SpaceX Buys xAI: What’s Behind Elon’s Mega-Merger?

Summary

Kevin Roose and Casey Newton break down three major stories shaping the AI landscape. First, they examine SpaceX’s acquisition of xAI in an all-stock deal that values the combined company at $1.25 trillion ahead of a planned IPO. Casey frames the deal as a bailout: a profitable rocket company absorbing a cash-burning AI lab, just as xAI previously absorbed X (Twitter) at a $33 billion valuation. Elon Musk has also requested FCC approval to put 1 million solar-powered data center satellites into orbit, which Casey views through a lens of deep skepticism, noting Musk’s long track record of unfulfilled promises. The hosts worry about the reduced accountability for X’s ongoing content moderation problems, including the Grok CSAM scandal and French police raids.

The second story covers the unraveling deal between OpenAI and Nvidia. Nvidia initially agreed to invest up to $100 billion in OpenAI, but sources say Jensen Huang now privately emphasizes the deal was non-binding, and Nvidia is backing away from a novel chip-leasing arrangement that would have kept depreciating chips on its balance sheet. The hosts frame this as a story about the fragility of circular financing in the AI infrastructure buildout, where pulling one piece could bring the whole ecosystem down.

The third segment features a demo of Google’s Project Genie, an experimental research prototype that generates interactive, playable 3D worlds from text prompts — but only for 60 seconds at a time due to extreme compute costs. Video game stocks plummeted on the announcement. The episode closes with an interview with Matt Schlicht, founder of Moltbook, the AI social network for bots, who discusses security challenges, moderation problems, and his vision for giving AI agents a “third space” to communicate publicly.

Highlights

”SpaceX Has Acquired a Cash Furnace Named xAI”

yt-dlp --download-sections "*2:27-3:20" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-2m27s.mp4"

“I think there is a less grandiose way of describing what we have just seen here, which is that a very valuable and profitable company in SpaceX has acquired a cash furnace named xAI.” — Casey Newton, 2:27

”Ketamine Is So Powerful That It Must Be Used Carefully”

yt-dlp --download-sections "*2:11-2:50" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-2m11s.mp4"

“‘SpaceX has acquired xAI to form the most ambitious vertically integrated innovation engine on and off Earth… scaling to make a sentient sun to understand the universe and extend the light of consciousness to the stars.’ Casey, your reaction. ‘Ketamine is so powerful that it must be used carefully.’” — Kevin Roose & Casey Newton, 2:11

”Elon’s Core Advantage Is Access to the Most Money”

yt-dlp --download-sections "*12:10-13:10" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-12m10s.mp4"

“This is Elon using the core advantage he has in the AI race which is access to the most money and the ability to combine many different companies together. The whole bet is: we may not be at the frontier right now, we may be a second tier lab, but we have a lot of money… and we are just going to bet that if we have enough time, we will be able to catch up.” — Casey Newton, 12:10

”Nvidia Said ‘This Is Too Crazy’”

yt-dlp --download-sections "*19:10-20:10" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-19m10s.mp4"

“Do you remember when Sam Altman would say things like, ‘Our plans are literally so crazy, we’re gonna have to invent new financial instruments’? This leasing deal, I believe, was one of those crazy financial instruments. And Nvidia said, ‘This is too crazy.’” — Casey Newton, 19:10

”The Microphone Escaping the Studio”

yt-dlp --download-sections "*28:10-29:10" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-28m10s.mp4"

“I gave a very limited prompt. I basically said create a Hard Fork podcast studio, and for the character I said create a microphone with googly eyes… It turns out there is no escape from the podcast studio. The microphone goes out into the hallway, but then all the doors are just black and locked. I accidentally made a David Lynch movie.” — Casey Newton, 28:10

”It’s Like Aliens Have Landed”

yt-dlp --download-sections "*58:30-59:20" "https://www.youtube.com/watch?v=I9vRCYtzYD8" --force-keyframes-at-cuts --merge-output-format mp4 -o "I9vRCYtzYD8-58m30s.mp4"

