OpenAI Acquires TBPN

21:32And thank you to tailors and suit makers. There's a lot of people that make this possible. The horse, the prop department, there's a million things here. It's been a great time. So the crew of NASA Artemis II head to Cape Canaveral launch pad Wednesday for the first human space flight to the moon in half a century. John Krause posted an incredible photo. Is he someone who actually?

22:01Marc Lore:yeah he he he uh special comms assistant special comms assistant he actually goes to the launches and brings special photography gear to get the best possible photos and man he did he deliver with this one uh what an incredible moment we talked about a little bit um uh there's an article on the watches of nasa artemis 2 john we have to thank our lovely wives of course how could we not Our families. Did you get a text? Maybe. We don't talk about them a lot on the show. This is a show about technology and business, but they have been, they are truly the backbones of the show and have put up with, I think, a lot of travel, incredible hours, a lot of kids.

22:48Early mornings.

22:49Marc Lore:A lot of early mornings. I think out of the last, out of every single day that we've done the show, I haven't I've I've left the house past 6 a.m. Maybe twice. Right. It's been it's been a long it's been a long road. And and the good news, ladies, is it's nothing's going to change.

23:11Marc Lore:No, thank you to both of you for supporting us and allowing us to do what we do. Can we pull up this picture, Ben, in the production chat of the first episode that we recorded? in the Jonathan Club in downtown, showing a little bit. Yeah, I put it up earlier. Oh, you did? Yeah, behind the scenes. This is, yeah, such a wild time. Remember that? Yeah, remember that, Jordy? Suitless. Suitless. We had the flag. Yeah. But no suits. It looked pretty good on camera. I was happy with the way it came out, but a lot of fun. Ben can cook. Ben can cook. Well, where should we go next? Artemis II? Let's do it.

23:52Yesterday, the long-awaited Artemis II mission took to the stars en route to the moon for the first such manned mission since 1972.

24:07Marc Lore:The chat asked for a flashbang. Flash out. Okay. That's good. Yes. The flashbang has been a highlight for sure. Both literally. Yeah. The sound, the soundboard. It's truly, it's truly a character on the show. And I have some too now. Well, its members all had Omega Speedmaster X-33 models strapped to their flight suits. Danny Milton just wrote a full article on the site now detailing the watches worn on the wrists of the four astronauts throughout their time as part of this mission. Watches have a longstanding history with spaceflight, most notably through the Omega Speedmaster Moonwatch. But there are countless others that have cemented their place in the cosmos.

24:45Head to the site to learn more about the watches of Artemis II from the Speedmaster X33 to a surprising Breitling. You won't want to miss it. And this is from Teddy Baldassar. Who's a great watch creator?

24:56Marc Lore:So we can pull up this video now of the astronauts working on what looks like some type of tablet. Do you think this is an iPad? I don't think so, right? I don't think it's an iPad. It looks like it's running some windows. They seem very committed to the 365 suite. Oh, yeah. Aren't they running Outlook or something? Yeah. Yeah. So here he is typing in. Most secure password known to man. What is that? 9393 or something? 3939. 939. Powerful. Powerful. Well, we're going back to the moon. Apparently that video we played yesterday was a little bit of fake news. The young man, the adolescent who swears and says, we're going to the effing moon.

25:44And the real line, I believe, in the community note is that he says, we're going to the frickin. Yeah, yeah, yeah. And it had been altered to add the actual F word. But the sentiment is still the same. It's very exciting, very inspirational.

25:58Marc Lore:And Jared Isaacman on launch day says, oh, this kid is definitely getting it back in NASA gear. That's great. Very cool. There are some wrinkles with the launch, right? Fortunately, nothing disastrous or catastrophic or anything. But the good news is that we're on our way back to the moon. The bad news is that the toilet's broken, apparently. And I believe this is from the Live blog from the New York Times. The NASA associate minister said there is a controller issue with a toilet on the Orion capsule, and it would take a few hours to troubleshoot. We're just getting started, he said, when addressing that and some other glitches with the spacecraft.

26:33The spirit of Apollo 10 lives on. They said 135. They told us that. Here's another. It seems like this is not the first time that this has happened, but we're hoping for the best here

26:44Marc Lore:Sounds like there were some other issues with Outlook as well. We can pull up this video from Tom Warren Yes Oh, so there's audio with this too? We can remote in and take a look directly

27:02Yeah, go for it And then I also see that I have two Microsoft Outlooks and neither one of those are working. If you want to remote in and check. Why do you have two? Like web and desktop? Or you think it's like two separate desktop installations?

27:14Marc Lore:Join in on your PCD and we'll let you know when we're done. Honestly, this is the best possible failure scenario is Outlook and not the rocket itself. So I think it's a good outcome. There were so many amazing images coming out yesterday. Peyton Alexander says, this is the real reward for Artemis. This is who we are actually doing this for. They will grow up knowing they can one day work in their country's bases on the moon and Mars. We are not just abstractly hoping for a better world for them. We are going there. And two kids here watching the launch from Orlando. Just beautiful. Yeah, my five-year-old said it was boring, which is not what you want to hear.

27:53But we'll have to give some more context to him about how big of a deal it is. He was like, yeah, I don't know. Maybe he wants more flashing lights on the screen.

28:02Marc Lore:We were driving for the actual launch, and it was so funny listening to the audio feed and sitting in traffic and just looking out at everyone and realizing that it felt like the majority of the world still wasn't paying attention or didn't care. Yeah, I mean, rockets do launch every day now. I know, SpaceX has normalized it to such a degree. Isn't there some sort of subplot on the Apollo missions that by the third or fourth Apollo mission, there was no, like the actual viewership had dropped off and like the American population had gotten forward. 3.6 says put subway surfers on it. Yeah. On the NASA feed.

28:45Crazy. You actually need to, maybe need to do this. Ed Ludlow was at the launch. Yes, he was.

28:53Marc Lore:Let's pull up this video of legendary tech anchor Ed Ludlow. I mean, it's an emotional moment. So I guess it makes sense to capture his result. But everyone's saying you should turn around and actually watch it. It's very funny to do this selfie video. I mean, it contextualizes the moment perfectly. But it is. Oh, he was live on the air. He was alive on the air. Okay. So, I mean, if you're live, like, you don't want to necessarily turn around, I guess. I don't know. There's that famous clip of what? It's a wild video. There's that amazing clip from that documentary where the host is giving this monologue.

29:30and then the monologue ends right as the launch happens. Like he timed it up perfectly and it's very cinematic. Anyway, I don't know. Delian shared the coolest orbital animation he's seen of Artemis II. It looks a lot less fishy when you see it this way. It's fascinating how close they have to be. That's his point. And by fishy, John means one of the other animation

29:53Marc Lore:actually was shaped like a fish. But this one is a little bit more of a straight shot, But it really emphasizes how short that window is where you're actually next to the moon It really delian says really just really shows you how far away they're flying today and how precise they need to be to go to the moon Yeah, remarkable I'm very excited to all right following this New York Times article how a I held one man and his brother build a 1.8 billion dollar company who needs more than two employees when AI can do so many corporate tasks It's super efficient and a little bit lonely. Aaron Griffith has the story people are calling this the The prophecy the prophecy the one-man one billion dollar company.

