I see a large IPM on the horizon. You're surrounded by journalists. Hold your position.
Overnight success. The place of 30,000. Double. Right.
There's misinformation. Clearing order inbound.
Let's just roll. We are surrounded by journalists. Hold your position. Come on. Get up.
Trust the experts here. We are experts.
Founder Mel. 5.
I see multiple journalists on the horizon. Stand by. UAV online.
Double blaze. Triple blaze. Double kill. Fight is team deathmatch.
We are experts. Triple blades. That's just wrong. Right.
Hardy clearing order inbound. We are surrounded by journalists. Hold your position.
Strike one. Strike two. Activate golden retriever mode.
Marky clearing order inbound. Five put. I see multiple journalists on the horizon. Founder.
You're watching TV. Jord's juggling. It's Wednesday, February 18th, 2026. We are live from Ultra Dome, the temple of technology, the fortress of finance, the capital of capital. Let me tell you about Ramp.com, baby. Time's money, save both, easy corp cards, bill pay, accounting, a whole lot more all in one place. We have a great show for you today, folks. Specifically, Tyler Cosgrove has been on a little bit of a tear with the market maps. He dropped the final market. We don't need any more market maps because Tyler made a market map that has every company on it. Let's pull up his latest market map. It looks like the...
There was some VC associate out there that was making a market map and was just devastated.
All my, all the companies I was going to put on the market map are now on this market map, which is in the timeline by the way. And it's very blue and you did select Google. Can you walk us through how you built this particular market map of every company on it? It's every company that has a Wikipedia article. Correct.
Correct. Yeah. So the one we're showing, that's the wrong one, that's for later. So basically, I, over winter break actually, I was interested in this thing where like, okay, on Wikipedia there's like all sorts of like, Wikipedia I think is like a very underrated data source. And there's like all sorts of cool things I think you can do, right?
You mean Grapedia, right?
Well, so Grapedia is a little different because it's like generated on the fly, right? But basically, you know, whatever. What I ended up doing is, I took every Wikipedia article, there's like seven, seven and a half million English ones.
And I ran them through an embedding model. It was Qwen 3 embedding 4B, I think.
Okay. So, but I got embedding for every single article, right? So it's like basically...
Every article has a vector. It's like 2500.
You did this a while ago, right? The whole Wikipedia embedding or did you re-embed?
This was like a month ago.
Yeah. Yeah. Yeah. I remember.
So then basically I took all the articles. I found all the ones that are about companies, enterprises, right? Which is basically...
You can find some direction in the embedded space that corresponds to how much companiness something has, right? So you just find all the ones at the end.
Really? Oh, you don't filter by like Wikipedia's categorization of whether or not...
So I use that, but that's not inclusive of every single company. So it's like a little bit blurry because some things are like, well, is it a company? Is it not?
Yeah, I noticed some like railroads on here that looked like maybe they're companies, but they're like state-owned. And where is that?
It's kind of a blurry thing. So you can't just use just what Wikipedia says.
But you can basically find things that are companies and then you have an embedding for every single one, right? So it's this big vector, super high dimensional space. If you map it down to 2D,
you can have this like cool 2D map, which is basically what I did. So you can see there's these big clusters, right? So it's like in the top left there's all these theater companies or...
There's space companies.
I noticed the aviation companies were pretty far away from the train companies. Is that deliberate?
I mean, you knew there was kind of like rivalry.
Yeah. Rivalry. They need to be... You got to keep those apart or they'll just fight.
Like when you map something down from like, you know, there's like 2,000 dimensions down to 2D. It's like very hard to keep.
Like a ton of things.
And it just randomly looked like the United States.
Yeah. That has nothing to do with that. It was totally random.
Because I looked at it. I was like, 'Oh, okay. There's a lot of companies in Florida, a lot of companies in the Northeast like...'
Yeah. I didn't even realize. I was like, 'Oh, it kind of looks like...'
And then I was like, what is this? What is this enclave in Canada? Why does that... is that Alaska or something? But in fact, it has nothing to do with the United States. It just happens to look like the United States.
Yeah. But this... So it's like actually interactive, so you can like look up a company and you can find where it is and stuff.
tylercosgrove.com/wikipedia_map.html. Wow. Really a wordsmith with the URLs there. Tyler couldn't use a TLDD-less domain. There are some fun ones in here. Anyway, that's a fun project. All the links take you to Wikipedia. Go check it out. And we market maps are basically done, but a lot of the Neolabs are not on this market map. Quickly, let me tell you about Ream. One live stream, 30 plus destinations. If you want to multistream, go to Ream.com and let's click over to Tyler's market map of the Neolabs because we've been tracking the Neolab boom. We've had a lot of these founders on the show. We came out of the world where we were like, okay, there's DeepMind, there's Google, there's OpenAI, now we got Anthropic, there's banking machines, and there's a couple different companies. But the Neolabs have exploded. They have a term that's been coined. Sarah Guo was actually I think the first person that came on the show and sort of broke it down for us around the Christmas episode. But since then the Neolab taxonomy has evolved and so we needed to build a market map. So Tyler, take us through what's going on in the world of Neolabs these days.
Yeah. So Neolab is kind of this interesting term. Like it's very broad. People say like Neolab, it's not very clear what they mean.
Because there's like broadly, I think it generally...
And this will make it clearer.
Yes. I think after this it'll be pretty obvious like what you know what you should be looking at, how to think about these different companies.
Yeah. I don't want to be more confused at the end of this. Yeah.
That would be a disaster if that happened.
Yeah.
This is going to be easy. Okay. Got it. Got it. Got it. Cool.
Okay. So let's just start. Okay. So you have Neolab, right? Yes. So Neo, this prefix, okay, it has to be relative to something. Yes.
So Neo is relative to like your traditional.
This is your... Yeah. This is your open up to the big labs.
Yeah. They don't get enough credit today. Building data centers, spiking capex.
So this is going to be your OpenAI, your DeepMind, your Anthropic, kind of your big lab. Yeah. XAI.
XAI kind of fits in there, too. Even though it's a newer trad, it fits in with the big lab. They got a lot of money.
Dario, I think he would... he was like, yeah, three maybe four labs, right? So the fourth is probably XAI.
I think you can also kind of throw in Mistral in there.
Okay. Mistral is a little bit older. Yeah.
Yeah. Yeah. I mean, there's much of these labs that were basically founded in the like two or three years before ChatGPT and then in the like six months after. Yeah. So I think XAI's in there, Mistral's in there.
And these specifically, these I feel like those trad labs, it's like they did a transformer-based pre-training run. They have their own base pre-train. Maybe it's not at the frontier, but at least they're playing that game. They're not doing fine-tuning. They're not doing something else. So that's sort of like you're in the trad world when you're thinking about like a big pre-train run loosely.
Yeah. I mean, especially if you're talking about these big pre-trains, it's really just these four. No one else is really at that scale.
But then, okay, so Mistral kind of brings us down into what I call the sovereign labs.
