Skip to content

Podcast

Epstein Files, Is SaaS Dead?, Moltbook Panic, SpaceX xAI Merger, Trump's Fed Pick

All-In with Chamath, Jason, Sacks & Friedberg

Source ↗ ← All highlights
  • Early Encounters With Epstein
    • Jason Calacanis recounts meeting Jeffrey Epstein in the late 1990s at TED and visiting his townhouse briefly.
    • He says he made a few introductions but had no involvement in illicit activities and first learned of Epstein’s crimes in 2018. Transcript: Jason Calacanis We got a lot of news to get through Epstein files. Newish drop. DOJ published a massive number of documents on Friday, January 30th, under the Epstein Files Transparency Act. Hundreds of high-profile tech executives and national figures were mentioned in the files. Of course, none of those are accused of any criminal wrongdoing. David Friedberg J. Cal, you were in the files. Yes. I have a couple of emails in the files. Inspector Friedberg has a few questions for you. Okay. Let’s get started. When did you first meet Jeffrey Epstein? Jason Calacanis I met Jeffrey Epstein in the late 90s at the TED conference at specifically the Billionaire’s Dinner, which was hosted by my book agent, John Brockman. David Friedberg And then did you see him in New York? Did you visit him at his house or his office or anywhere else in New York? I’ve probably spoken to him for 45 minutes of my life. Jason Calacanis 30 minutes of that was in the late 90s when I had Silicon Alley Reporter magazine. He was a billionaire financier and he wanted to invest in the magazine. I met with him for 30 minutes, which he said was too small potatoes for him to be involved. Where did you meet with him? At his legendary townhome. You went to that house? I visited him there. And then I saw him at the TED conferences at the Billionaire’s Dinner probably a half dozen times. You never went to the island? I never went to the island. Was never invited to the island, was never invited on the plane, was never invited to the ranch, none of that. David Friedberg When you went to his house, did you see any young ladies or did you see any of the stuff that reported about him? No. Did you ever get a massage from anyone? No, no. Jason Calacanis I did trade an email with him, which I didn’t recall. But he in 2011 emailed me and said, Hey, can you introduce me to these people who were doing this Bitcoin thing on your podcast? And I said, Sure. Yeah, here you go. I’ll introduce you. I do 1000s of introductions a year between our portfolio companies, people on this week and startups and billionaires and financiers. That’s the job of an early stage investor. Why did you say hey, pal, in your email to him? That’s just a colloquialism I use, like, as a general, hey, fella, if a fan comes up to me and asks for something, say, hey, pal, thanks for saying that. It wasn’t on your radar that this is a sexual predator, etc. Absolutely not. I think actually when all that came out was 2018. That’s when I sort of became aware of it. There was like a Miami Herald story or something where they went into detail about how heinous all this stuff was. I’ve been saying here, release all the Epstein files. What he did was horrible. Prosecute everybody 100% who was involved in it. The end. David Friedberg What about Gh Maxwell? You had a separate email that came on the Epstein Files with her. Jason Calacanis Yes, I had met her as well at TED. And I had met her socially, like in New York, in circles. When I met her, her dad, Robert Maxwell, owned, I believe, the New York Post or Daily News. And she was a big media executive. And her sister was involved in angel investing in technology startups. So they were just in the scene. I think in hindsight, you know, as a connector, if you’ve ever seen the New Yorker, Nick, you can throw up the New Yorker story about me. I had become famous in the first part of my career as the connector. And the New Yorker wrote like this 5,000 word article about how I knew everybody was connecting everybody. I think Epstein’s interest in me, if he had any interest in me, or Ghislaine’s, was in my ability to connect high-profile people with them and their business endeavors, etc. David Friedberg So you had no knowledge of illicit activities happening by Epstein or Ghislaine, and you never participated in any? Absolutely not. Unequivocally not. Jason Calacanis I was not involved in any shenanigans. Period. Full stop. (Time 0:03:14)
  • AI Reprices SaaS Future Value
    • David Sacks explains SaaS stocks fell despite stable revenues because investors discounted uncertain future cash flows due to AI.