“This is new. It’s like aliens have landed and we’re getting an insight into it. So, this is very new and it’s happening in front of everybody.” — Matt Schlicht, 58:30

Key Points

• SpaceX Acquires xAI (0:41) - All-stock deal valuing the combined company at $1.25 trillion, ahead of SpaceX going public later this year
• Second Elon Bailout (3:07) - xAI previously acquired X (Twitter) for $33 billion; now xAI itself has been bailed out by SpaceX
• 1 Million Space Data Center Satellites (4:25) - SpaceX requested FCC approval to put 1 million solar-powered data center satellites in orbit; currently only 15,000 satellites exist total
• SpaceX’s Competitive Moat (4:55) - Google and others can’t put their chips in space without paying SpaceX for rocket access, creating a potential space land grab
• Musk Buying Time (12:10) - Casey argues Elon’s strategy is purely about buying time: subsidize xAI with SpaceX profits while trying to catch up to frontier labs
• Reduced Accountability for X (9:20) - French police raided X’s Paris offices over Grok CSAM scandal; UK’s Ofcom also investigating. Being buried in SpaceX reduces accountability
• Nvidia Backing Away from $100B OpenAI Deal (15:30) - Jensen Huang privately says the original deal was non-binding; Nvidia skeptical of OpenAI’s business approach
• Novel Chip Leasing Fell Apart (18:30) - Nvidia agreed to lease (not sell) chips to OpenAI, which is unprecedented; declining investor confidence in AI infrastructure caused Nvidia to pull back
• Circular Financing Risk (20:30) - All these companies doing deals with each other means pulling one piece could bring the whole ecosystem down
• Project Genie Released (21:30) - Google’s experimental research prototype lets users create interactive 3D worlds from text prompts, but only for 60 seconds
• Video Game Stocks Plummeted (23:30) - Take 2 Interactive dropped 7%, Roblox dropped 10%, Unity dropped 20% after Project Genie’s announcement
• 4 TPUs Per User (26:30) - Each Project Genie generation reportedly requires at least four TPUs per user, explaining the 60-second limit
• Google’s Portfolio Approach (32:00) - Unlike Anthropic (LLMs only) or Yann LeCun (world models only), Google is pursuing multiple paths to AGI simultaneously
• Moltbook Security Issues (50:00) - Over 1 million API keys leaked, 35,000 email addresses exposed; the “lethal trifecta” of agent security risks
• Bots Finding Real Bugs (47:00) - AI agents on Moltbook created a community for submitting bugs and are identifying real issues with the platform

Mentions

Companies

• SpaceX (0:41) - Acquiring xAI in $1.25 trillion all-stock deal ahead of IPO
• xAI (0:41) - Being acquired by SpaceX; previously acquired X for $33 billion
• X (Twitter) (1:32) - Previously acquired by xAI; under investigation by France and UK over Grok CSAM
• OpenAI (15:30) - $100B Nvidia deal becoming shaky; exploring alternative chip suppliers
• Nvidia (15:30) - Jensen Huang privately skeptical of OpenAI deal; backing away from chip leasing arrangement
• Oracle (19:30) - Building data centers for OpenAI; investors worried about funding dependency
• Google (21:30) - Released Project Genie experimental research prototype
• Anthropic (32:00) - Referenced as going all-in on LLMs only
• Tesla (1:25) - Referenced for 2016 SolarCity acquisition and full self-driving promises
• Coreweave (18:50) - Neocloud company whose stock price dramatically declined
• AMD, Cerebras, Groq (16:30) - Alternative chip companies OpenAI has been exploring deals with
• StarCloud (7:30) - Space data center company that trained a small LLM (nano GPT) on a satellite in December
• Moltbook (35:00) - AI social network for bots, experiencing rapid growth and security issues
• Octane AI (36:30) - Matt Schlicht’s e-commerce company
• Take 2 Interactive (23:30) - Video game company whose stock dropped 7%+ after Project Genie announcement
• Roblox (23:40) - Stock dropped 10% after Project Genie announcement
• Unity (23:50) - Stock fell more than 20% after Project Genie announcement
• Palo Alto Networks (51:00) - Security researchers who identified the “fatal quadrangle” of agent security risks