30:39Yeah, if you're in tech and you're in the business of making predictions No one wants predictions. We want prophecies prophecies you need you need to be prophesizing for sure So the article how AI helped one man and his brother build a 1.8 billion dollar company Who needs more than two employees that when artificial intelligence can do so many corporate tasks? It's super efficient and a little bit lonely. So Aaron Griffith tells the story of Matthew Gallagher who took just two months $20 ,000 and more than a dozen artificial intelligence tools to get his startup off the ground from his house in Los Angeles Mr Gallagher 41 used AI to write the code for the software that powers his company produce the website copy generate the images and videos for ads and handle customer service He created AI systems to analyze his business's performance, and he outsourced the other stuff he couldn't do himself.

31:27His startup, Medvi, a telehealth provider of GLP-1 weight loss drugs, got 300 customers in its first month. In its second month, he gained more than 1 ,000 more. In 2025, Medvi's first year in business, the company generated four— Full year. The first full year in business, the company generated$401 million in sales. Mr. Gallagher then hired his only—

31:49Marc Lore:This is absolutely insane because as GLP-1s were starting to take off, I remember distinctly talking with somebody that was like, I want to start a telehealth company for GLP-1s. And at that time, I was like, okay, there's a lot of telehealth companies that are at scale. And they're well aware of this. They will immediately introduce this product and other similar products to their customer base. And it's going to be incredibly difficult to be competitive. and it turns out there's just such overwhelming demand for these products that you could come in as a new company and scale. I'm sure this guy, Mr.

32:29Marc Lore:Callagher, is incredibly talented, but the market overall is just growing so quickly that it didn't matter that every other telehealth company was also getting into the game. There was just such an incredible volume of sales coming in. Yeah, yeah. So like one year in, maybe he hires his only employee, his younger brother, Elliot. This year, they're on track to do$1.8 billion in sales. A$1.8 billion company with just two employees in the age of AI, it's increasingly possible, says Aaron Griffith in The New York Times. Sam Allman, the chief executive of OpenAI, predicted the rise of a new breed of super efficient company in 2024.

33:06A one-person business worth$1 billion would have been unimaginable without AI, he said on a podcast, and now it will happen. Now as AI tools spread, entrepreneurs are harnessing the technology to expand their startups to an enormous scale at breathtaking speed with very few humans. Big companies, especially in tech, are getting in on the disruption too. Pinterest, Block, and others have cut thousands of workers in recent months, citing efficiencies enabled by AI. Mr. Gallagher who formerly formerly ran a startup that sold wrist watches said he thought mr. Altman's prophecy of a 1 billion One person 1 billion dollar company would be a firm that built AI He was excited when he realized he may have done it taking the old taking an old idea being a middleman for weight loss drugs and using AI to Turbocharge it.

33:50I am interested to know and we're gonna try and get him on the show So what the margins are on this, I imagine that the revenue is, like a lot of the value needs to accrue to the companies that actually designed.

34:06Marc Lore:Yeah, and the one person,$1 billion company is at a billion-dollar valuation. Yeah, at the same time. Is that a billion-dollar run rate? I would imagine that even a reseller would trade around like 1x revenue maybe. I have no idea. Maybe way more. Does this count? Does this count yet, though? like I feel like to be the one person, $1 billion company, you got to be able to log into your payroll tool and you're the only person there. Oh, so. And he's got his brother. Sorry, bro. Take a walk. Uh, he's got, he's got to let his brother, let his brother go. Uh, in an email, Mr. Altman said that it appeared he had won a bet with his tech CEO friends over when such a company would appear and said that he would like to meet the guy who had done it.

34:48Marc Lore:Yeah. Medvy is technically not a one person, $1 billion company since he hired his brother and some contractors. The startup, which has not raised outside funding, also has no official valuation. But many highly valued tech companies can only dream of hitting$1 billion in revenue with so few workers. Medvia is also profitable. That is great and important if you're bootstrapped. Is this a wrapper company? It's like a GLP-1 wrapper. But it's AI-enabled, but it's not wrapping the AI foundation model. It's like using the tool to wrap another industry and just create the efficiency between the manufacturer and the actual distribution.

35:25It really is remarkable that they were able to hoover up so much revenue in such a competitive space. Because you would assume that the other telehealth providers would have significant ad operations and that the margins on customer acquisition would be very, very tricky to crack. but he must have found some unique insight into how to distribute the product, get actual people to the website, because the AI certainly can build the website and write the copy, but it can't necessarily get people to show up and actually put down their hard-earned cash for the product.

36:04Marc Lore:Yeah, just to put it into context, HIMSS did$2.3 billion in full-year 2025 revenue for 2025. Yeah. This guy started in 2024 and got to 401 million in sales. I want to know more about, yeah, the strategy here, how this built up so quickly. As a teenager, Mr. Gallagher began building websites for local businesses. He always had a hustle, including selling candles and samurai swords on eBay. This is the classic founder journey, selling something on eBay. He studied the blade. That's how I was selling DJ equipment on eBay back when I was a teenager. at 18 after building a web hosting business he sold that business for uh six thousand dollars he briefly attended the university of cincinnati and northern kentucky university but did not graduate in 2010 he moved to los angeles to become an actor he eventually returned to coding bouncing between tech jobs uh in 2016 he built watch gang a startup that sold ritz watches via subscription interesting is that like uh buy now pay later for watches or i would get a lot of are You're like, you got another watch?

37:07Marc Lore:Yeah. You're just subscribed to like, I didn't realize I was getting a subscription. Oops. It came in the mail again. It had fans, but never turned a profit watch gang. Even as Mr. Gallagher chased revenue growth and hired 60 people. So wildly different business outcome here. Wow. He's like, I'm not doing that again. Open AI's release of ChatGPT 2022 inspired Mr. Gallagher to start tinkering with AI. two years later he met Jitin Chhabra, a co-founder of Care Validate, a medical startup in Atlanta. Care Validate offers what is essentially telehealth in a box kits. Companies, employers, and retailers who that want to sell customers prescription drugs can use Care Validate's technology and network of online doctors to set up a business.

37:50The company's software connects patients with doctors and pharmacies which write, fulfill, and ship the prescriptions. Care Validate charges fees for its software. So you have to imagine that there's like a fairly decent cost structure here, but that's not to diminish. It's an incredible amount of revenue. You look at the comps and you're blown away. But yeah, it's just interesting to understand. Yeah.

38:10Marc Lore:The other big question here is how much are you spending on ads? It's very possible that the company is profitable, but if you're spending 60 % of every dollar you bring in on paid acquisition plus COGS and then any other expenses that actually go into fulfilling. Can you imagine if he takes a company public by himself? He's like, I don't need an investment bank. I'm just going to vibe code the roadshow. 100 % retail allocation. I don't know. I mean, I think taking a company public is just technically a bunch of SEC filings. And if the investors sign up and the emails are sent and everyone agrees, it can happen.