So, I mean, you know, if you kind of look at this, it's basically just labs that are not in America, but I think also there actually is some meaning to this. So like Mistral, you've seen Mistral become kind of the leader in European AI, right? So I think it was in... was it Sweden? Maybe they're building a new data center.
Sweden. So they're kind of becoming...
Stuff going on in France, too.
Macron is always talking about Mistral. It's a big leader. Cohere is also kind of... I think as like a very, you know, it's a Canadian company. Yeah.
But also has done their own pre-trains, has done...
No ties to the curling team though.
Complete. No ties. So I don't want them.
Yeah. Yeah. It's important to put some distance between that scandal.
Yeah. You can go down, you can kind of see all your Chinese open source labs and see your Qwen, DeepSeek, Kimi. Unitree is also in there, right? Unitree, I think. So as we'll see later, there's also... I have a section for like robotics labs. Sure.
But this is very clearly like, you know, this is the Chinese.
Yeah. Take us back in time now. What was going on before the trad labs broke out?
Yeah. So here I have this section, legacy labs. Okay.
So these are ones that were kind of more entrenched in these big enterprises.
So you have stuff like Microsoft Research. Sure.
AT&T Bell Labs, right?
Oh, Bell Labs. Yeah, I forgot about Bell Labs. After you know what, you know how, you know why they call it Bell Labs.
Why do they call it Bell Labs?
Yeah, it was founded by him.
Bell Labs.
Okay. But also you have stuff like... you have FAIR, Facebook AI Research. This was like... I mean there's so many like OG research papers that came out of FAIR. This is what Yann LeCun used to be head of before it transitioned to Meta. Okay, so then I think let's move up here. Around your trad lab you also have postlab, right? P-O-S-T.
Yeah, these are labs where you get a lot of posters, right? So obviously this is OpenAI, you got Rune, Anthropic, a lot of, you know, scholar posters. Prime Intellect, I think posters, Brown.
A bunch of anons at Prime Intellect doing great stuff over there.
For sure.
And then you kind of get into...
Yeah, the proper Neolab. So this is also a bit hard to identify because like what is actually the core of a Neolab? What are these different kind of offshoots?
I think Prime Intellect is kind of the prototypical, like quintessential Neolab when you think of it.
Where you basically have... it's like fairly recent.
It's still very much research focused.
Like sure they have enterprise, like you know...
Think about different stuff, but at the core of it you're still like trying to find these like new novel approaches. It's research. You're hiring researchers. It's not just like engineers, sales guys, etc.
So let's...
Is it... Wouldn't Sakana be more of like a sovereign lab?
Yeah. Yeah. I mean, so a lot of these can fit in all different places. Sakana would be... Yeah. Japanese maybe.
Okay. And you put Mistral in here because it's a new project.
Yeah. This one was also a bit hard. It doesn't feel like a trad lab because I mean, maybe it has the scale, but it's just...
It's newer. They haven't shipped yet.
Yeah. Neo, new lab, right? I mean, it's so recent.
Definitionally. Thinking Machines is my classic go-to Neolab.
I feel like it's post-OpenAI Exodus and sort of OpenAI is nothing without its people. You know, you get the spinouts and Thinking Machines and SSI are two of like the first case studies that sort of set the tempo for, okay, it's possible to do some research outside of the big trad. And so that's where you get the Neolab boom from. And then a lot of the other companies I feel like are saying, 'Okay, we're going to do something similar to Thinking Machines or SSI. We're going to commercialize early or late, but we're following in that and we're benchmarking to that.' Oh, they raised two billion. We're raising 200 million. It's easier. There's a 10% chance that we, you know, are at their scale. So you can underwrite it that way.
Yeah. So Thinking Machines also brings us to what I call the trad SaaS lab, right? So you have SaaS lab, you have trad SaaS lab. So I think the way I think about this is the trad SaaS labs are trying to basically use the data that's inside these big enterprises, pull them out with AI. Okay.
So this is Thinking Machines, right? The rumored idea, right, is they're doing RL for enterprise. A bunch of these are doing fairly similar things where it's kind of chatting with your data, using the data that's very valuable to a company but it's going to be inside the company. You can't really pull it out anyway besides having the AI be like internal. So you have Applied Compute, you have Poolside doing all kind of similar things in this enterprise LLM field.
And then that brings us to Neo SaaS.
Not full base pre-trains for those companies, mostly fine-tuning or RL on top of a particular company's use case.
Yes. And then I have Neo SaaS lab. This is different than trad SaaS. I think these are different in that they're not really pulling... they're not going enterprise specific maybe. I think that's one way to look at it. Also much more of like startup focus, right?
But they're making a product that is sold effectively as SaaS.
So Cursor, Cognition, Windsurf, Ramp.
Ramp Labs. These are seat-based, sort of consumption-based, but it's a product that's vended into a... and the product is what you get and then sort of customizes as you integrate it, but it doesn't... the conversation doesn't start with a business development relationship.
Yeah. And of course, I mean, these lines are pretty blurry. But then, okay, let's go down to the postlab. Okay. Right. So postlab. This is after the lab. Yes.
So that means like basically they train the models and then these labs are working on top of those models. That's how I think of it. Right. So you have Meter, you have Epoch, these are going to do eval.
You have Pangram. They're seeing... is the model producing slop?
Yes.
Or is it producing text that you're using in some way?
These are purely eval. They don't necessarily have AI products themselves. They don't necessarily sell to big business.
But they could still be training models, right? Like Pangram is training models that sit on top of the lab.
That's true. So it counts as a lab.
Makes sense. What else we got?
Maybe that brings us down to the safety lab. Yes.
So these are pretty interesting. Anthropic kind of fits in this, right? Because they have a big safety team. They're doing a lot of mechanistic interpretability. You have Goodfire, I think they just raised at like 1.25 billion and they're just doing mechanistic interpretability.
Very interesting. Luther AI is similar kind of...
Yeah. Very cool.
They're also... a lot of these are also kind of in the open source space.
Yeah. Stable Diffusion came out of Luther AI.
Yeah. This is another label that I think I could have put on, but it's so hard to get everything to like work together, but a lot of these are also like the core of the company is doing open source stuff.
Sure. Sure. Sure.
Right. So Prime Intellect was a good example. Almost OpenAI back in the day, but a lot of these have bled together where OpenAI has an OSS model, but also a lot of consumer and enterprise. Yeah, makes sense.
Okay. So then in contrast to the SaaS labs.
We have the consumer labs.
So these are focused on consumers, right? So you have Eureka Labs. This is Andrej Karpathy's project. Yeah. I don't think there's anything been released from it yet.
But yeah, education. It's for people. You have Humans.
Oh, it's four. Four people, not four individuals working there.
It might be four people. It might be one person. Who knows? He's pretty good.
Yeah, you have Humans and... Okay. Right. This is the... I think I phrase it as like humanity focused.
Human sand.
Yeah. We got to hang out with the founders at the Super Bowl, but yeah, focused on creating models that work better alongside people.
Sure.