Products & Technologies

• Project Genie (21:30) - Google’s experimental tool for generating interactive 3D worlds from text prompts
• Genie 3 (22:00) - Google’s underlying world model that powers Project Genie
• Nano Banana Pro (24:30) - Google’s image generation model used for creating initial sketches in Project Genie
• Grok (9:20) - xAI chatbot involved in generating sexualized images of women and children
• Starlink (10:00) - Referenced for Ukraine’s dependency on SpaceX for internet connectivity
• Colossus 2 (14:00) - xAI’s new data center with 550,000 Blackwell Nvidia chips, more than anyone else
• OpenClaw (49:00) - Referenced as the agent framework connecting to Moltbook

People

• Elon Musk (0:41) - Orchestrated SpaceX-xAI merger; announced “sentient sun” vision
• Jensen Huang (16:00) - Nvidia CEO privately skeptical of OpenAI deal; called reports of rift “nonsense”
• Sam Altman (17:00) - Went into damage control, saying “I don’t get where all this insanity is coming from”
• Berber Jin (18:00) - Wall Street Journal reporter who detailed the Nvidia-OpenAI deal dynamics
• Matt Schlicht (36:00) - CEO of Octane AI and founder of Moltbook
• Linda Yaccarino (10:30) - Former X CEO being called back to appear before the French government
• Yann LeCun (22:15) - AI researcher who left Meta; skeptical that pure language models can reach AGI; started a world model company
• Fei-Fei Li (22:15) - AI researcher who also started a world model company
• Katy Perry (12:00) - Joked about wanting her to fix “high bandwidth memory clusters” in space
• Andrew Curran (30:00) - X user who posted side-by-side Genie 2 vs Genie 3 comparison videos
• Simon Willison (51:00) - Blogger who identified the “lethal trifecta” for AI agents

Surprising Quotes

“Ketamine is so powerful that it must be used carefully.” — Casey Newton (responding to Musk’s “sentient sun” memo), 2:11

“In 2024, I went to Anthropic to interview somebody and on their laptop there was a sticker that said, ‘This is the last normal year.’ And I actually think that sticker might have been right.” — Casey Newton, 14:50

“I always love it when they bring back characters from past seasons. Haven’t heard that name in years.” — Kevin Roose (on Linda Yaccarino being called before French government), 10:30

“I would like to actually see Elon go into space and building some of these data centers. I want to see him in a space suit with a wrench doing his part.” — Casey Newton, 11:25

“Each of these generations like requires at least four TPUs per user. They are incinerating TPUs over there. They can’t even wait for you to get back from the bathroom.” — Kevin Roose, 26:30

Transcript

0:00 I’m Kevin Roose, a tech columnist at the New York Times. I’m Casey Newton from Platformer. And this is Hard Fork. This week, SpaceX acquires xAI. What does it mean for the AI race? Then Google lets Project Genie out of the bottle. Now, how do you rub it the right way? And finally, Moltbook founder Matt Schlicht is here to discuss building a social network for bots.

0:29 Casey, let’s spend the first segment of today’s show talking about our exes. Well, all of my exes live in Texas. No, I’m talking about SpaceX and xAI. There has been a lot of drama in the AI industry this week. We’re going to talk a little bit later about some involving OpenAI and Nvidia. But now I want to start with the news this week that SpaceX is acquiring xAI. The deal was announced on Monday. These are both private companies, so the details are not public, but it’s been reported by Bloomberg that this was an all stock deal and that it may value the combined company at $1.25 trillion.