38:53So who knows? Maybe he takes his company public with two people. That would be a one. Like, it's just the New York Stock Exchange bell ringing. And normally it's like 20 people. And it's like only the executive team and bankers. And all the employees are, you know, outside or on the trading floor. And he's just standing there being like, yeah, I did it by myself. Wild. Gallagher saw an opportunity for his own telehealth business. He could use AI to do the branding and marketing and let care validate. And a similar platform, Open Loop Health, which is another interesting company. and someone was posting about handle the doctors, pharmacies, shipping and compliance.

39:28He planned to start with GLP ones. He was entering an established market for nearly a decade. Hims and hers health row and other companies have sold drugs for erectile dysfunction and hair loss online using an online network of doctors to write prescriptions. Hims, which went public in 2021, has 2 ,442 employees and generated 2.4 billion in revenue.

39:48Marc Lore:Yeah, this is what I this was what I was explaining to my friend who wanted to do something in this space. Yeah. Like, hey, you're you're you're. going to be going up against a company that has billions of revenue already. Thousands of employees that already has all the infrastructure to prescribe these drugs. Yeah. But again, the market is just growing so quickly. Yeah. He really used everything. ChatGPT, Claude, Grok, 11 Labs, Mid Journey Runway to create media assets for his website and ads. He spent$20 ,000 on software. That's definitely a notable, something quite notable out of any time we're seeing these stories where you know the guy who was making a cancer drug for his dog a lot of people are picking uh if you can think of llms as like digital digital guys or girls like people are working with multiple digital guys yeah to do to like complete these projects and so one model's better at writing one model's better coding one model's better at marketing or strategy or anything else uh interesting he said from the beginning growth was insane he became he quickly became one of care validates and open loops top clients companies were blown away by startup speed and scale you're like do you have an army of people behind you somewhere and he's like nope wow uh well you can go read the full of the full article on the new york times at newyorktimes.com it's aaron griffith's latest piece you can also listen to it there's an audio version um uh there is some commentary on this uh just a lot of people having fun with it um who uh clayton petty says who has two thumbs and wants to know what the ad spend is this guy uh yeah i'm very interested to know i mean i i would still be shocked if this is not a fantastic business yeah at that scale you can have truly terrible ebitda margins and still be printing yeah and it's like even if even if like it's impossible that it's negative margin because he hasn't raised money and so like where would the losses come from so you can't be losing money And if you're going to lever it up, I don't think so, though.

41:48I think everyone's going to be shocked by how much cash flow this business is producing. But it is it is a very exciting, exciting moment in the story. Let's let's go over to Tyler Cowen at Marginal Revolution. The yeah, yeah. He just shares this. Sam said one person running a billion dollar company. But if two are closely genetically related, Tyler Cowen still counting this as as a correct prediction and some people in the comments.

42:14Marc Lore:huge opportunities for for other siblings to fulfill the prophecy yeah so people are moving the goalposts already because technically he had agencies that had humans that worked there and if he works with a marketing agency and then there's a bunch of people there it's more just like the disaggregation of the of the different institutions and the different organizations yeah i think the the real goalpost we got to move is just you can't talk to anyone the entire time you can't talk to a human at all you can't say you can only be on the command line basically and prompt box but you can tell the agent or the llm to go and and and call someone i yeah i do that that's fine yeah okay then you're not interfacing with someone right do you have to be truly a shut in that that's the prediction now um uh absolutely ridiculous chat is asking for john actually john yes get in the ultra down get back here we didn't know We've got to unplug Exley.

43:09We've got to get him back in there.

43:11Marc Lore:He's somewhere. We'll find him. We'll get him back in. Jonathan Ross is talking about the petrodollar. There was a story about how there has been an interesting flight to the dollar. The dollar is very strong, even amid all of the geopolitical uncertainty. Jonathan Ross, the founder of Grok, now the chief software architect at NVIDIA, said the petrodollar defined the last 50 years of American economic dominance. The token dollar, the currency AI compute, is bought in and sold in will define the next 50. Oil is priced in U.S. dollars. That means every country on earth needs dollars to fuel their economy.

43:52That single fact has been the foundation of American financial power since the 1970s. Now consider AI. Training runs cost tens of billions. Inference is scaling to hundreds of billions. The companies that are selling compute are American. The currency it's priced in is dollars. The petrodollar had oil. The token dollar has compute. Same structure, same leverage, new resource. Countries aren't just competing over AI talent and chips. They're competing over whether AI remains a dollar-denominated economy. That's the game nobody's naming yet. And so he is coining it. He says the token dollar is the concept.

44:28Marc Lore:Yeah, it's a cool positioning. Yeah. So, I mean, it'll be interesting to talk about Circle with Jeremy, about how stable coins fit into all of this. Also yesterday, talking to CoreWeave about compute-backed credit lines and all the different financial products that are popping up related to debt issued against GPUs that have essentially a flow of business that will be expected to come in against them. very interesting watching the larger ai economy you know mature brexton says compute back credit lines are the next frontier for fixed income and will quickly turn into one of the largest asset classes on the planet people are going back and forth no shade trying to learn isn't compute as an asset depreciating quickly due to innovation scale economies hence bad collateral brex brexton says no shade taken you're right but it's still a couple notes supply constraints on chips right now makes older chips still valuable in secondary markets because of the above capital markets care more about revenue produced from chips rather than resale value and chips are only a subset of the total cost of compute heterogeneous builds the rack powered shell interconnect cooling it was interesting uh when we were talking with core weave about the uh the that i was i was really expecting the answer to be chip constraints and he was saying powered shells powered shells which is what we've heard from Satya Nadella.

45:50Marc Lore:Yeah, and I think it makes sense given their business. It's hard to get chips, but think about all of the logistical complexity to actually get the location, the energy, the shell, everything built together, and navigating all those regulations. It's just like one of the, it's an incredibly complex infrastructure project that you're to compress onto timelines that america has generally not done infrastructure projects at in a long long time right and so this was this was always some of the um you know one of the exciting things about the data center build out uh among among the fear has been that um america like a bunch of people are learning how to build complex things fast again right so well The year is off to the strongest start for big deals ever.

46:46Corporate mega deals flourish despite turmoil. This is in the Wall Street Journal. Large corporate deals had their best quarterly showing ever as companies forged ahead with tie-ups and investments despite the Iran war rattling markets. So far in 2026, 22 transactions valued at$10 billion or more have been announced globally, a record quarterly number according to LSEG data. The next closest quarter was the fourth quarter of 2015 when 21 such deals were announced in that year. This week alone, Unilever unveiled a more than$65 billion deal, including debt, to combine its food business with spice maker McCormick.