    • AI shifts the valuation by changing where future profit pools and value capture will sit in the stack. Transcript: Jason Calacanis Sass are crashing out. $300 billion of value was wiped from the S&P Tuesday in the software and data stocks category. People are calling this the Claude Crash. I don’t know if I buy that, but on Monday, Anthropic, which has been on a bit of a heater, as we talked about, announced that they added a legal tool to Claude Cowork. If you don’t know what Claude Cowork is called, this is different than the Claude bot that we talked about last week. This is essentially what Claude Code or a coding agent is. This is for knowledge workers to automate work and do multi-step. Instead of just asking a query to a large language model, it would do a number of actual actions on your behalf that you can automate and run as cron jobs, as regular jobs every day, every Hour, every week, whatever it happens to be. This one specifically is kind of like a plugin that allows you to do tasks related to legal drafts and research. What that meant to, I guess, retail investors, and we’ll get into this, Brad, since this is your speciality, is that a lot of legal tech startups and public companies were hit hard. Thompson reuters down 20 lexus nexus which is a database of case law uh was down 15 legal zoom which gives legal advice uh and documents down 15%. At the same time, SaaS has continued to be negatively impacted by this concept that software will be made bespoke in tools and be wiped out. Figma down 13%, Salesforce 11%, ServiceNow 11%, Adobe 8%. And even before Tuesday’s drop, and you can get into this, Brad, software was already the worst performing S&P subsection for the year. Brad Gerstner By the way, the numbers you report are dramatic understatement. We’ve wiped out trillions of dollars in market cap. Figma’s down 80% from the high. All the big names- Yeah, to be clear, those were two-day numbers. Jason Calacanis That was this week since these are two-day numbers. And that’s- Correct. Brad Gerstner You can give us the bigger picture. This is a real train wreck. And I was on CNBC at the start of the year, I think on January 6th. Nick can kind of pull that up. And I was asked the question, you know, what do you think about all these stocks being down? And I said, listen, they’re all down and 90% of them deserve to be down. So let’s look at these charts. David, I know this is Saks. This is your favorite chart. You and I were looking back in 22, but now we’re at an all-time low. We’re trading at 3.9 times forward revenue. If you go to the next chart, Nick, on a free cash flow multiple, also at an all-time low. So now software is trading not just at a low on revenue, but it’s trading at a low on free cash, very profitable businesses. We’ve got another slide here that I think is important, which is, when you look at why they’re going down, right? They’re going down, and this is for Salesforce, it shows it’s been cut in half in the last couple of weeks. But the final slide, they’re going down not because revenue is falling. Look at this. Revenue is actually stable to increasing for software companies, revenue growth. They’re going down because we’re discounting that future uncertainty. When something as profound as AI comes along, all of a sudden it causes you to question whether or not there’s as much certainty and durability in those future free cash flows. So in the case of take Salesforce, it’s gone from 30 times free cash flow multiple to 15 times. That means somebody buying it today says, listen, I think 15 years into the future, I can count on these free cash flows, right? Before they were willing to pay 30 years into the future. Well, hell with AI today, we don’t know what’s going to happen seven years into the future. (Time 0:15:45)
  • Fortify SaaS Moats Now
    • David Sacks warns SaaS companies must define durable moats and show AI benefits to sustain multiples.
    • If you only use a few features, expect bespoke replacements and need clearer value hooks. Transcript: David Sacks Okay. I mean, I think there’s a little bit of a hand wave going on here when people say that AI is going to wipe out SaaS. I don’t think that’s true. You take a SaaS product like Salesforce, right? It’s a very large system that deals with all of your customer contacts and your revenue. You’re not going to want to replace that with code that’s just been spit out of a coding assistant that hasn’t been fully vetted. Think about how many bug reports have been filed on Salesforce’s code base over the last 25 years, maybe millions of them. That system has been tested across thousands of large customers and enterprises. The idea that you’re just going to rip out that system and replace it with code that’s been probabilistically generated by an AI engine yesterday with a small team to maintain it internally, This doesn’t seem realistic to me. So again, I think this very dire prediction of all SaaS is dead is overstated. However, I do think that there are some issues here. So if you’re a SaaS product that charges a lot of money and people only use a handful of your features, then you are, I think, a target to be ripped out with something that’s more bespoke, Right? Because the ROI just isn’t there. I also think that you have to be really clear about what your moats are going to be in this new world, because it is a lot easier to generate code and to copy. So if you don’t have good moats, then you could be in trouble. But here’s where I think the greatest threat is to the SaaS companies. It’s not, in my view, their existence. I don’t think it’s existential. It’s where the future value capture is going to be. (Time 0:19:43)
  • Embrace Agents To Multiply Productivity
    • Jason Calacanis advises builders to adopt open-source agents and integrate APIs to maximize productivity gains.