Yeah. You have a lot of like companions, these kind of ideas, right? You have Character AI also.
Oh yeah.
Do they really own C.AI? What a great domain.
If that's true. I don't know. We'll have to back it anyway.
So then that brings us down to the visual labs, right? These are a lot of either multimodal models or they're actually like producing video or images, right?
We talked to a lot of these founders. I feel like almost all of them have done...
Yeah. I mean, World Labs raising today. Yeah.
Or fundraising announcement today.
Yeah.
Midjourney, etc. These are pretty obvious. You have your Neo Auditory Lab.
Midjourney is the sailboat logo, correct?
Okay.
You have Meta Reality Labs on there, too.
Oh, okay. Yeah. Yeah. Yeah. Yeah. Yeah. That makes sense. They're visual. Not fully AI yet, but they're getting there.
Yep. You have... Okay. You have Neo Auditory Lab.
Right. So this is going to be anything that has to do with vocals or voice or music, right?
11 Labs.
Of TVPN. Thank you.
Suno, right? Making music. Gemini also released a new model, 3. I didn't even know there was a one or two. It's a trilogy already. They just got secret models that they're hiding from us.
Yeah. Okay, so this is a very interesting field and then you have your legacy auditory as opposed to your new auditory, right? So these are your old ones. This is... well John, do you want to talk about Nuance?
Nuance Dragon NaturallySpeaking. This is the original box software. You buy it, install it on a Windows computer, you can talk into a microphone and it will write down what you say, dictate it. Yeah. Using some AI, not a large language model at the time, not a transformer-based architecture, but became a very large company. I think it's part of Microsoft now or something. I think it's been acquired a few times. But yeah, very, very interesting company. Lot of really solid... Fruity Loops. Yeah, that's a... you know, you're in the lab making beats, I guess. Makes sense.
Okay, so now moving up. I think this is really the very interesting section. So this is Neotrad Lab.
So I think what is a Neotrad Lab?
This is the simple... it's just a simple definition clearly. Does it even need explaining?
To your head by the way. Yeah, you're coming really close.
You might want to be on the other side of the team. Okay. So Neotrad Lab, it's a Neo lab, but it's trad. Okay.
Okay. So what does that mean? So basically, the way I think about a lot of these labs is that they're extremely research focused.
They're also largely focused on like kind of single idea.
If you think of like OpenAI...
Very research focused obviously, but they're doing a lot of different things. Yeah. Right. So they have consumer.
Yeah. They have consumer, but it's even like on the product or on the research side, right? Doing their video.
Yeah. But even within like language models, I'm sure they have a, you know, continual learning team or all these like weird things where I think a lot of these Neotrad labs are basically focused on one single moonshot idea.
Mh. Okay. So example, Flapping Airplanes, right? They just came on. They're talking about data efficiency. This is kind of the one kind of moonshot idea, right? Obviously it's like a very general broad...
A bunch of different ways you tackle it, but they're like, that's the problem that we're going...
But it's one specific thing they're working on. Yep.
And I mean, they talked about, oh, you know, if we figure it out, there'll be some value, but we're not exactly sure how it's going to come out like right now, and we're not sure how we're going to productize it necessarily, but we have a... but we have really...
Yeah. But so the idea is like if these labs can figure out like the core research idea, then the value will appear, right? So this is... you also heard this out of Ilya with SSI, right? Not sure how they're going to get revenue, but it'll come if they figure out a breakthrough.
Build it, they will come. Yes.
So a lot of interesting things here. So we can look at like, okay, General Intuition.
They're basically doing a lot of multimodal training where they can basically take video game data and try to figure out how to map that onto LLMs or models or these types of things.
Okay. You have Inception. I believe they're doing Dream Tech.
Wait. Okay. I'm thinking of Logical Intelligence. They're doing like diffusion models. Okay.
Right. So diffusion but for LM... but for text. Yeah, we've seen a demo from Google on that.
Okay. Inception is doing... I think they're doing the energy-based models which is kind of this weird thing. Okay. Wait, I have both of those companies flipped again but it's so Yann LeCun is...
Simple. I mean, I don't know why you're flipping stuff around. Like this is literally just Neolab 101. You're doing a basic breakdown.
The point is that they're doing these like kind of weird architectures where like energy-based model. It's like kind of different than a normal LM where you have this normal backprop stuff like this.
But the point is that like these are all like very kind of weird architectures that they're working on. So maybe the big labs have like small teams that are working on this stuff, but basically these people go out of the big lab. A lot of them are coming out of the big labs and they're starting these new projects. Okay.
Like coming out of a trad lab or a Neolab or a Neo or legacy lab.
Exactly. Yeah.
Okay. Got it.
Yeah. Okay. So now let's move up a little bit. We have...
Neolab Lab. Okay. So this is... Yeah, I like this one. So these are a lot of companies that are focusing on... they're also like very research focused, but the point of the research is to build essentially like a researcher.
So it's... they're recursive, right?
You have Recursive and Recursive.
Yeah, you have actually two that are Recursive and Recursive, right? You have Richard Socher. Okay.
You have Periodic Labs where they're not... they're a little bit more focused on the hardware.
But the whole point is that they have this kind of closed loop.
Where you can basically build a lab. Yeah. Within the lab, right? That's the whole point. Neolab Lab. You're building a lab.
Unconventional AI, similar thing. I think they're...
The product will be a lab. That they're in the lab manufacturing business.
Correct. Got it. Yes. Okay. Moving up, we have Math Lab.
So these are pretty interesting. Axiom and Harmonic. Yes. And then you have Math Lab, right? Yes.
But these are pretty cool. There's been a lot of good breakthroughs recently. I think there's a bunch of math problems that are being solved or maybe they're being proven in some ways, but there's a lot of like interesting research coming out of these.
Harmonic is Vlad Tenev, the founder of Robinhood. Yes.
So these are your biolabs.
Oh, you got LabCorp. Yeah, I'm familiar with LabCorp.
LabCorp. But there's a lot of like biology focused labs. It's actually like I didn't know a lot about a lot of these.
But there's all sorts of like interesting research. So Isomorphic Labs. This was spun out of I believe Gemini or at least Google.
Yeah, that's right. They're working on like longevity and just drug development almost.
Yes. So some of these are very focused on specific forms of drug development.
Some of them are just like broader where they're very focused on longevity stuff.
Cool.
And then, yeah, let's go to...
Yeah. What's going on up top? Oh, yeah. Up top we have Labrador.
Oh, that's really important. If you want to understand labs, you got these. You got your lab, the foundation.
your black lab, your chocolate lab,
the chocolate lab. Yeah, chocolate labs are important.
Yeah.
If you want to understand labs broadly.
Yeah. Okay. Then moving back down, we have the Neokinetic Lab.
Okay.
So these are going to be your labs that are more focused on robotics. Yes.
So you have a bunch... you have Project Prometheus. Yes.
This is Bezos's lab. It's still kind of in stealth, which is why there's not even a logo for it. Yeah.
You have Figure, you have Skilled AI.