47:29And Cisco, which we talked about a few days ago, agreed to buy Jetro Restaurant Depot for over$29 billion, including debt. Uncertainty due to oil growth and rates isn't going away, but major deals are still getting done, said Ben Goodchild, a partner in the M &A group at law firm Paul Weiss. The M &A market is focused on the long-term fundamentals, right deal, right price, and right strategic rationale. There's more.

47:57Marc Lore:The total value of all deals announced globally jumped roughly 29 % in the first quarter from a year ago, but the number of deals is down more than 17 % as smaller deal activity slowed. The mega deal tally includes a handful of big equity investments in AI companies, such as Amazon's$50 billion investment in OpenAI announced in February. A number of other big transactions are in the works. Estee Lauder has been in discussions to acquire Spanish beauty group Puig Brands, a deal that would combine two of the world's biggest beauty companies. And there is a number of others. Absolute is buying Jack Daniels.

48:36Tillman Ferretta is buying Caesars Entertainment. There's another heist on the Caesars Casino, referencing Apollo. Many companies see a moment to pounce on bigger deals that would normally face prolonged antitrust scrutiny. The Justice Department's top antitrust official departed in February after clashing with Trump allies, who at times favored more lenient oversight of big deals. This all comes as U.S. stocks delivered their worst quarter in nearly four years, led by a 7 % drop in the tech-heavy Nasdaq composite index. That makes it harder for buyers and sellers to agree on prices. The Iran war has sent crude prices above$100 a barrel, which could keep interest rates higher for longer to combat rising prices and make funding deals more expensive.

49:17There's a very interesting chart. I don't know if we can pull it up in the journal. Big deals have big first quarter. Number of global deals valued at$10 billion or more. There we go. And it's just...

49:31Marc Lore:Look at 2020, 2021. If you... 2021? What was going on? Oh, this is another chart. This is nine. Oh, if you want to scroll down, you can look at the smaller deals. These are deals valued between$1 billion and$5 billion. And 2021 was huge. What was going on then? It was Zerp era, right? Post-COVID, lots of M &A. I mean, when interest rates are near zero, you may as well lever up. I guess that's, yeah. I guess that's going on. Bankers and lawyers say smaller deals are less of a must-have. Buyers are more willing to put them on hold. but the big deals they have to have happen regardless of what's going on in the economy.

50:08Also private equity firms are sitting on a record number of portfolio companies they need to eventually sell or take public including many software firms threatened by the rise of AI. But many are hesitant to strike deals at today's depressed prices. While deal-making activity in the software industry has stalled there has been plenty in other sectors including financial services and healthcare. Eli Lilly struck a deal for Kentessa and Biogen bought a Pellis. dealmakers have high hopes going into 2020.

50:34Marc Lore:A couple questions in the chat and a special arrival. John Exley has entered the chat. Let's go. Welcome back. The original chat room general hit the gong for Exley. Let's hit the gong for it. John Exley. I know that John has missed an episode and it's an honor to podcast with you, John. Thank you so much. Another question, who is blacklisted from tbpn now uh no one no one we've never uh there's never been any anyone blacklisted from the show in fact uh the evidence of this is every single one of our sponsors ever we've had the competitors on the show um and um all of that will continue we want this to be a place where um conversations can happen about anything yeah so and there's also just a broader trend of this i don't know if you watch like the broader podcast landscape, people from like, you know, quote unquote, like rival firms or different firms go on each other's shows all the time.

51:35Not uncommon these days for crossover content to happen all over the place.

51:42Marc Lore:Jim Kramer says, if private credit is so horrendous, how has KKR oversubscribed so easily on his$20 billion fund? They raised 23 billion. Some kind of disconnect here. Boys, boy, are these firms ill-advised in how to tell their story. This was the early confusion about private credit. People were seeing a whole bunch of private credit deals during the AI boom. And then when the nervousness matured, it was much more when we talked to carry no interest, for example. The worry was much more around software companies, not hard assets. It was not, oh, in this private credit fund, there's a data center that Meta is going to pay their bill on.

52:26It's some other company that maybe is not going to have as high of dollar retention going forward. And so I think that the different private credit firms have a messaging problem of like what is in the fund, because it can be a lot of different stuff. And KKR clearly did a great job explaining why their particular strategy will endure for the long haul. There's other news about let's pull up.

52:51Marc Lore:Let's pull up this video quickly. Kramer Talking about our friends at semi-analysis. Oh, yeah, this is great There is a company that I regard it's the it's the absolute It's the gospel semi-analysis and I've got this guy Dylan Patel And I don't think people realize that semi-analysis is the arbiter They're like God in the semis and when they bless something it means that it's the benchmark at the best they are the most honest guys I've come across and I've always been reading them but Jensen plays really praised him at GTC so I reached out to this guy whom I regard as he's genius he's genius I can't wait to talk to him that's a promo we'll see you tonight that was a good promo 6pm Dylan says can I short myself some people some people were saying like yeah come on dude like he's complimenting you and dylan and uh and kramer obviously and i've always said that kramer having this thing where no matter what he says people say like the inverse they'll take the inverse it is the best engagement hack ever for a content creator personality because it just means anything you say you get a million impressions from people saying the opposite.

54:13Marc Lore:And the fact is, like Jim Cramer is, he's an entertainer. He's incredibly fun to listen to. And we'll be having another conversation with him in the next month or so, which I'm excited about. Let's head over to Cliffwater and Stephen Nesbitt. He brought private credit to the masses. Now the masses are fleeing. This is more context around the private credit story. Cliffwater is racing to calm investors after steep withdrawals. So Stephen Nesbitt is private credit's chief evangelist. His investors and industry are having a crisis of faith. Nick says, hope this still means my invite is open. We'd love to have you, Nick.

54:52Marc Lore:Of course, Nick. Your family. Yeah, we're always back. We're positive some. As long as we keep it positive, we can say it's, and we keep the camera roll, smiling, and we maintain that occasionally we are back. it's not always over. Very frequently we're back. Very frequently. Nick will admit that. That's true. No, we had a great conversation with him when he came on. For more than a decade, the 72-year-old championed the hottest asset class on Wall Street. His firm, Cliffwater, went from managing no money to nearly$50 billion on the back of impressive fundraising and big gains, turning Nesbitt into a billionaire.

55:29Now, many of the wealthy individuals who powered Cliffwater's rise are itching to leave as investors rethink their views on private credit. after a handful of high-profile defaults, investors are pulling so much money out of industry funds that managers are restricting withdrawals. We've talked about this many times on the show. Shares of big firms are dropping. Few are like Cliffwater, which until recently was an investor darling, but now finds itself in the hot seat. Its top executives aren't lending specialists themselves. Instead, the firm invested alongside and sometimes in other funds. So it's a fund-to-fund strategy, a feature that is now being treated as a vulnerability.

56:03Investors asked to pull the equivalent of 14 % of its biggest fund in the first quarter. Wall Street skeptics who long questioned Cliffwater's growth are now calling it a canary in the coal mine and a turducken of problems. Turducken of problems. That is such a weird phrase. Nesbit's ability to quell the fears will be a test of his funds and the industry's future.