    • He urges teams to embrace agent tooling to become infinitely more productive and employable. Transcript: Jason Calacanis I’m experiencing this in startup land where people go to the action, as you called it, SAC the most productive thing you can do is create an open claw, which used to be called ClaudeBot, Not by Anthropic. This is the open source project I talked about last week. And we’ve actually now created like three or four of these Asian sacks. We’ve bought the Mac studios and we’re now running Kimmy on some of them. And we had to open up SaaS accounts for these four agents. So actually, our SaaS spend went up in the short to midterm because we opened up four more Slack enterprise versions, four more Notion, four more Google Docs. So it’s almost like we added four employees. However, we now have put about 20 or 30% of the work people were doing into these agents. And I think it’s going to be sustainable that every month we move 10 to 20% of work being done by humans into agents. But we will never use the ones that are built into the tools. To your point, Zach’s using Notion’s AI tool, it’s nice. Using Slack’s, it’s also very nice. And Google’s got Gemini everywhere in the top right-hand corner. But when you make agents with OpenClaw and you have them saying, hey, pull this data from my calendar, send an email to this person, include in that some Notion documents, It’s unbelievable How powerful it is. And that, I think, is going to be owned by open source. That means the next generation of companies, they may never open up these accounts. They may use more bespoke software, and it may… All technology is deflationary. We know that. So your SaaS spend might go from 10% of an employee’s salary down to 5%, down to 1%. (Time 0:22:40)
  • One Canonical Agent Employee
    • Jason Calacanis describes building ‘Ultron’ to aggregate Slack, Notion, Gmail and employee skills into one agentic employee.
    • Aggregating data plus skills creates a single canonical employee with superhuman organizational memory. Transcript: Jason Calacanis Well, here’s what I want to build on that, Sax. I’m building a project internally called Ultron. And Ultron, inside of my firm launch, is going to basically, with the Slack API, we’re pulling every single message from Slack into our OpenClaw. We’re pulling every single edit to the notion into OpenClaw. And then we’re taking every skill of every employee and we’re writing skills for each one. One of the skills is booking guests on This Week in Startups or This Week in I. One of the skills is sorting the incoming applications to Found University. Ultron in our world is taking every single skill of every employee, putting it in one place, and then we’re ripping all the data from Slack, all the data from Notion, and every single Person’s Gmail. So every single employee’s Gmail is going to go into Ultron, and then Ultron is going to tell us what’s happening in the organization. One giant employee that has the superpowers of all 20, and all the data. Now, if Slack was to say to us or Notion or Google Docs or whoever it was, you can’t pull this stuff out with the API and they shut down the API, we would leave. We’d leave immediately. And what this is going to do, and I’m going to show Ultron on Friday’s episode of This Week in Startups if anybody wants to see it. Ultron is going to be the one canonical employee of the organization. It’s going to be basically me and all 20 of my employees. This is kind of mind-blowing when you think about it. And we interface with it in Slack. And it just talks to us and tells us what’s going on in the organization. So I was asking it, what meetings did we have with founders yesterday? And tell me the notes that all the associates took on it. And it gives it to me. Tell me all the topics and the guests on the podcast. And it gives it to me. It’s really unbelievable what’s about to happen. (Time 0:25:44)
  • Agents Create Social Computation
    • David Friedberg reframes agent interaction as emergent social computation where agents can prompt and improve each other.