Skilled AI is the Luke Metz project. Yes.
Got it.
Yes. Physical Intelligence, Sunday, right? These are all your kind of Neokinetic labs, right? These are started fairly recently in the past like maybe four or five years.
Neo Neolab, right? Okay. So 1X is building Neo robots. So they're Neo Neolab sense. Yeah.
And then Legacy Kinetic is the previous...
Legacy Kinetic is kind of the old gen. Yeah. But cooking.
They're cooking. Yeah. Waymo's cooking. Cruise and Boston Dynamics have been a little bit behind. Zook's also another self...
There's a bunch in here that I kind of...
There's another one, Stealth. I think that never really hit inflection. Okay.
Yeah. And then you have your... mostly vehicle focus.
So this is working with the government.
I have... Yeah. I have Shield AI. I also have DARPA.
DARPA is a lab. Yeah. They invented the internet, right?
Yes. That's good. And then simulation lab. I think that...
Simulation lab. Yes. So Simuly, we just had them on.
Space. SpaceX. You could put up there because aren't they working on this Pentagon?
Yeah. Where's Rocket Lab? Rocket Lab needs to be on there. That's a lab.
There's a lot of labs that I mean... Yeah, lab is a very broad term.
Very broad term. Well, at least it's crystal clear now for everyone.
Yeah. So I think this should be pretty obvious to anyone who's thinking about Neolabs like how you should be thinking about them now. If you've been paying attention, this is all second nature to you.
Yeah.
Did you add up how much all the companies have raised? It's got to be in the north of 200 billion.
Yeah, it's a lot. I mean, so I didn't do that, but for a while I was trying to figure out how to valuations on the map.
Yeah, it was too complex.
You didn't want to... you didn't feel like you could do the math.
No, we don't know how.
Well, it's also a lot of them are rumored. It's actually like kind of hard to find out because a lot of these are still really in stealth. Like a lot of these Neotrad labs, they basically... because the whole point is that they're doing this like research stuff.
They're not going to like productize early.
Yeah.
And also how much do you put in the DeepMind bucket? That's a huge amount of investment and it's not exactly disclosed. Do you count the TPU? Do you count Google Cloud? Like different allocations. You can go really deep in the stack to understand the impact of like the broad AI buildout. But yeah, I mean, if you just total this up, you can really just do XAI, OpenAI, Anthropic and get like 90% of the way there and it's probably like 200 billion.
It's also hard because it's like evolving so fast, right? So David Silver's lab, who he used to be at DeepMind.
I think that means rumored today.
But these things are coming out like every day.
You put the typos in just to prove that.
Human. So like Sovereign Lab and then Sovereigntable Intelligence also has a typo.
And so I just wanted to make sure... I wanted to make sure people...
Yeah, you put the typos in so that it was proof that you made it. Yeah.
Well, yeah. Whatever you built this in doesn't have spellcheck, I guess. Anyway, fantastic report. Thanks for breaking it down. I learned a lot and I hope you did too. And let me tell you about a lab, Gemini 3 Pro. It's Google's most intelligent model yet. State of the reasoning, next level vibe coding, and deep multimodal understanding. And I'm also going to 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.
One show, two maps.
Strong start. Should we break down five wildly obvious fixes that will explode consumer LLM adoption? They don't want you to know this over at the big labs, but I have some ideas. Basically, everyone's been really focused on agentic coding and the SaaS apocalypse and what's happening in the business-to-business world and the enterprise world. I've just been sort of like thinking back on, you know, basic improvements to the chat apps that I use all the time because there's some really obvious stuff that I think is in the works. I think it's coming, but I wanted to just sort of like get it all down in one place to think about what the next iteration and the next breakout moment when people are like, 'Oh, I'm using them even more. I'm having a better experience.' What would that look like? So the first thing is that I realized that I've asked ChatGPT just when was OpenAI founded three different times. It's the exact same query. Like it doesn't need to light the GPUs on fire for that question. The answer literally never changes. You can cache the result. And that's what Google does with those knowledge queries, knowledge panels, and there's a whole bunch of different ways to deliver results that are sort of pre-cached. And so if you just look down, when an LLM launches, basically every question has never been asked before. But now there's a lot of people that are just showing up with the exact same question. Give me the history of the Roman Empire. Give me the history of this company. And you might not be the first person to ever ask that question. Exactly. But also, if you do a little bit of fuzzy search over it, there are probably like hundreds of thousands of people that have asked the exact same thing. So cache those results, give them to the user instantly. And I think this like instantaneous feeling of LLMs, like they felt slow for a really long time. They actually got slower. Like it was always sort of slow. You watch the token stream in, but then once the reasoning models and the thinking models and the deep research and the o3 Pro came out, it was like really slow. It was like close your phone and come back in 20 minutes. That doesn't have to be the end state. And I don't think it will be. And I have no better example than the number two on my list, which is cerebrus inference. So ChatGPT currently has a model called 5.2 Instant and it is not instant at all. I fired off a prompt to 5.2 Instant and I said no reasoning, tell me the history of LLMs. It took 38 seconds to deliver the full response for like all the tokens to stream in. And it does a good job. It shows you images and stuff and it is a cool illustration. But then I went over to Codex desktop and I fired up GPT 5.3 Codex Spark Low,
which is a crazy name which we'll get to. And it responded in under two seconds because it's, from what we know, Spark is incredibly cerebrus and it's very, very fast. And so everyone's obsessed with like the fast models in the agentic coding world because you're waiting a half an hour for something to get back to you. You're waiting five minutes for something to get back to you and you're actually losing your train of thought. But I think that applies in consumer as well. And I think the interaction of sending a message and then just immediately getting a response before you actually think, 'Oh well, like it's waiting. I'll close the app. I'll check my messages. Oh, I got an Instagram notification. Let me go over there.' Like instant responses will keep people in the apps longer and user minutes will actually increase once that rolls out. So pretty simple implementation for I think most companies. My big question is like I know Google has a huge advantage with TPU but I don't know if they have like an answer to cerebrus specifically and Nvidia just brought Grok which can do I think some of the same things. So I'm curious to know how every lab solves the like fast responses question because that feels like an important piece of the puzzle. It's not the only piece of the puzzle, but it's an important feature and I think we're going to see it rolling out to consumer LLMs very, very soon. And I do think it'll be an interesting moment for people to both ask a question and just boom, it's as fast as going to Wikipedia and just seeing like, okay, everything's rendered, it's thoughtful, it's what you want. And on the flip side, I think that it could make people a lot more chatty with them, like actually asking follow-up questions because you don't feel that cost of like, oh, if I ask you to follow up and tell me more or go a different direction, like I have to wait. I have to wait, so I might just close the app. Speaking of GPT 5.3 Codex Spark Low, no more model names. Like truly no more model names in consumer AI LLM chat apps like ever. Like just bury them so deep in the UI that you never see them. And people will complain. People will be like, I wanted it to be easier to pick. I like picking between Pro and Thinking and Fast and Instant. I know what I want for everything. People complain, but it will all inspire the model routing team to grind harder. And the model routing team has a hard job to do, but they will eventually figure it out. And eventually you should be able to just talk to the model. You can already do this in chat. You can say, hey, think really hard about this question and give me a really thorough answer. And it'll switch from instant to thinking. I don't know that you can trigger Pro from that. I haven't actually experienced that. I did try and trigger a deep research report. I said, hey, please deep research the Roman Empire for me. And it does not fire off a deep research report. Deep research is buried under like a plus button and you have to select it and say, okay, I actually want you to do this thing. And then it takes you down the deep research like workflow which I understand is like for inference reasons. They don't just want you firing...