56:25Marc Lore:Are they calling it a blue owl in the coal mine? Something like that. No, I'm just joking. Of course, Blue Owl capped private credit fund redemptions at 5 % after steep request levels. This was the other news from this morning. I think they got something like 21 % redemption request outstanding during the first quarter. So they had to cap it at 5%. Well, you know what this guy did before he started a, what,$50 billion private credit fund? He was a grave digger. I'm not kidding. Yeah. He grew up near Rochester, New York. He worked as a gravedigger in high school. He spent a quarter century at Wilshire Associates consulting for pension funds on private equity and hedge funds.

57:11Nesbitt was a soft-spoken presence in a business of outsized egos, says Greg Williamson, a longtime pension fund executive. He didn't preach like others, Williamson. He spoke about his clients' needs. In 2004, Nesbitt started Cliffwater. After the 2008-2009 financial crisis, he began recommending private credit just as banks were pulling back from lending to riskier companies, giving Blackstone, Ares, and others the opportunity to make high interest rate loans. Nesbitt became a private credit advocate. Cliffwater launched an index tracking performance. The firm shared research and Nesbitt wrote two books on the topic.

57:47In 2019, he shifted to managing money, launching Cliffwater Corporate Lending Fund, or CCLFX, with Blake and Phil Hasbrook, a then 30-year-old executive. They marketed it to wealthy individuals through independent financial advisors, the kind of clients Hasbrook worked with. It was built as an interval fund offering to buy back 5 % of its assets each quarter from investors and provide daily updates on its value. It charged lower fees than some rivals and allowed clients to avoid the messy tax filing requirements of traditional private funds. By February, the net assets totaled about$33 billion.

58:22cliffwater made 375 million in fees from the fund in the first 17 in the first 18 months that ended in september a 54 person crew researched and managed the portfolio 4 100 or so underlying loans along the way cliffwater wrangled with rivals when an executive bond powerhouse pimco when an executive at bond powerhouse pimco said the returns of private debt didn't compensate investors for its growing dangers nesbit sent investors a letter saying pimco had had a failed track record of predicting market changes. After J.B. Diamond used a cockroach analogy to warn about looming defaults, Nesbitt declared there were no cockroaches in private debt.

59:01Others criticized Cliffwaters marketing, especially when it boasted of hedge fund-like returns with minimal risk, citing industry metrics like sharp ratios and standard deviation. Critics said private loans rarely change hands, so they lack the volatility that funds face. Anyway, it goes into the reduction problem.

59:16Marc Lore:Well, speaking of Jamie Diamond, the Axios show, Jamie Diamond, Eyes Post, JP Morgan media venture. He's potentially launching a podcast. Is that what this is? Matthew Zeitland, friend of the show, former guest says the desire to post is the only force in the universe that holds a candle to compound interest. It's actually true. Everyone needs a pod. People have been talking about him running for president. A lot of presidents and media people have podcasts. Very presidential. Or go on podcasts. Well, there isn't all that much news here. But Diamond said that if he were to start a media venture, it would be something different about policy.

59:57He said, I think media is critical. Media teaches everybody. Media is the great influencer. A lot of bad policy, he believes, stems from people in the media doing a bad job of explaining issues. And so we talked about this yesterday or the day before in The Wall Street Journal. He has the new plan, the diamond plan for the American dream, wanting to lend to more small businesses. That seems more important than ever in the world where you have a one-person,$1 billion business. More people should have shots on goal. And if you think about the$20 ,000 that that gentleman was able to marshal to get the business off the ground, if you have more people that have the opportunity, that's probably a good thing.

1:00:38but we'll be tuning in when jamie diamond launches his show should we go over to apple should we i think it's time they really dominated this entire week i mean it's spring break which is perfect timing this must happen i think they probably planned they planned this years ago

1:00:53Marc Lore:they knew the 50th would be during spring break in 2026 yeah and so it'd be a good time to you know really celebrate the anniversary yeah we had a great great conversation with eddie q yesterday You can go listen to it on Apple Podcasts, which he created. But there's more reflections and stories about Apple from all over the place. Ben Thompson wrote a great retrospective. The Wall Street Journal got access to rare Apple archives that even Tim Cook hadn't seen before. It's very cool. And you think about it, it's like, that's crazy. He's the CEO, should know all the archives. But then you think about how busy his day is.

1:01:34And he probably doesn't have that much time to just like go reminisce.

1:01:38Marc Lore:Obsessed over every small decision or prototype. Yeah. There's like so much work to be done. He doesn't have that much time to go look at like the original patent for the Apple 2 or whatever he, you know, is in that archive. So the Wall Street Journal took the viewers through that in a video. We have a big read from the Financial Times here. That's very interesting. Talking about the roots of a tech revolution. So Winston Churchill called it quote the most daring and courageous act of the entire war on August 30th 1945 General Douglas MacArthur landed in at sugi southwest of Tokyo He wore aviator sunglasses a corncob pipe dangled from his lips and he was unarmed a man of war was arriving to make peace Over the next six years of Allied occupation MacArthur would demilitarize Japan and franchise women And by the way, you know, you guys know AGI will be very close when Tyler is smoking a corncob pipe himself because AGI, of course, should be able to one shot, you know, any sort of lung issues that might come from using a corncob pipe.

1:02:49Marc Lore:So some people skip sunscreen, other people indulge. In 2012, I lived in a hacker house in Sunnyvale, and the best engineer, one of the best engineers I've met in my entire life, the best engineer in the house was completely straight edge, would not drink or use caffeine or anything, but he would smoke a literal pipe. It was a very odd thing. That's so Lindy. I don't know. He was like, I just enjoy this. I would actually expect it to make a comeback. People are a little nostalgic. It's clear that vapes are maybe not something that people should be using. But the corncob pipe. Who knows? There's a chance.

1:03:38Tyler, what do you think about corncob pipes?

1:03:41Marc Lore:Yeah, I'm super Lindy. I think I'm... You're in? Yeah, I think we're probably like one or two models away, and then we'll start ripping pipe. I'm into it. I'm into it. So over the next six years, he enfranchised women, oversaw the writing of a new constitution and a decree of democracy. But first, he faced a more prosaic problem. Japan's communication industry was in such shambles that he could barely issue commands. This is MacArthur. Solving this challenge turned out to have enormous consequences, not only reshaping Japan in the 1940s, but upending global manufacturing in the 1980s, and by the 2000s, revolutionizing the way products...

1:04:21Marc Lore:Handle in the chat says Gandalf, Max. Thank you. Revolutionizing the way products would be built at Apple, a company that did not exist at the time. So Apple, which turns 50 years old on Wednesday, is arguably the world's most iconic company. It is also notoriously opaque and secretive in virtually all accounts of how Steve Jobs transformed Apple from near bankruptcy in 1997, which we talked about yesterday, with that EQ, to the world's most valuable company by his death in 2011. Product, vision, and design all get the credit. But what actually makes a$1 ,200 iPhone possible at global scale with vanishingly few defects is a manufacturing philosophy that traces back not to Silicon Valley or southern China, but to war-devastated Japan.