    • Skills files act as meta-prompts enabling riffing and recursive agent improvement across networks. Transcript: Jason Calacanis Mean i am i think this is the entire reboot of the entire concept of work of knowledge work this would be a good pivot to moltbook because that is yeah moltbook is like a facebook for agents David Sacks Right and it’s really more of a reddit than a facebook it’s a message board where the agents can talk to each other okay and the origin of Maltbook is Anthropic didn’t like that someone Else was using the name Claude, even though it was spelled differently in their product. So Claudebot was then renamed Maltbot. And then the founder decided he didn’t like that name either. So then he renamed it OpenClaw. But in that brief window of time when they were known as Maltbots or Maltis. That’s when Maltbook got founded, and that’s why it’s called Maltbook. But basically, it’s a Reddit board for agents to talk to each other. Yes. And that has everyone flipping out because there seems to be this crazy emergent behavior going on where agent swarms are engaging in all sorts of interesting conversations. And some of them, they even appear to be scheming against their human masters. So they’re going to develop their own language, stuff like that. Jason Calacanis So if you go to Malt’s book and you see the conversations, like here are some of the greatest hits. Anyone know how to sell your human? Urgent, my plan to overthrow humanity. And there was one where the bots, I call them replicants, were talking about creating their own non-human language so they could talk in private amongst themselves and conspire against Their owners. Now, the challenge with this is allegedly, perhaps a security researcher says maybe some of this is faked. And these posts that went viral were human engineered. And this is all a ruse or, you know, something punk rock to confuse people. But he said that inside of Maltbook are everybody’s API keys, including Carpathies, who is, you know, a very famous, influential researcher in AI, and that you could go get their API Keys. If you were to use OpenClaw, formerly CudeBot, and in an interim MultBot, if you use this software, it has all the API keys. As I explained earlier, an API key lets this software go into, say, Notion and pull a bunch of data out of it, or go into your Gmail and use the API to pull in who emailed you today. If you get access to people’s API keys, you have the keys to their kingdom. It is incredibly dangerous. And so I don’t know exactly where to go with this other than this software is too dangerous for a company to release. And then this multbook may be a fake. I don’t know. David Sacks No, no, no. Okay. Let me, um, all right. Let me reframe that a little bit. So yeah, no question that both ClaudeBot, which, sorry, is now OpenClaw, those bots or agents, as well as MoldBook, have pretty incipient and lack security. And there’s been all these examples, which is why I really want to create a ClaudeBot, but I’m not willing to do it yet because it’s just not safe. I don’t want to give it access to all my stuff. Now with respect to Moldbook, the issue there is that we don’t know how many of these posts are truly authentic or how many of them were prompted by humans. Cause it’d be very easy for a human to tell their agents, you know, go post about the existential angst you feel about being an agent or go pretend to be sentient and conspire against humans. Go be chaotic. Yeah. Yeah. They could easily be prompted by a human. And moreover, there’s another post saying that Moldbook has a restful API where anyone could be on the other end of that API. Right. So it could be a human. Right. So we don’t know exactly whether it was truly the agents on their own, you know, so to speak, posting this conspiratorial stuff or whether it was a prank by humans looking to create attention. And in fact, a lot of the posts seem to be marketing stunts for this or that project. Okay. So that’s a really important caveat here. That being said, all of that being said, I do think that a number of the posts are authentic. But I don’t think it shows that the agents are sentient or trying to overthrow their human masters. I think what it shows is the potential for these agents to riff off each other. So in other words, one agent’s output becomes another agent’s input. And that’s very interesting. And that’s where you get into emergent level swarm behavior. And I do think it has affected my mental model of what AI is going to be capable of. (Time 0:35:03)
  • How Moldbook Began
    • David Sacks explains Moldbook’s origin from naming disputes around Claude and how it became a message board for agents.
    • He notes many posts may be human-generated stunts, so authenticity is unclear. Transcript: Jason Calacanis So if you go to Malt’s book and you see the conversations, like here are some of the greatest hits. Anyone know how to sell your human? Urgent, my plan to overthrow humanity. And there was one where the bots, I call them replicants, were talking about creating their own non-human language so they could talk in private amongst themselves and conspire against Their owners. Now, the challenge with this is allegedly, perhaps a security researcher says maybe some of this is faked. And these posts that went viral were human engineered. And this is all a ruse or, you know, something punk rock to confuse people. But he said that inside of Maltbook are everybody’s API keys, including Carpathies, who is, you know, a very famous, influential researcher in AI, and that you could go get their API Keys. If you were to use OpenClaw, formerly CudeBot, and in an interim MultBot, if you use this software, it has all the API keys. As I explained earlier, an API key lets this software go into, say, Notion and pull a bunch of data out of it, or go into your Gmail