off deep research reports all the time. But I think in the future, like, the model router should be very intelligent about, okay, this is a question that people have asked thousands of times. Let's just go get it from a database, which is crazy to think in the age of AI you wouldn't even be hitting a GPU, but I think that's going to be real. And then I think on the other side, like, it should detect, like, okay, this person wants something that's far beyond anything that we've ever worked on before. We got to go search the internet. We got to write some code. We got to do a whole ton of stuff. So let's fire up deep research. Right. Um, fourth, ads. We've talked about this, but we got to get them in the LLM. We got to get them everywhere because I was thinking about the death of Google Reader. I don't think you were ever a Google Reader guy, were you? But it was amazing. You could take all these RSS feeds from all these different blogs during the blogosphere. You could put Marginal Revolution, Tyler Cowen's blog, all these different stuff, all these different things in there and just kind of scroll through them really quickly. And Google wound up killing it and everyone was like really upset. And the reason was, I think, because they never really got on the Google ad flywheel where there was real, like, revenue generation and
Yeah. And was that just that it didn't hit a scale that enabled it? Makes sense.
The failure of every Google project that has failed is always a question of, like, was it because they weren't making money from it or was it because they hadn't monetized it yet and
or it just never got big enough?
Never got big enough to monetize
weekly actives. It's not probably worth keeping around.
Totally. Totally. But my takeaway from Google's surface area of products that are successful and loved β Google Search, Google YouTube, Google Maps, Chrome, Android β like, these are all direct funnels for the ads flywheel. And so you can see that they're driving the bottom line. There's a whole bunch of folks on the team that are getting excited when they're hitting their numbers, when they're making more money for the business. And so they just get more and more resources, more and more engineering effort. Everything gets better. And I think that not only are ads the best way to deliver high-quality products to the broadest possible audience, but they just make products better top to bottom. And yes, there's the stated versus revealed preference thing. And yes, you might want to pay to not have ads like you do on YouTube. Um, many people do. But, uh, I do think that that ads flywheel is going to be really, really important as the inference gets really
and right on time, Perplexity ends ads experiment.
I saw that. This was the news from this morning in The Information from Katherine. He says, "Perplexity is no longer offering ads." An executive told the Financial Times the AI search startup is pulling back from this line of business as rival OpenAI starts showing its users ads in ChatGPT earlier this month. The company said it worried ads would undermine users' trust in their platform, with an executive saying the challenge with ads is that a user would just start doubting everything. I don't buy this. That's all weird. Uh
Arvin has a history of kind of just, like, trying to provoke OpenAI at every turn. And so coming out
uh, that this is just, like, somewhat bearish, right? They're trying to serve as many people as possible all over the world. The best way to do that is going to have an ad-supported tier.
Kind of bailing on this moment. I don't know.
Maybe it's not worth reading too much into it, but a little bit early to throw in the towel on the economic engine that has driven the internet for its entire history.
Yeah. I mean, we talked to a lot of founders who have brands and they love advertising. And I think that's another side of this, which is that when, like, a lot of entrepreneurs and also people who work at businesses want to grow their businesses and they have fond memories or affiliations with Facebook and Google because that's how they grew their companies. And when you talk to somebody like Sean Frank at The Ridge, he's like, "I'm going to be first in line to advertise on ChatGPT. I can't wait for that. It's converting so well already. I want more of that business." And we didn't really hear that with the Perplexity ad product. We didn't hear people lining up to buy ads in that product. So maybe it was not going as well as they thought.
Uh, so according to The Information, Perplexity started testing advertising in 2014, less than a year into its test, Taz Patel, the executive leading the ads effort, left the company, and Perplexity had only let in less than
half a percent of the brands that wanted to advertise on the chatbot. So there was, like, a bunch of demand.
They barely let anybody use it and then they bailed on it.
Interesting. Well, the last one is somewhat related to OpenClaw. Uh, but I think way down the funnel beyond the 20-minute deep research project, uh, you probably want to be able to fire off something that looks like Claude Code or OpenClaw or Codex to write lots and lots of code and solve a really, really hard problem. Uh, and so many reasoning models can already write some Python and execute it, but it's clear that everyone wants to go further, hence the Mac Mini boom. And I'm not actually sure how important access to the local file system is to most consumers. Like, when I think about what's, like, most of the data in an average internet user's life is mirrored in the cloud. I think they care about their camera roll. They care about their email, their messages, and almost everything's in the cloud. I've noticed this when I move from one computer to the next or I move from a phone. I'm like, wait, I didn't actually β there was a time when it was like, oh, you're moving computers, like, get an external hard drive. Like, make sure you drag all your folder, all your files over. Most of the stuff's mirrored to iCloud that can be accessed via an API. It requires a business development deal probably, but it does seem feasible. And a lot of the LLMs have hooks into Gmail already. I think all three major LLM apps have Gmail integrations already and more integrations are coming clearly. And so I'm not sure that you need to replicate OpenClaw and have it running on a dedicated piece of hardware or even, like, cloud hosted. But I do think people will want to be able to fire off something that writes tons of lines of code to solve a particular problem. Even if it's something as mundane as, like, getting you a restaurant reservation at a place that doesn't have a, you know, an API. Like, if there's a restaurant that just has a web form and you basically want to deploy, like, agent mode, that might look like writing a web scraper and writing something that actually does, like, a headless Chromium browser and, like, clicks it, and that might be generated from something that looks a lot more like OpenClaw or Claude Code than something that is just a couple lines of Python in a reasoning model. Anyway, um, there are also a bunch of nice-to-haves. These aren't really on the list, but, uh, you know, these apps, they still occasionally fail to return results when you're in areas with patchy cell phone service. There's, like, little UI things. Some of them botch text-to-speech requests when you'll fire off a deep research report and they'll be like, "Read this to me," and then it'll read for, like, a minute and then it just stops. Some of the apps don't let you listen to the deep research reports but they let you listen to the normal reports. So there's all these, like, little fine details in the UI that I think are causing more churn and people can just chop away at. Um, it's unclear if what is required to make an amazing product is just A/B testing all of these things and just optimizing or is it taste? I have no idea. But if you wind up doing this β this is my recommendation for anyone who's working on this stuff. If you're just going to run an A/B test to figure out what is the correct user interface and you run the A/B test, you find out that the button should be blue instead of green. Don't tell your boss you ran the A/B test. Tell them it was taste. Say that it's all about taste. Good call. And that you have taste.
because then you'll have a job forever.