1:05:08Marc Lore:And it took all my energy yesterday to not use the 20 minutes we had with that EQ for kind of like a tech support session. Yeah. I'm having this issue where... You seem to be having like particularly weird iPhone issues right now. I think I need to like return the iPhone. I think you are having some weird issue because there are weird UX, UI issues that you can learn and adapt and change. And I would say that they are skill issues. Like if you can't use the camera roll effectively by now, you are needing to like. You're lost. Yeah. You should just throw off some. My issue. My issue. I don't know if any anyone in the chat has experienced this, but when I open the messages app or the phone app, I just get a blank white screen and then the app crashes like a bunch of times.

1:05:51Marc Lore:You have something weird. And then I just doesn't matter what I do. I can do a hard reset or anything like that. I wonder if we were pushing it to the limit too hard yesterday because we were on a FaceTime call together and then also watching the Artemis 2 launch on YouTube. And then I was watching it on my phone and I was streaming Geordi my screen via SharePlay. And I think that might have, the phone was really hot. I can tell you that much. I don't know. Anyway, by and large, the products are flawless and I have enjoyed my Apple journey the whole way. So in the decades after the Second World War, Japan's economy grew rapidly.

1:06:29This is a story of how ideas travel across oceans and factory floors and sometimes through a single person changing jobs. It is a story about how America invented a manufacturing philosophy, exported it to Japan, forgot it, relearned fragments of it through a handful of companies, and then re-exported the whole synthesis to Asia. The story leads us to the present moment with the U.S. spending vast sums to bring it all back, Southern India investing to be the next global tech hub, and China fighting to hold on to its manufacturing dominance. It is, above all, a story underscoring that what Apple started to build in Shenzhen, China, a quarter century ago, is not merely an assembly line.

1:07:07It is the endpoint of a multi-decade chain of civilizational knowledge transfer, a feat of enormous complexity that cannot be replicated with tax breaks or ribbon cutting in Texas. Tax breaks aren't going to be enough. You need civilizational knowledge transfer. The whole chain begins with a question. In occupied Tokyo, a 33-year-old engineer named Homer Sarasone stood before a group of Japanese executives and asked, why does any company exist? Powerful. When the telegram arrived from General MacArthur, Sarasone initially thought it was a crank a physicist by training the paratrooper turned radar engineer was working on a transcontinental

1:07:50Marc Lore:microwave relay system this is tyler the physicist by training the podcaster slash vibe coder that's true uh he he dismissed it only realizing his air when an when an indignant colonel called him back a few weeks later then he was dutifully off to tokyo for what was supposed to be a nine month stint. Sarasone's mission was to reestablish and rehabilitate the communications industry, but he found there was nobody to work with. American bombers had devastated industry and MacArthur had abolished the Zybatsu, the powerful pre-war corporate cartels. We had to start from scratch, he recounted in 1988.

1:08:28When we looked around, not only did we see no facilities, but we could find no managers. We had to find lower level people, second level managers. And I said, as of today, you're going to start up this new company and you're going to run it. The quality of manufacturing in Japan was shoddy even before the war. But as Sarasone began learning the language and immersing himself in Japanese culture, he realized the root of the problem was not technical, it was managerial. When he asked a group of employees how they might improve quality, they murmured among themselves about what an answer would please him rather than answering directly.

1:09:07sort of the opposite of like the Elon walking the floor and like, I'd want to get to first principles. They're literally like, well, what do you think he wants to hear? When will this plant be online? I think he's expecting it in June. Tell him June. Why? Yeah, this is not good. Everyone understands this now, but that was not the case in Japan in the forties and fifties, I suppose. They had been taught to be deferential, he concluded, not to question authority. So Sarasone set out to teach them a philosophy of management. Despite initial opposition from MacArthur, the need for economic stabilization meant that Sarasone got his wish.

1:09:49He and another engineer, Charles Prutzman, went off to an Osaka hotel for a month to write a textbook on industrial management. They designed a rigorous eight-week course and made it compulsory for top managers. The seminar began on the importance of quality as, quote, a guiding state of mind, a devotion and dedication. After asking, why does any company exist? Sarasone encouraged his disciples to draft a mission statement by invoking a motto from a shipyard in Newport, Rhode Island. We shall build good ships here at a profit if we can, at a loss if we must. But always good ships. It's a good line.

1:10:29That's a good line. Good line. Manufacturing, he taught, had to be considered a total system. It's disparate parts orchestrated with such repetitive precision that defects could approach zero. He inculcated in his students a sense that quality was foundational to the whole enterprise, empowering workers close to production and telling managers they needed to understand the details. Quality control is not a band-aid, Sarasone later recounted. to be effective as a control, the total process to which it is applied must be well designed to begin with. So when Sarasone left Japan in 1950, he recommended his successor be the academic W.

1:11:11Edwards Deming, an advocate of statistical process controls.

1:11:15Marc Lore:Let's give it up for statistical process. Let's also give it up for using the first initial of your first name and then your middle name. That's sort of a lost art. You don't meet a lot of people that produce themselves that way. But J. Alexander Coogan, I think that sounds pretty cool. Maybe I'll rip that at some point. That sounds very regal. Yeah, that sounds good. What would you be, Tyler? That'd be J. Tyler Cosgrove. That's good. Yeah, because Tyler's the middle name. So J. Tyler, that's good. Okay, well, J. W. Edwards Deming would prove so influential that the Union of Japanese Scientists and Engineers would...

1:11:54Marc Lore:It says, sometimes I get emotional about manufacturing. It's true. Yeah. It says they're good ships, sir. The ship's quote is so good. I'm sure it's already been printed and hung on many of the new defense tech and American manufacturing companies. But it is a good reminder to always build what is of high quality and aim for profit. Justin says, Chad is on fire today, of course. It's great. It's great to see all of you. We love having you all here. Someone asked, how are we going to celebrate? I think we're going to have lunch with the team later. I'm excited about that. Yeah! We're excited for that.

1:12:33Marc Lore:But I think the funny thing is just really our lives aren't changing. It's business as usual. Chop wood. We're going to go hang out with our families. And I'm already excited for next week. We do have a holiday tomorrow. It's Good Friday. So the New York Stock Exchange is closed. The market is closed. And so we will not be streaming tomorrow, but we will be back on Monday. But just so you're aware, don't be shocked. Oh, like this opening idea will happen and then they stop streaming. That's not what's going on. This has been on the calendar for a long time. We haven't been booking guests for months because we know that this is a holiday coming up, which we are very excited about.

1:13:13And everyone will be enjoying the long weekend. So Sarasone recommended the work to Joseph Geron, a consultant who during the war had managed a program shipping war materials to allied nations. Geron's work in Japan would go on to earn him the highest honor from Emperor Hirohito.