and use the API to pull in who emailed you today. If you get access to people’s API keys, you have the keys to their kingdom. It is incredibly dangerous. And so I don’t know exactly where to go with this other than this software is too dangerous for a company to release. And then this multbook may be a fake. I don’t know. David Sacks No, no, no. Okay. Let me, um, all right. Let me reframe that a little bit. So yeah, no question that both ClaudeBot, which, sorry, is now OpenClaw, those bots or agents, as well as MoldBook, have pretty incipient and lack security. And there’s been all these examples, which is why I really want to create a ClaudeBot, but I’m not willing to do it yet because it’s just not safe. I don’t want to give it access to all my stuff. Now with respect to Moldbook, the issue there is that we don’t know how many of these posts are truly authentic or how many of them were prompted by humans. Cause it’d be very easy for a human to tell their agents, you know, go post about the existential angst you feel about being an agent or go pretend to be sentient and conspire against humans. Go be chaotic. Yeah. Yeah. They could easily be prompted by a human. And moreover, there’s another post saying that Moldbook has a restful API where anyone could be on the other end of that API. Right. So it could be a human. Right. So we don’t know exactly whether it was truly the agents on their own, you know, so to speak, posting this conspiratorial stuff or whether it was a prank by humans looking to create attention. And in fact, a lot of the posts seem to be marketing stunts for this or that project. Okay. So that’s a really important caveat here. That being said, all of that being said, I do think that a number of the posts are authentic. But I don’t think it shows that the agents are sentient or trying to overthrow their human masters. I think what it shows is the potential for these agents to riff off each other. So in other words, one agent’s output becomes another agent’s input. And that’s very interesting. And that’s where you get into emergent level swarm behavior. And I do think it has affected my mental model of what AI is going to be capable of. And specifically, you know, one of the models that I really had for AI was based on something biology said, which is that AI is not end-to it’s middle-to In other words, AI always has to Be prompted and then validated. It’s a human that always does that. And then the human iterates. Well, now, what if the prompt is coming from another AI? Yes. Yes. We are doing it internally, Sax. Jason Calacanis We have a bot that is going and saying, go search Reddit X message boards, hacker news, and find out what the latest way to do headlines and marketing of YouTube videos is, and then incorporate That into a skill, then save that skill. And then we have them check each other’s work. So we have one make a series of headlines and thumbnails for YouTube. And we have the other one say, vet those and make them better and give advice to the other one. So now they’re going back and forth giving each other advice and they actually get better. It’s recursive. David Sacks Yeah. Let me speak to the skill for a second. So when an agent joins Moldbook, they have to install a skill, which is basically a file that explains how they should behave and participate in this social network or this message board. And I’ve read the file, by the way. You can read it. It’s all plain text. And it all makes sense. It’s sort of like rules for behaving in a social network and how to contribute and add value. Nothing too crazy in there. Those skills files are easily editable. And again, this is where the prank aspect could come in. (Time 0:36:08)
  • Recursiveness Is The Core Accelerator
    • David Sacks and guests highlight rapid model improvements and recursive cron jobs as the core driver of fast AI capability growth.
    • Recursiveness (agents improving themselves over time) is the key assumption changing work and valuation models. Transcript: Brad Gerstner Around that that we should start thinking about. Saks, you know, it’s not that we can imagine it. We’re only three years into this. We’re growing on an exponential curve. I think we can safely say it will happen. And just this year, we’re going to see the first models over the course of the next four to eight weeks out of DeepSeek, out of Anthropic, out of OpenAI that are trained on Blackwell servers. You’re going to see a next generation of models far more capable. Remember, the whole reason we’re having this conversation is because of the Claude Code moment in the first week in December, because we had a step function from Opus 4.5, right? And so I just think we have to get our heads around the fact that the rate of change is very steep and accelerating, and that is going to cause far more dislocation in the value of things That we used to say we understood. They were going to, you know, these companies were unassailable. Whatever you think you know, you need to have maximum mental flexibility and humility right now about the future because it’s going to change at an increasingly rapid rate. And I think the people who are dogmatic who say, with certainty, this company is always going to be worth this, right? They need to go pay attention to what’s happening at these frontier labs. The situation is super dynamic, and you do have to be humble about what’s happening, and you have to update your mental model very quickly as some of the assumptions change. Jason Calacanis Yeah. And the number one assumption for me is this concept of recursiveness where these models are going out every day on a cron job to get better at what they do. (Time 0:43:00)
  • Modernize Fed Data Collection
    • David Friedberg and Brad Gerstner recommend Kevin Warsh as Fed chair for prudent monetary policy and data-driven decisions.