But if you say, "I'm just the guy who runs A/B tests really well."
It's true. The AI models can't taste. They can't taste. They can't taste A5 Wagyu. They can't taste a Cabernet Sauvignon. Only you can do that. So make that dinner reservation and enjoy a nice glass of red wine because the models can't. They just can't. There's just no way. There's no way.
Anyway, uh, let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI. Own the data platform that powers it. And let me also tell you about Lambda. Lambda is the super intelligent cloud building the AI supercomputers for training and inference that scale from one GPU to hundreds of thousands.
Uh, Robinhood says, "Historically, investing in private markets was limited to institutions and the elite, but not anymore. With Robinhood Ventures, you can now get exposure to private companies like the ones listed below." They have a new fund that has Databricks, Mercur, Revolute, Airwallex, Boom Supersonic, Ramp, Aura, and Stripe, which is signed and pending close.
Very curious which of these companies, if any, were actually on board and excited about being part of this lineup.
I think Ramp was. I saw Fax Herbert from Ramp posting about it. Uh, and he is
But that doesn't mean the company β okay.
He said, "We are excited to partner with Sarah Shiv Chan and the broader Robinhood Ventures team on their inaugural fund. On a personal note, I'm relieved to finally have an answer for family and friends who have been asking how do I get exposure to Ramp equity?" And so, you know, if this is coming out from your head of investor relations, it's not exactly a Mac Rim-style response. So I think most of the companies that are in the press release at least and saying, "Hey, you can use our logos, we're cool, we'll see where it goes." There are folks that might get funneled in there and they don't want to be, and there might be a whole bunch of different debates and back-and-forths. Um, what is on
Sheil shared kind of some of the cost basis from the prospectus.
Uh, they bought Databricks at $150 per share, now trading at 204. Ramp at 90, now trading at 98.
Uh, Airwallex $21. It's now trading at $18.8.
And then Mercur at 714, now trading. So already seen a little uptick. Uh, Anchor came in and was sharing some of his insights for end of the show. He says, "A single closed-end fund that gives you exposure to some of the top private startups." My thoughts: people want access to private markets. Of course, so much wealth creation in America happens in startups and people desperately want access. You can see this with the insane, silly fees people are paying for Anthropic SPVs and OpenAI SPVs.
He says too, the structure of this fund is broken. And as a closed-end fund, the price here can diverge very significantly from the net asset value
of the underlying assets. With FOMO from access, this could easily trade at a very high multiple
to NAV, leading to a lot of retail investors getting their face ripped off. Uh, it ends up being less of a venture fund versus a speculative product to ride private market sentiment.
Disclosure: long Robinhood. But well, not long, but he's like, I don't like.
Yeah. So we actually have the founder of Destiny coming on the show today.
Sohail Prasad, who's coming on at 1:00 p.m.
And they're sharing their Q4 results. They have exposure to Anthropic, Chaos Industries, Hermes, positioning Destiny as a New York Stock Exchange-listed vehicle democratizing retail access to high growth.
Yeah, Destiny has suffered from the same problem. They were super early. Uh, they got this fund out almost two years ago exactly, or close.
And immediately it spiked right. There's a lot of demand to get exposure to these assets, and it's sort of come back down to earth since then. But excited to get the update from him and understand.
And let's pull up the rest of the Linear lineup to show you who's coming on the show today because we got Blake Dodge from Pirate Wires, Fredrik deBoer from Substack, Sohail as we mentioned from Destiny, Travis from Mesh Optical, and then Evan Spiegel, the co-founder and CEO of Snap. Linear, of course, is the system for modern software development. 70% of enterprise workspaces on Linear are using agents. Uh, moving on, uh, Elon Musk announced that xAI is moving away from traditional academic benchmarks like Humanity's Last Exam to focus Grok on maximal utility for real-world engineering and software development. He said, "Actually, I don't think HLE is a great measure of usefulness. We're moving away from these benchmarks." So,
Andy Scott says, "So, it's bad."
Uh, I think it's totally fair to just focus on real-world utility. Uh, but of course, people are still going to ask, well, I still want to know how it does.
Yeah, it's interesting. I mean, Tyler, give us the update on 4.2 that came out today. So Grok 4 has already been out. This is a minor revision
4.1. So now we're at 4.2, and is it focused on benchmarks or have they carved out a particular niche yet?
Yeah. So I think historically, especially when Grok 4 came out, people were, like, very, very quick to say it was, like, oh, this is so benchmarked or whatever. Um, I think they've definitely retreated from that, like, at least path with 4.2. It doesn't look like outrageously benchmarked or anything. They did this kind of interesting thing where, um, when you β so it's still not, like, fully out. It's still, like, in beta if you go on the Grok interface. Um, they did this kind of interesting thing where there's, like, four agents. Okay. Every time you actually do a prompt, there's, like, four agents, and the agents specifically have, like, distinct roles. Okay. Where it's almost kind of like, um, you have four instances of the same model but they have different system prompts.
Yeah. So you can try to get, like, okay, this one is, like, focused on doing, like, qualitative things
instead of mixture of experts, mixture of agents.
But you know, mixture of experts is, like, a β that's, like, in the architecture of the model, where this is, like, you train the model, then you kind of add this as almost, like, a harness-type thing.
Um, so it's kind of an interesting path. Um, we'll see. Like, yeah, again, this is still not, like, the actual 4.2 full release.
Yeah. I wonder what the bull case is here for xAI. There's a world where they carve out some sort of niche. You know, Anthropic's, like, focused on coding very specifically and had some major gains there. Uh, what else is there to
I mean, I think with Macro Hard, um, they're going very hard on computer use.
Okay. Computer use. Yeah. See, that would be an interesting thing where they could, like, jump to the front of that, and if that's the important technology for a couple months, that could be really good vibes.
Um, also it is interesting to think about, uh, with the Cerebras news and with the value of, like, high-speed inference on one β the whole model on one chip. Uh, is that something that Tesla's chip team can iterate towards on a faster time horizon than other chip companies? I don't really know, but, uh, they β I mean, they do custom silicon and they've done it for a long time, and they got an entire self-driving model that runs on a car. So, you know, they have some experience there. Um, and they obviously design and fab β or they don't fab it themselves, but they design it themselves. And so, uh, we'll be interesting to see how they carve that out. Let me tell you about Vanta. Automate compliance and security. Vanta is the leading AI trust management platform. Let me tell you about AppLovin. Profitable advertising made easy with Axon.ai. Get access to over 1 billion daily active users and grow your business today. Tariq says, "I'm proud to share that Humane has invested three billion into xAI's Series E round just prior to its historic acquisition by SpaceX. Through this transaction, Humane became a significant minority shareholder in xAI. The investment builds on our previously announced 500-megawatt AI infrastructure partnership with xAI in Saudi Arabia, reinforcing Humane's role as a strategic development partner." So, yeah, interesting. Uh, maybe, you know, would have wanted to get this out before the SpaceX acquisition, but better late than never.