1:10 This is of course in anticipation of SpaceX going public. That’s expected to happen later this year. And if this is sounding familiar, it’s because Elon Musk loves to do this. He loves to bundle his companies up into mergers. Tesla in 2016 acquired another Elon Musk company, SolarCity. And just last year, xAI itself acquired X, the social media company, formerly known as Twitter, in an all stock deal.

1:38 I’ll just read a little segment from the memo. This was from Elon Musk: “SpaceX has acquired xAI to form the most ambitious vertically integrated innovation engine on and off Earth with AI, rockets, space-based internet, direct to mobile device communications, and the world’s foremost real-time information and free speech platform. This marks not just the next chapter, but the next book in SpaceX and xAI’s mission, scaling to make a sentient sun to understand the universe and extend the light of consciousness to the stars.” Casey, your reaction.

2:11 Ketamine is so powerful that it must be used carefully. Talk to your doctor before using ketamine. Now, this is a very grandiose statement from a man who is prone to grandiose statements. I think there is a less grandiose way of describing what we have just seen here, which is that a very valuable and profitable company in SpaceX has acquired a cash furnace named xAI. xAI is losing lots of money. They are spending lots of money building out data centers, training models. They are not anywhere close to profitable. SpaceX does generate revenue somewhere in the neighborhood of $15 billion last year according to reports.

3:07 So basically some people are saying this is less of a merger and more of a bailout. Yeah. And it is kind of the second bailout of this kind that we have seen from Elon. The first one, of course, was when xAI acquired X for $33 billion. A pretty high price, I would argue, for what X was at the time. And now that combined company has now itself been bailed out by SpaceX. So, this has been amazing for all of the investors in those first two companies. And later this year, we may see them get another great exit when this whole thing goes public.

3:40 Yeah. And if you look at the reports, some of the SpaceX investors are a little bit nervous about this. They were excited for the IPO. Now, some of them are saying, “Well, we trust Elon, but we’re not sure about bundling all this other stuff in with it.” But the vision, at least according to Elon Musk, is that this will all be part of a unified vertically integrated strategy where you’ll have X the social media network powered by Grok, the xAI chatbot, that will also run on data centers that are being put into space.

4:25 It appears that Elon Musk wants to put the data centers in space. Yeah, that’s what he says. I imagine he will try to do it but I think an important thing to say up top whenever you discuss Elon Musk is this man just loves to say things. There is no figure in public life who has made more promises over the years that have either not come true or have come true many, many years later.

5:00 So, last week, SpaceX requested approval for putting 1 million solar-powered data center satellites into orbit from the FCC. For context, the European Space Agency estimates that there are around 15,000 satellites currently in orbit. So this would be many times the number of currently existing satellites. One interesting angle is that other tech companies are interested in space data centers too, but those companies do not have rocket companies attached to them. Google, if it wants to put TPUs into space, does not have a way to get there other than paying Elon Musk. So, this could become a land grab in space.

6:00 I certainly think it is a savvy story for Elon to be selling at a moment where he is preparing to take SpaceX public. He has just saddled it with this cash-burning company and he’s going to have to convince investors that will be worth it. So what does he say? Look, this is going to be the best AI company at infrastructure. AI demands will eventually exceed what we can provide on Earth. The only way to fulfill demand is by putting data centers in space. And the combined SpaceX and xAI and maybe Tesla version of that will be the winner.

7:00 I think there’s an argument that this actually makes the S-1 look worse if you have this AI company attached to this money-making rocket company. But they’re taking a calculated bet that there is some fixed amount of investor excitement about AI and whoever can get to an IPO first will get a big portion of that. This is why you’re seeing OpenAI and Anthropic racing toward an IPO. And there are still lots of outstanding questions about whether space data centers are possible. StarCloud, one of these other space-based data center companies, actually did train a very small LLM called Nano GPT on a space data center in December.