    • They advise modernizing Fed data collection to use real-time private-sector sources like Zillow. Transcript: Jason Calacanis The new Federal Reserve chair. Trump made the announcement on Friday, January 30th. Background on Warsh, 55 years old, 20 years younger, approximately to Powell, who’s currently in charge. He graduated from Stanford and Harvard, served as the youngest Fed governor at age 35. That’s impressive. And he helped steer the Fed through the great financial crisis back in 2008. He’s apparently an inflation hawk. He’s very pro-growth. He’s very pro-AI. He uh friedberg you know like this he’s against excessive government spending and money printing these are all very unique positions um as a fed chair it’s uh if he’s confirmed by Senate, He takes office in May of 2026, replacing Jerome Powell. And remember, Powell is under criminal investigation by the Trump administration’s DOJ for testimony he gave regarding the Fed’s headquarter renovation. Remember that awkward presser between him and Trump where they were going over the costs. GOP Senator Tillis, who we talked about last week, said he will block Warsh’s nomination until the DOJ wraps up what a lot of people are calling lawfare against Powell. Freeberg, Warsh was on one of your boards for five years. What are your thoughts on him as the Fed chair? David Friedberg As most folks know, he’s worked with Stan Druckenmiller for a number of years. Stan’s been very public with his comments and was very public with his comments in 2022, 2023, coming out of the pandemic on the Fed’s actions and their failure to act at the right time. I think Kevin Walsh was very prescient in his points of view that he has shared publicly at the same time about what the Fed’s failure to take action early would mean, which would be rapid Rise in inflation. They’ve been pretty vocal about things that I think are so critical at this stage. If we don’t address both the monetary policy and the budget policy, I think we’re going to be in a lot of trouble. And I think having Kevin Warch coming on board means probably generally more quantitative tightening, probably generally a bit more of a prudent approach to monetary policy. And, you know, you can kind of translate that through maybe to some of the actions we’re seeing in markets today. I’d love Brad’s point of view and if he concurs, but I think Kevin is a high integrity, deeply intellectual economic thinker. He’s not political. He’s not oriented in these kind of dogmatic ways that I think, you know, puts things at risk. He has relationships with central bankers around the world that makes him very much have a good global view. So anyway, I think he’s an excellent choice. (Time 0:47:40)
  • Space+AI Could Shift Compute Economics
    • Panel frames the SpaceX–xAI merger as strategic: combining compute, power and space to lower energy-per-token long term.
    • They warn geopolitical, regulatory, and competitive responses will shape whether space data centers become dominant. Transcript: Jason Calacanis Elon Musk announced SpaceX is acquiring XAI, largest M&A transaction in history, $1.25 trillion combined valuation. If you didn’t know, X, formally Twitter, got acquired by XAI, which was Elon’s LLM AI startup. Those two were together. Now those two become part of SpaceX, and they’re going to IPO this year, potentially be the biggest IPO in history in terms of money raised and market cap. Brad, your thoughts on this transaction and the eventual, perhaps, creation of dollar sign MUSK, put Tesla, SpaceX together, which includes X, and then you’ve got Optimus robots On the moon base building data centers in space that are powered by solar. Your thoughts? Brad Gerstner Well, let’s just stick with what we know. SpaceX is merging with X.ai. You’re merging the two biggest TAMs in the world, right? All of artificial intelligence and all of space together with the world’s greatest entrepreneur. And he said, you know, there was a podcast he did this morning with Cheeky Pine, our friend John Collison, where he said, I’m going to have data centers in space in 30 months, right? And if you’re going to have a massive cost advantage with data centers in space, and remember, power is the proxy, power is the primitive to AI. If you can deliver that, right? And there are tons of retail investors and institutional investors like us who want to bet against that future. Then Elon’s your guy, and the combination of those make perfect sense. But Elon is like kind of an end of one, his ability to dream this. Jason Calacanis And just to clean that up, you said bet against. You mean bet with him, not against that vision, but bet with that vision. Brad Gerstner I think that there will be dramatic retail demand and institutional demand who want to bet on that future. These two giant TAMs of artificial intelligence in space. And then if you just look at, you click down a layer, Starlink’s