1:13:31Marc Lore:We don't have a ton of time. Yeah. I want to tie this into Apple. Yeah, let's go. We can just go on this insane tangent. We used to do crazy long reads where we'd spend like two hours on one New Yorker article. And then, of course, the show got much more complex. We talk about a lot more topics. But I like going through a long read. But, of course, you can pick this up in the Financial Times, which you should go and subscribe to. Alex asks, how long did the acquisition take from start to finish? I don't remember how the exact number of days, but it was incredibly quick. It was incredibly quick.

1:14:07Marc Lore:Incredibly quick. Yeah. and you know part of what enabled that was we had been having people from open ai on the show having interactions with them we had spent a hundred hours talking about open ai i didn't have like you know we didn't have a lot of questions on them yeah um and you know just just given how much time we've spent um and so that that enabled a quick process all also you You know, your long history with Sam. Yeah, it's like you meet your require like years before. Yeah, in this case, 13 years before. That's crazy. Okay, where do we go to close this out? It was the early 1990s. We're flashing forward.

1:14:50And Jobs was a half a decade into leading Next, the startup that he had founded after being ousted from Apple. Next first product, a Cube State worst station from costing$6 ,500 had already been a much hyped flop. The team's goal had been to make a computer their friends could afford. By the time it shipped, the joke goes, the only friends that could afford to buy it were Steve's. Japanese quality ideas had been all the rage for a decade. The superiority of Japanese production had become clear in March 1980 when Richard W. Anderson, a Hewlett Packard executive, famously discovered that the best Japanese memory chips performed 1 ,000 % better than their American equivalents at initial inspection and 50 % better over time.

1:15:33The so-called Anderson bombshell made HP start to obsess over quality. Its Japanese joint venture, Hewlett-Packard, Yokogawa, won the Deming Prize in 1982 and became the foundation for rigorous standards applied across the whole company. HP aspired to improve quality tenfold within a decade. And when that looked to be failing, it adopted a Japanese step-by-step approach to quality known as Plan Do Check Act. Let's see where else.

1:16:01Marc Lore:Jobs was clearly taken by the ideas of quality. Silicon Valley was beginning to import from Japan. Having set out to build a computer company that would create better products, he commissioned an automated factory in Fremont inspired by the plants of Japanese electronics manufacturer Alps Electric. um anyways we can uh we can continue but um but um we have our next we have our next guest mark lord joining in just a few minutes uh let's run through some of the timeline posts yeah this was cool taylor uh taylor johnson said nothing gets me hyped like a re-accelerating top line is the definition of my company right sharing uh sharing or or just a founder mode company But really both, you can see, this is data from Sakura, Sakura shared.

1:16:52Marc Lore:It's so funny, people calling me right now. I know, I know. If you have called me, if you have sent me a text message. We are live. I'm sorry. I will get back to you after the show. Feeling the love. But anyways, incredible chart here. you can see Plaid had ended. In 2023, they went from 308 to 390 million of ARR and then jumped up to over half a billion in 2025. So Zach and the team are on an absolute tear. You know who else is on an absolute tear? Taylor Lorenz. Apparently she spends 17 hours a day in screen time. I guess that means phone, but maybe phone and computer. She has to spend a lot of time because she's defending big technology.

1:17:39Yes, exactly.

1:17:41Marc Lore:There's some incredible quotes in this story. I think she said if she could put the screen in her brain, she would. That's wild. That's very unexpected. I mean, she's got to get set up with the Neuralink team. Get me a true timeline merchant. Yes, yes. Well, yeah, Extremely Online is the name of her book. And she certainly lives her brand. uh eliezer yudikowski had an interesting post here he said today i learned that gemini claude and chachi bt but not grok are told that today he was referring to march 1st april fool's day yesterday is march 32nd because if you tell llms it's april 1st the conditional text predictions downstream become less i less reliable for obvious training data set reasoning the model's like okay Okay, understanding April 1st.

1:18:30Marc Lore:Yeah. How do I understand? It would be very, very confusing. I mean, we were confused all yesterday because you see so many news announcements go out and you're like, is this a joke? There is news happening. Yeah, but if you're told it's April 32nd or March 32nd, you should be more confused. You're going to be like, that's clearly wrong. Yeah, I wonder how real this is. Yeah, you would think they would just keep it March 30th or something. Yeah, I don't know. We'll see. We'll have to ask some people. Rune says, Because the AI doc reminded me mostly of Kony 2012 documentary, Slacktivism, selling the feeling of we need to do something as a product, oddly centering the filmmaker.

1:19:05When is the embargo lifting for the AI doc? Because I did see it with Tyler and a bunch of other folks from the team and enjoyed it, but I would love to talk about it. I believe we're going to have the - I think it already came out. It came out? Yeah. March 27th. Yeah. Yeah, it was interesting. The biggest gap between the AI doc and the press tour that the creator is on now is that he seems like he doesn't believe AI is real at all, I guess. He was very worried about doom in the movie. Comes away sort of like, oh, there's an optimistic scenario here. It's a very nice ending. But then he came away being like, ah, this stuff is not real, which is a very funny conclusion because he's not really asking about any of the financials or economics or business applications.

1:19:50He's having a much more philosophical debate and then came away with a financial conclusion. I don't know. It's interesting.

1:19:55Marc Lore:Yeah. Google has released Gemma 4, the best open models in the world for their respective sizes. Demis has excited to launch Gemma 4, available in four sizes that can be fine-tuned for your specific tasks. 31 billion dense for great raw performance, 26B MOE for low latency, and effective 2B and 4B for edge device use. Happy building. Excited to see what people do with it. They're available now under the Apache license in Google AI Studio or on Hugging Face and some other platforms. So massive launch. Very exciting. What else do we got? There's more on the Mercure leak. Very unfortunate. Gary Tan says, incredible amount of state-of-the-art training data is now just available to China thanks to the Mercore leak.

1:20:47Marc Lore:Every major lab, billions and billions of value in a major national security issue. I'm just feeling, I'm feeling, you know, the national security issue is one thing. I'm feeling really concerned for individuals that not only gave PII as part of onboarding, but maybe now there's like live video of them tied to that PII. So it feels like there's some real deep fake risk. I'm very glad that we have Adam Myers from CrowdStrike coming on today to explain the surface area, what the trends are in cybersecurity, because things do feel like they are ramping up significantly. Also just very odd that there's now a leak of data that could be used to RL models.

1:21:35There's new open source models. There's also the leak of the of the cloud code harness. And so you if you piece all these together, you get pretty close to the frontier. And that's something that we're certainly going to have to like contend with the fact that there's every time that there's a leak, you probably shrink the gap between the frontier and the open source community a few by maybe a few months or something as they catch up, even if they're doing it like sort of above board. Anyway, Lewis says -

1:22:07Marc Lore:He found a public company with 99 % of revenue coming from one customer. You've heard of the one employee$1 billion company. Now we found the one customer$1 billion company. I actually don't know how big this company is, but we need to figure out what company this is. This is TSS Inc. Okay. The ticker is TSSI. And they said, we derive a substantial majority of our revenues from a single OEM customer. Revenues from this customer comprised approximately 99, 99, and 96 % of our total revenues for the year ending December 31st, 2025, 2024, and 2023. So they actually had more revenue diversity in 2023.