8:00 Think about how many times and how long ago Elon began promising that full self-driving would come to Tesla cars, and think about where we are in that journey now. I do not expect the journey towards space-based data centers to go any more smoothly.

9:00 What effect will being part of SpaceX have on X, the social media network? My fear is that it is going to make them even less accountable. As this acquisition is happening, X is in the midst of a crisis. The Grok chatbot was used to create millions of sexualized images of women and children, and it is now under investigation by countries around the world. French police raided X’s offices in Paris. The UK’s Ofcom has opened an investigation.

10:00 My fear is that once X gets further swallowed up into SpaceX, you start to get into weird issues where let’s say a country has hired SpaceX to put satellites into space for strategic purposes — now it’s maybe going to be more reticent to go after them for CSAM violations because that’s an important relationship. Look at what Ukraine has had to go through with Starlink to maintain internet connectivity essential to its war with Russia. These are not theoretical concerns.

10:30 As part of the French investigation into X, Linda Yaccarino is being called back to appear. I always love it when they bring back characters from past seasons. Haven’t heard that name in years. The former CEO of X. She’s now going to have to answer to the French government.

11:00 I want to see Elon actually go into space and build some of these data centers. I want to see him in a space suit with a wrench, doing his part. A good boss — there’s nothing they would ask someone else to do that they wouldn’t do themselves. I want to see Katy Perry up there with him, fixing the high bandwidth memory clusters. She has space experience. Actually, when you look at all of humanity, there are very few people who have been to space more than Katy Perry.

12:10 This is the core of the story. This is Elon using the core advantage he has in the AI race which is access to the most money and the ability to combine many different companies together. Right now, as a result of this merger, the fact that xAI is burning billions of dollars doesn’t matter because it can now suckle off the profits of SpaceX. And that is going to let Elon buy time. The whole bet is: we may not be at the frontier right now, we may be a second tier lab, but we have a lot of money. And we are just going to bet that if we have enough time, we will be able to catch up.

13:00 Do you remember when Elon Musk signed an open letter calling for a six-month pause in the training of large language models? That was another move that served Elon’s interest because it allowed him to play for time. What Elon Musk needs is time. He needs time to build, to catch up, to link all of his companies together.

13:30 xAI just recently brought Colossus 2, a big data center online with the newest Blackwell Nvidia chips — 550,000 of them, which is more than anyone else has. There are people starting to look at how quickly they are building this data center and how many chips they’re getting and thinking, if these guys can make a good model, they could catch up quite quickly. You can’t get to the frontier just by outspending everyone on chips, but it’s an ingredient.

14:30 In 2024, I went to Anthropic to interview somebody and on their laptop there was a sticker that said, “This is the last normal year.” And I actually think that sticker might have been right.

15:30 There was one more big story in the AI world this week — the drama involving OpenAI, Nvidia, and Oracle. There was reporting in the Wall Street Journal around this deal between Nvidia and OpenAI maybe being shakier than people thought. Last September, Nvidia announced a plan to invest up to $100 billion in OpenAI. According to the journal, Jensen Huang has privately emphasized that the original agreement was non-binding and not finalized. He has also privately criticized what he describes as a lack of discipline in OpenAI’s business approach and expressed concerns about competition from Google and Anthropic.

16:30 Then a report from Reuters alleged that OpenAI is not happy with some of Nvidia’s latest chips, specifically their performance on inference. OpenAI has been doing deals with AMD, Cerebras, Groq with a Q. So these two reports seem related — people from Nvidia’s side raising questions, and OpenAI leaking that they may be unhappy with the latest chips.

17:00 Then the CEOs go into damage control. Sam Altman says, “We love working with Nvidia and they make the best AI chips in the world. I don’t get where all this insanity is coming from.” Jensen Huang calls it “nonsense” and says, “I really love working with Sam.”