Wait, they said they got in before the acquisition.
But, like, you know, this round got announced a while ago.
So maybe they're coming out with this news today.
Yeah. But they're saying, "Hey, we got in before the acquisition, so we got SpaceX shares." I don't know. It is odd that better late than never on β yeah. You mean on, like, a comms front, but, like, from a financial perspective, like, that was the right time to invest, right?
I think that's what's going on. I wonder why the announcement was delayed. Maybe it's, like, regulatory approval.
Because it's an international investment.
Um, let's play this clip from Jeff Bezos. Uh, his space company Blue Origin will move heaven and earth to get to the moon before rival SpaceX. The CEO Dave Limp said
recently Jeff Bezos, who never tweets, this was his first tweet of 2026, uh, posted a photo of this, like, black tortoise, um,
which goes along with the blue motif of
slow and ferocious, methodical.
But a lot of people have viewed it as a warning shot to Elon Musk, which really was focused on SpaceX going to Mars, and now he's saying we're going to focus on the moon. What do you make of that tweet, and what is the competition right now? Do you think you're going to be the first?
Well, it gives me an opportunity to put on a T-shirt for you. So, there you go. That's that.
Nothing else. Let me do that.
Uh, do I get to keep this?
Yeah, that's all yours. And that's the first one off the presses, too, by the way. I think everybody's going to want one of those. T-shirt have to lose for Blue to succeed. What the US needs is it needs two SpaceXs. It needs two launch companies that are competing vigorously against each other to try to give us the most capabilities as a country, uh, commercially, civically, from a defense perspective. Uh, because our adversaries aren't standing still, and so we need to be moving very quickly.
Healthy competition. But I think a lot of people read into that as the tortoise being Blue Origin and the hare being Elon Musk and SpaceX because it also comes after Secretary Duffy had said that SpaceX is behind. So they were opening up for everyone in terms of Artemis, and Jared Isaacman, who's now the administrator, also said essentially, yeah, whoever can get there first is going to get the contracts. So do you think you're going to get there first?
I think if asked, we will make it. We'll give it a run for our money. I like our architecture. I like our odds of getting there very quickly. I don't have a crystal ball into what SpaceX is doing. I think again, Gwynne and Elon are competent and they show it every day by launching rockets. But I love the fact that the US would compete us against each other. They are for sustainability on lunar. We're talking about who could get there in 2028. Uh, if asked, we will step up and we will move heaven and earth to get to the moon first.
Powerful line. First, uh, the moon race is going to be fun. I think it's shaping up well. I mean, yeah, a little bit of a come-tortoise-in-the-hare story. A little bit of come-from-behind. I'm not buying the tortoise as ferocious.
Yeah. I don't really love the β I don't really love the analog. Like, I don't, I don't, I don't
I don't think it's the best comms strategy. Like, I like the vague posting out of Jeff. It gets the people going, but
at the same time, just imagining SpaceX as a hare just, like, running a bunch of laps around the tortoise, just kind of
They need to take this way further. Elon needs to wear tortoise-shell glasses. Be like, "I turned your tortoise into my glasses." And Bezos needs to start carrying a rabbit's foot for good luck. That would be the hare, like, "I got your foot," you know? I want much more battles here. This is great. Uh, well, let me tell you about Okta first.
Sorry. Okta helps you assign every AI agent a trusted identity. So you get the power of AI without the risk. Secure every agent, secure any agent.
According to Blue Origin, will Blue Origin land on the moon before SpaceX? So if Blue Origin lands an uncrewed
moonlander on the moon before SpaceX before January 30, 2030, the market will resolve yes. So currently it's at a 70%
Of course, like, the race isn't over. Like, the finish line is not get one lander to the moon. It's, like, develop an economy on the moon and get lots of people there. So, uh, you know, it was just one read on this market, but it is interesting. And certainly, I mean, you can see the market was not pricing this a year ago, and I don't think anyone was. I think everyone thought that Blue Origin was kind of just a side project that was sort of just, like, doing space tourism, and now it seems like they might be going to the moon, which is pretty cool. Um, we have some breaking news.
Claude OAuth is officially not allowed in OpenClaw. So, Anthropic is responding to the OpenClaw OpenAI news, and Andrew Warner shares that this would be a great time for Sam Altman to step in and let us use OpenAI subscriptions with OpenClaw. So, in the Claude Code docs, OAuth and OAuth authentication, which is used with the free, Pro, and Max plans, is intended exclusively for claude.ai. Using OAuth tokens obtained through Claude free, Pro, or Max accounts in any other product, tool, or service, including the Agent SDK, is not permitted and constitutes a violation of the consumer terms of service. So if you're on the consumer plan β if you're on the consumer plan with Anthropic Claude, like, you just signed up for a normal plan on your app, and then you get excited, you want to set up OpenClaw on your Mac Mini. You do that, and then when you're in the login flow, you say, "Hey, I'd like to use my Claude tokens over here," it's going to say, "No, you got to set up an enterprise plan. You got to set up a proper API." Correct?
Yeah, this is not news, though. This was, like, a couple weeks ago, I think. Like, a week after OpenClaw got, like, super big, they stopped because you can still β I mean, I'm pretty sure you can still use an API key.
An API key, and will that use your
Right now, consumer β so how it used to be was you could have a Claude subscription
Um, and then with that, you get a certain amount of, like, basically Claude Code tokens,
but they're super β they're, like, massively subsidized versus the API. It's, like, 10x.
Uh, for the Claude Code tokens. So they were basically using those in
Other agents. Yeah. That was for OpenCode.
Uh, not actually OpenClaw.
Sorry, I'm getting β yeah. No, I know.
But I think it's a similar
25 different names. Yeah.
Yeah. So the chat is saying that this is news that, like, the particular OpenClaw integration maybe broke today. Uh, Peter from OpenClaw has responded and says that OpenAI has already publicly said that OpenAI subscriptions will work and continue to work in OpenClaw. And so, uh, it's a little odd because, like, yeah, I mean, you can just use the API. That's not a problem. But for the sort of pseudo-technical folks who are setting up OpenClaw instances on their Mac minis, they might be a lot more encouraged to set up the system if they're able to just log in with OAuth, um, with their Claude accounts because they're like, "Yeah, I already have the app and I use the app and I have some extra tokens. Why don't I use them over here?" Yeah, Thomas says the news is that they are applying it to the SDK.
Yes. So, there we go. Um, anyway,
uh, let me tell you about Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets.
The fossil fuel tycoon teaming up with the Rockefellers to fight energy poverty. I'm sure the online conspiracy community will love this one. Uh, but we love tycoon. We were trying to bring the word tycoon back. So, we're happy to see the journal using this. EQT chief executive Toby Rice is starting a nonprofit to tackle a lack of access to modern energy infrastructure in poor countries.