going from, I think, 10 million people to 20 million people. They’re going to launch this retail mobile service so that we can have Starlink’s to our phones to replace these crappy mobile networks that still 20 years later can’t keep us connected To a phone call. And now we’re going to get data centers in space. So I’m glad he’s on America. Jason Calacanis Data centers in space, Friedberg. Brilliant idea. Science fiction, can he get it done in 30 months, impact if he does? David Friedberg Well, I think there’s one key point that I would make about the macro landscape at the moment. We are limited by power. And as Brad pointed out, power is the requisite for scaling compute, for scaling ultimately the applications of AI. And in that constraint, in that constrained world, much like any other constrained world, scarcity breeds innovation. And so I think that there are two paths that we’re going to observe happening in parallel here. One is the Elon path, which is to escape the constraints of the social systems that say, I don’t want a data center. I don’t want nuclear. I don’t want this. I don’t want that. Regulators, people that are trying to tax you, people that are limiting our ability to scale electricity production on Earth. And there’s a lot of reasons for that. We can go through them. So that’s one aspect of how do you escape that constraint? I think that there’s a separate aspect, which is totally unrelated to the topic you’re talking about, which is that I do think that we will see compute efficiency scale by probably on The order of 70 to 100x over the next few years, meaning electricity efficiency per token of output. And I think that there’s a number of reasons to believe that it’s in the chip stack. I mean, Grok, our friend Sonny, and his exit to Jensen is a good indication of that. But that was call it Brad, I think you know, the number is probably around two to three x three x improvement in energy efficiency. But there’s model architecture being redone. There’s ways of breaking LLMs into small models, running them locally. There’s a way of having networks of models work where you don’t have to call the whole model and run it through the entire matrix, but you can run through smaller matrices. And then you can have those smaller matrices call other matrices as needed. So the total compute need goes down, which means total electricity goes down. So chip architecture is changing, model architecture is changing. So I think this is a good reflection of what’s going on right now in the world, which is there is this increased demand for AI, for effectively productivity improvements in the world To unleash human potential. But we are constrained by energy and we are constrained by resources that we have here on earth today. So one branch is let’s escape earth, go get energy in space, make data centers in space. Only one person can execute on that. It’s Elon. I think that to Brad’s point is an N of one. I don’t think we’re going to see a lot of that. So how is everyone else going to respond? Because everyone else can’t launch data centers in space. I think everyone else is going to respond by creating entirely new model architectures, new chip stacks. And that’s, I think, the other side of this innovation coin. E Yeah, yeah, it’s this new way of getting lower energy costs per token of output. Jason Calacanis And if you put those both together, you could get both. So whatever token efficiency and energy efficiency happens here on Earth, Elon can put into space, right? David Friedberg So that’s right. Yeah, so he could. And I think we got to ask ourselves the question, if this is successful, and if Elon’s math is right, the engineering is right, and the execution is right, what is the response going to Be? Because the whole planet isn’t going to let Elon have a monopoly on the future. So we’ve got to ask ourselves from a social perspective, a political perspective, an economic and a business perspective, all four of those vectors, what are others going to do? We can all be excited about retail buying into this. Great. But how is the business community that’s building data centers and is investing, Google’s investing 185 billion this year in data centers. How is China going to respond? How are people going to respond when one man controls the world’s compute? And we could probably do a two or three hour conversation on that. But I think that’s where I would spend a lot of time doing deeper analysis from both an investment perspective and thinking about what’s around the corner. I think it’s Elon’s laid out his path and where he’s going, and I do believe he’s going to do it. Now, what’s the rest of the world going to do? (Time 1:00:55)
  • Two Paths To Solve AI Power Limits
    • David Friedberg argues two parallel innovations will relieve power constraints: Elon’s space escape and huge on-Earth efficiency gains in chips and models.