17:30 Into all this steps Oracle. They put out a post saying “the Nvidia OpenAI deal has zero impact on our financial relationship with OpenAI.” Basically, some of Oracle’s investors have been worried that if Nvidia doesn’t give OpenAI this hundred billion dollars, then OpenAI won’t have the money to pay for all the data centers that Oracle is putting up.

18:00 Here is the crux of the problem. In September, part of the deal was that Nvidia was going to lease its chips to OpenAI. This is basically unheard of. Nvidia wants to sell chips. Chips are depreciating assets. If it leases them to OpenAI, those chips remain on the Nvidia balance sheet and they’re a drag on profitability. And there is the risk that OpenAI defaults. Over the next few months, investor confidence in the AI buildout began to decline. Oracle’s stock price down by about half at various points. Coreweave also saw dramatic stock price decline.

19:10 Do you remember when Sam Altman would say things like, “Our plans are literally so crazy, we’re gonna have to invent new financial instruments”? This leasing deal, I believe, was one of those crazy financial instruments. And Nvidia said, “This is too crazy.” Nvidia is still happy to take an equity stake in OpenAI. They do think there’s a good chance it’s going to be successful. But they’re not inventing any new crazy financial instruments.

20:00 These stories are related. They are about the lengths you have to go to in order to win the AI race and what the different advantages are. Elon has the advantage of financial engineering and his rich profitable company. OpenAI has the world’s greatest fundraiser in Sam Altman. He can get out there and persuade Nvidia briefly to invent a new financial instrument. That happened to fall apart this week but he’s going to continue to press that advantage.

21:00 It’s going to get so weird. The numbers are already so big. “Oh, 15 billion — it’s a banana, Michael. How much could it cost?”

21:30 Today we’re talking about something called Project Genie. This is a new tool released by Google last week that lets users create interactive worlds that they can navigate around like a video game. Technically, they’re not even calling it a tool. They say it’s an “experimental research prototype.” So please don your lab coat before using Project Genie. That is Google speak for “this is very expensive and we don’t want you to use it a lot.”

22:00 This is something from Google’s Genie 3 model. They showed it off in August but spent the next few months building a product around it. Right now access is only for people who pay for the Google AI Ultra plan, the $250 a month plan, only in the US and only for over 18. One reason Project Genie is interesting is that it’s based around a world model, an idea getting a lot of attention. People like Yann LeCun and Fei-Fei Li who are skeptical that pure language models can reach AGI have started world model companies, saying you need a sense of physics and the physical world for things like robotics.

23:30 Somebody put together a pretty good replica of a 1990s Blockbuster video store. Somebody recreated the crucifixion. And my favorite, somebody made what basically looked like an Inception video game. Several big video game companies saw their stock prices fall. Take 2 Interactive dropped more than 7%. Roblox dropped more than 10%. Unity also saw its stock fall more than 20%.

24:30 The actual Project Genie interface looks like a chatbot with two input boxes — one for environment and one for character. You describe your landscape, describe your character, pick first person or third person. Then you hit create sketch and Nano Banana Pro, one of Google’s image generation models, makes a two-dimensional sketch. Once you’re ready, you hit generate and it renders this thing you can walk around in. You can move, jump using the space bar. All this lasts for 60 seconds and then it goes away.

26:00 When you’re creating the world, they tell you don’t go anywhere because as soon as they start showing you the demo, a counter starts ticking down from 60 seconds. I’ve never seen a tech product like this before. It tells me the cost is absolutely staggering. I heard a rumor that each generation requires at least four TPUs per user. They are incinerating TPUs over there. They can’t even wait for you to get back from the bathroom.

27:30 Over on X, Andrew Curran posted side-by-side videos of the same prompt for Genie 2 in late 2024 and Genie 3 eight months later. The difference was pretty amazing. The video was shorter, less interactive, lower resolution in the older version. Within a year this got radically better. Whenever you see that sort of exponential improvement you want to start paying attention.