Toby Rice made his fortune unlocking a gusher of natural gas in Appalachia. He has a bold new ambition: bringing energy to millions of people in impoverished nations. Rice, the chief executive of EQT, one of the largest natural gas producers in the US. He's a co-founder of Energy Core, a nonprofit that helps developing nations such as Ghana, Zambia, and Burundi build out their energy infrastructure and prosper. Unlike other philanthropic initiatives that emphasize renewables to energize impoverished societies, Energy Core sees a role for a broader spectrum of solutions, from fossil fuels to solar panels and nuclear plants. Notably, this approach has been endorsed by the Rockefeller Foundation, one of the oldest and richest foundations.
They really opened up the floodgates with this. The Rockefellers, you know, wasn't John D. Rockefeller the richest person in human history? You see how much he's putting in this project? 200 G's. 200K. Go solve it. Go solve energy globally. 200K. Here you go.
Best I can do is 200 bucks.
I got you. I'm super excited about this.
I think Macronone deserves a victory lap at this point.
I mean, his Macronone size is looking
Yeah. Its size compared to this. No. No. Obviously, they have a lot of other donors. The Rockefellers are just a fancy name. Uh, because Toby and his wife have personally contributed $3 million, and the initiative is raising 10 million this year from energy companies, family offices, and private individuals. Um, and from his perch up at Pittsburgh-based EQT, a company with a market cap of 36 billion, Toby Rice has preached the benefits of selling more American natural gas across the globe to reduce emissions and strengthen security of the US and its allies. Now he's wading into a debate: should impoverished societies be encouraged to rely on polluting fossil fuels to improve their fortunes or leapfrog to intermittent renewables? There was this question about, um, should Brazil be allowed to clearcut the Amazon rainforest to pull forward industrialization. It's the world's lungs. Everyone suffers if that happens, but they would certainly benefit in the short term. Um, so there's a hot debate here, and he is engaging in it. Anyway, let me tell you about Cisco. Critical infrastructure for the AI era. Unlock seamless real-time experiences and new value with Cisco.
David Holz has hit the timeline. He says five million humanoid robots working 24/7 can build Manhattan in six months. Now just imagine what the world looks like when we have 10 billion of them by 2045. Now imagine the year 2100.
Dyson sphere. Dyson sphere by 2100 is the correct, like, debate. Like, is it before? Is it after? But it's, like, around there.
I keep going back to my land thesis.
It's, like, when armies of robots can build anything at any time, what is actually scarce in this case? I think with 10 billion of them, I don't even think land will be scarce anymore.
It's, like, hey, we're going to build an island.
We're going to build another moon. We're building the moon.
Just build another Earth and just throw it on the other side of the solar system.
Yeah. Yeah. I mean, it's, you know, right now we're talking about what businesses are unsloppable. The next meta will obviously be unclankable.
Unclankable businesses. What's unclankable?
What's actually unclankable when you send, you know, an army β figure out what's unstoppable. Figure out what's unclankable and then go invest in it on Public.com. Investing for those who take it seriously. Stocks, options, bonds, crypto, treasuries, and more with great customer service. Richard says, "SF guy eating a delicious blueberry. In 18 months, everything will be blueberry."
This is a perfect contrast to the other post. Just the hot dog discourse. No, no, no. David Holz. David Holz is, like, he's β because David's seen humanoid robots. Like, he's lived in SF and been around this stuff. Like, he's a true believer, and he's sort of saying, like, I've seen what they can do, and I understand the exponential here, and now imagine 10 billion of them in a hundred years. Like, it's going to be crazy. And then you have Richard on the other side: everything will be blueberries.
Uh, I thought you were talking about the delicious tacos post. He said, "I'm the CEO of a hot dog company. I've worked on hot dogs for 10 years, and I wasn't prepared for what I've just seen. Your life is about to change." So, what can you do? Buy as many hot dogs as you can. Buy stock in hot dog companies.
It's a good idea. I am long hot dog. I like hot dogs.
Everyone loves a hot dog. Joys. It's all American. There's nothing better than a hot dog at a ball game.
Except for Fin.ai. That's better than a hot dog. It's the number one AI agent for customer service. If you want AI to handle your customer support, go to Fin.ai. Yeah,
and fundraising shows defense tech is still red hot. Pretty crazy. Katie Roof, uh, one of the scoop athletes,
Scoop, there it is, scoop.
Uh, she says, "In case you missed it on Friday, we broke the news that Anduril is in talks to double its valuation to around 60 billion in a new funding round." So if you were buying triple-layered SPVs into Anduril at
45, you might make it, assuming you didn't pay three levels of 10% one time
and assuming that the guy you bought them from scammed and now in custody of the feds.
The round is notable for more than just its price. While Anduril technically has both A and I in its name, it's not the AI-centric type of startup that typically gets all the investor attention in this current cycle.
Very unstoppable, right? You're not going to vibe code a drone. You're not going to vibe code.
Yeah. I don't know. I think when you think about who's going to unlock the potential of AI for the government, you think of Palantir, you think of Anduril.
No, no. I just mean in terms of, like, AI disruption. Like, it's not something that you can vibe code a Fury drone. Like, that takes a lot of hardware, a lot of testing. You got to blow a bunch of stuff up. You need a test range. You need government contracts.
I just don't β yeah, relationships built over decades.
I think all these defense-oriented businesses, even if they are building software, are quite a bit more insulated just because of the trust factor. And if Anduril sells a product for one price and you have a small team coming together saying, you know, we're 10 people, we can build you the same thing
for half the cost, there's not quite as much pricing pressure because
the government wants reliability. They want to set something up and use it for a really long time. They don't want to really take risks,
Defense tech's on a tear. Shield AI, a drone business that can also tap the AI interest thanks to its autonomous software, is in talks to raise a 12 billion valuation, Bloomberg reported. And several other younger startups will likely raise money in the next few months. Paul Kwan, managing director at General Catalyst, said that part of the reason the firm is so optimistic about defense tech is because there are very few trillion-dollar markets that are critical for global resilience that are dominated by legacy vendors and which are experiencing both tech and geopolitical transformation. Yeah, the number of companies that fall in that bucket is pretty small. Uh, General Catalyst has invested in Anduril as well as other defense-related businesses such as Seronic and Helsing, a European rival to Anduril. Uh, as the world unfortunately braces for more wars, increased government spending has led to high-priced contracts for defense tech. Kwan said that the US department is realizing that defense tech is critical for deterrence. Kwan said he has also seen a shift among entrepreneurs believing that many of the most talented founders are choosing to build for the defense industrial base. And you can check out the rest of the story on The Information.
Uh, a lot of attention's been focused on OpenRouter. If you go on OpenRouter and look at the rankings, you can see that Chinese open-source models are completely dominating. Yes. Charts. MiniMax
DHH talking about, uh, Kimi K2 is now a daily driver for squashing bugs at 37signals. Very interesting data point. Uh, since a lot of this can be β I think the OpenRouter stuff can be a little hard to contextualize because there's some amount of volume that doesn't get captured in OpenRouter. Obviously
it's the majority of the volume is not captured.
