the system deliberately stops showing them high-paying orders. The logic is, why pay this guy$15 for a run when we know he's desperate enough to do it for$6?