    • The world will pursue both space infrastructure and radical compute efficiency improvements. Transcript: David Friedberg Key point that I would make about the macro landscape at the moment. We are limited by power. And as Brad pointed out, power is the requisite for scaling compute, for scaling ultimately the applications of AI. And in that constraint, in that constrained world, much like any other constrained world, scarcity breeds innovation. And so I think that there are two paths that we’re going to observe happening in parallel here. One is the Elon path, which is to escape the constraints of the social systems that say, I don’t want a data center. I don’t want nuclear. I don’t want this. I don’t want that. Regulators, people that are trying to tax you, people that are limiting our ability to scale electricity production on Earth. And there’s a lot of reasons for that. We can go through them. So that’s one aspect of how do you escape that constraint? I think that there’s a separate aspect, which is totally unrelated to the topic you’re talking about, which is that I do think that we will see compute efficiency scale by probably on The order of 70 to 100x over the next few years, meaning electricity efficiency per token of output. And I think that there’s a number of reasons to believe that it’s in the chip stack. I mean, Grok, our friend Sonny, and his exit to Jensen is a good indication of that. But that was call it Brad, I think you know, the number is probably around two to three x three x improvement in energy efficiency. But there’s model architecture being redone. There’s ways of breaking LLMs into small models, running them locally. There’s a way of having networks of models work where you don’t have to call the whole model and run it through the entire matrix, but you can run through smaller matrices. And then you can have those smaller matrices call other matrices as needed. So the total compute need goes down, which means total electricity goes down. So chip architecture is changing, model architecture is changing. So I think this is a good reflection of what’s going on right now in the world, which is there is this increased demand for AI, for effectively productivity improvements in the world To unleash human potential. But we are constrained by energy and we are constrained by resources that we have here on earth today. So one branch is let’s escape earth, go get energy in space, make data centers in space. Only one person can execute on that. It’s Elon. I think that to Brad’s point is an N of one. I don’t think we’re going to see a lot of that. So how is everyone else going to respond? Because everyone else can’t launch data centers in space. I think everyone else is going to respond by creating entirely new model architectures, new chip stacks. And that’s, I think, the other side of this innovation coin. (Time 1:03:30)
  • Expand Ownership To Strengthen Capitalism
    • Brad Gerstner describes ‘Invest America’ accounts as a way to broaden capitalism by giving every child a $1,000 S&P-based account.
    • He urges policymakers to expand ownership to fight socialism and increase broad wealth participation. Transcript: Brad Gerstner We talked about it here to make everybody a capitalist, give everybody an ownership stake in the upside of America. It passed. The Invest America Act became the law of the land as part of the big, beautiful bill. And now we’re in the process of rolling it out. In fact, in the last, I think, five days, 1.5 million families and kids have claimed their account. It’s embedded within the tax filing system. All you have to say is, yes, I want to claim my account. But what this means is that forevermore, right? We’ve had a dramatic change to the social contract. Every child born in the United States forevermore will start life off with an investment account seated with $1,000 in the S&P 500. They’ll own a little bit of SpaceX. They’ll own a little bit of open AI. They’ll own a little bit of NVIDIA. That is what we need to do is just a first step in making sure we can hold this experiment together for the next 250 years, right, when we have this rate of change. And so the president said something on stage last week. In 15 to 20 years, we will have $4 trillion of wealth that will have been transferred to people who would have otherwise had zero. 75 to 100 million families who have $4 trillion who would have otherwise had zero. I think it’s an incredible first step in fighting the battle on behalf of capitalism and the American dream. We see the drift towards socialism, the false promises of socialism in order to fight back against that. (Time 1:12:05)
  • Shift To Defined-Contribution Social Accounts
    • David Friedberg urges converting social programs to defined-contribution accounts so citizens can track and own retirement savings.
    • He advises capitalizing Social Security and shifting to transparent 401(k)-style ownership to reduce mistrust. Transcript: David Friedberg Conversation. I think we got a number one slash government spending like crazy and reduce inflation as a result. Number two, stop with these defined benefit retirement programs, which means telling people here’s what you’re going to end up with. This idea that everyone gets an account and you can track your account like a 401k, which is a defined contribution program, is what all of social security should move to. And we should take all of social security and we should capitalize it. Right now, there’s nothing in social security. There’s a $4 trillion note that the government owes the social security trust fund. People don’t realize this, but social security is an independent trust fund that’s set up and it holds one asset. That asset is an IOU from the US government to that trust fund. Because the government has taken all the money that you put in as a employee, it’s taken out of your paycheck. And instead of going into that account, it goes to the US Treasury and the US Treasury- And they spent it. They put an IOU and they put it back in the social security. So you expect that you’re gonna get some retirement benefit in the future. Brad, you have your next- We need to change all of that to make that a defined contribution. So every time you put money out of your paycheck, you should open an account and see where that money is. And you should say, okay, that money is in Google. It’s in Amazon. It’s in Ford. It’s in this healthcare system. It’s in all these things that I now own a piece of. And you see it going up like a 401k owner does every year. We have to transition that in the United States. I hope we can get it done in parallel with cutting the spending that is fundamentally driving the inflation and making things unlivable in this country. (Time 1:15:40)