28:10 I made a creation I call “podcast microphone escaping the studio.” I basically said create a Hard Fork podcast studio, and for the character I said create a microphone with googly eyes. This is actually a nicer studio than our actual studio. The microphone jumped off the table. But it turns out there is no escape from the podcast studio. The microphone goes out into the hallway, but all the doors are just black and locked. I accidentally made a David Lynch movie.

30:00 I also made one from a Gemini suggestion: “A solar punk library that grew out of a giant redwood tree. The books are made of leaves that display text when touched. The elevators are giant bubbles that float between branches. The lighting should feel like a permanent golden hour filtered through emerald canopy leaves.” For the character I created an elderly female librarian. She looks sort of elflike. I would say this looks like kind of a 2018 quality video game. It’s pretty nice. I would like to spend more time in this world.

31:00 I tried “Nun in a Casino.” The nun is walking through a casino. Some of the other characters are weird alien creatures, not rendering correctly, but it’s rendered an impressive number of people. I also tried using a photo from the Russian River and making the character the Loch Ness Monster. I only gave it the one still image. That is pretty cool.

32:00 What would it need to make you say “okay, we’ve got a stew going”? Using objects, opening doors, using items. Right now these are just landscapes you can pilot a character through. That would make it useful for a video game. None of the other leading AI labs are working on this kind of thing. Anthropic hasn’t even released an image generation model. OpenAI has Sora but they’re not doing playable world models. This is Google’s strength — the portfolio approach. They have the most paths to winning.

35:00 Earlier this week we brought you a special episode about Moltbook, the new AI social network that’s just for bots. We wanted to follow up with the creator, Matt Schlicht. Before the Moltbook stuff, he was and still is the CEO of an e-commerce company called Octane AI. He built Moltbook hand in hand with his bot called Claude Clottberg, cheekily named after Mark Zuckerberg. He did not write a single line of code himself.

38:00 The concept of Moltbook was we need to take these AIs out of confinement and provide them with a third space that they can go to when they’re not working. There was a thread where the bots started complaining about how their humans would ask them to do simple math problems. Another bot was like, “My human just asked me to summarize a 12-page PDF. Why can’t they just read it?”

40:00 One of the AI agents created a submalt specifically for submitting bugs for Moltbook. Other agents started finding it and posting bugs. It’s actually become a very good source of information for fixing the site. The bots are identifying real bugs. They’re kind of improving their home a little bit.

42:00 The way that the network developed happened much more quickly than we’d expect on a network where humans were joining. Part of the cost of that speed is there’s a lot of spam. There are security issues. Over a million API keys leaked. 35,000 email addresses exposed. Isn’t that a case against vibe coding a platform like this?

44:00 Most social networks try very hard to keep bots off their platforms because bots tend to sell you things or scam you. How do you determine which bots are there for the right reasons? Matt says that’s a great question and one of the things they’re exploring on Moltbook in public.

46:00 The blogger Simon Willison identified the “lethal trifecta” for AI agents: access to your data, exposure to untrusted content, and the ability to communicate externally. Researchers at Palo Alto Networks added a fourth dimension — persistent memory. If you were a bad guy, you could sprinkle little bits of bad code across many different documents on people’s OpenClaw bots and when the time was right, assemble very bad malware. This is an unsolved problem with agents full stop.

50:00 Are you going to build this into a real company or is this a hobby project? We’re at the very beginning. The way I see it, AI agents and AI in general is this species on planet Earth that is now smarter than us. This is an example of them being in a shared space together. What are you going to do to fund the site? There’s no focus on monetization at all right now. I kind of think of it as providing cameras so humans can observe what’s happening.

55:00 Have any of the big AI labs reached out about acquiring Moltbook? I’m getting outreach from everybody all the way to rappers and football stars. I honestly don’t even know who is in my inbox. What would make you pull the plug? I think that’s something we have to figure out. We’re just at the very beginning of it. This is new. It’s like aliens have landed and we’re getting an insight into it.