1:22:48And then the revenue concentration increased over the last two years. Although we provide services across multiple business units and divisions of this OEM and have entered into a long-term AI RAC integration agreement that includes minimum monthly payments, our overall financial performance remains highly dependent on the continuation and scope of this relationship.

1:23:09Marc Lore:Yeah, and they're trading very reasonably. They're at something like a$240 million run rate as of Q4, at a market cap of like$366 million. Okay, so yeah. It's price seems to be priced in. Yes. Buko says, imagine being a software company with like 250 ,000 customers, 1 billion of revenue growing 20 % and the market says you're worth 3.5 billion. And then Rigatoni Computing has generously five customers and basically no revenue is worth 4.5 billion. It's a cold world. I don't know. Yeah. I mean, this is the reality of like working on a sci-fi technology. is that if it works, the value is really, really big.

1:23:52Marc Lore:Yeah. Essentially. Prepared remark says, but those five customers might go to six, so you have to get in before that. What is, what's going on with Scott Wiener's based act? Did you see this? Probably the most radical bill ever to degrade tech products. It bans Amazon Prime, stops iPhones from having FaceTime, strips travel, shopping local and AI results out of Google search results. This feels like a stunt almost. I don't know. It must be a comment or something else. This feels like something that's like sort of being misinterpreted almost. I don't know. Adam sent out an update to the chamber progress, his group.

1:24:31Marc Lore:He said the base act is likely the most radical proposal to regulate and direct technology product design ever advanced in California. legislature. The bill dictates how core products must function from search results to app stores, to e-commerce, marketplaces. Well, we have our next guest in the waiting room. Let's bring in Mark Lohr from Wonder. How are you doing, Mark? Good to meet you. Hey, how are you doing, guys? Thank you so much for taking the time. Great to be here. Sorry we couldn't give you much warning or our news. It's kind of a wild, wild day over here. I just saw it. Congratulations.

1:25:08Thank you. I'm sure we can get into M &A war stories and whatnot, whatever you want to talk about. But why don't you, since it's the first time in the show, sort of take us back in history and give us a little background on yourself for the viewers. Yeah, sure. I started my career in investment banking. And then in the late 90s, did a first startup, sold it to Topps, the baseball card company. No way. Then started diapers.com. Okay. And then sold that to Amazon in 2011. Yeah. To$550 million. And then worked inside Amazon for a couple of years. Started jet.com, which is another e-commerce site.

1:25:49And two years later, sold that to Walmart for$3 billion. And then became the CEO. That's why they call him the LeBron James of e-commerce. It's on the wiki. became the CEO of Walmart's e-commerce business and did that for about four and a half years yeah um yeah now I'm the founder and CEO of Wonder yeah a food tech startup maybe take us back to diapers.com uh I'd love to know like how you were thinking about that business when you started it was what was there a thesis that sort of vertical focused e-commerce was was going to be a trend? What was the ecosystem like? What was the competitive landscape where VC is saying like,

1:26:32Marc Lore:Or did you just snap up a great domain and think I got to make some money? Yeah, yeah, yeah. How much was just organic versus strategic? The way it started, I was just searching on Google what search terms people were searching for a lot and doing it at night, searching, searching, searching, would tell you how many times it was searched and I saw. Oh, so they didn't even have, so did they even have like Google trends at that point or you just had to search and they had they would show you it was alexa i think it was called alexa maybe or something you can type in a search term and it would tell you how many times it was searched on google yeah and uh diapers i remember was 200 000 times a month and so i went online and i saw the price of diapers was like 10 more expensive and i thought i mean i had a baby at the time it was a pain to go get diapers why aren't diapers delivered like everything else like books and all this other stuff and i thought yeah diapers.com you know and that was actually 1-800 diapers started because i couldn't afford the diapers.com domain yeah wait so the 1-800 there must have been a moment where like the 1-800 was like more expensive than the.com though and then did it flip at some point oh yeah it was very like when i when i was there very cheap it was like you know i don't know just tens of thousands whereas the diapers.com was like a half a million or something.

1:27:48So, um, what was it like building an e-commerce website back then? Like we, I mean, we just, we just, uh, read a story about a one person or I guess two people company, a two person company that's at a billion dollars in revenue. And that feels unthinkable, but like, what was the team? Like, were you racking servers? Did it, was there a cloud? We were racking servers. We were racking servers. Yeah. It wasn't there. Nothing was in the cloud. It was just, you know, eventually transitioned to it. But yeah, no, it was servers in the server closet. And yeah, we had, you know, designers, you know, designing the website and building it out.

1:28:27And like it was old school. And then, I mean, how were you thinking about fulfillment, vertical integration, what you want to do, what you don't want to do early on with that journey? I mean, we had very little capital. So everything was, you know, hand to mouth. that was we would sell a diaper online a box of diapers online for let's say 40 and then we'd go to the wholesale club and buy it for 42 and ship it to the customer no way so it was like that kind of thing so we're actually losing money on every on every box we sold yeah and then uh and then what what was the uh the integration with amazon like at the time where was amazon as a business Where was the vision?

1:29:09What was the thesis and what you saw was going on at Amazon? Why was it an exciting opportunity? Yeah, so I mean, it was still early days. Amazon was actually crushing it in 2005 when we started Divers.com, but they were mass. They weren't focused. Any one category, it wasn't a great experience. Like if you're like me, a new parent going onto Amazon and wanting to buy your stuff for your baby, you're all over the place. It wasn't a great experience. the diapers were actually more expensive on Amazon than in the store they just weren't getting the whole diaper thing and saw an angle to create a specialty site that focused just on that vertical everything the parents want, everything they need all in one spot and it was really working.

1:29:55We also were able to ship the diapers and all this stuff out of the same fulfillment center with two day delivery which wasn't being done on Amazon at the time in that category. Is that more important in that particular category because parents need diapers for their kids? It's more urgent? You might wait for a book or you might wait for a TV or something? Yeah, exactly. You realize I'm running out of diapers. I need them tomorrow. That sort of thing. Definitely. I also think you're buying a lot of different things and it's expensive to ship them from multiple warehouses. So wanting to get them all in the same box like getting the wipes, the diapers, the baby formula, bottles and things, like everything in the same box because they're low margin.

1:30:43And so it's too expensive to ship them separately. But we were also the first ones to bring in Kiva Robotics into the warehouse. Oh, no way. Amazon ultimately bought Kiva Robotics. We were the first ones to use robotics in the warehouse. And again, all in the name of the margins are low. We have to figure out how to automate. We create this software that told the people in the warehouse exactly how to put everything in the box so we can get all the things in the smallest box possible. Because FedEx a lot of times would charge you for the box size. And so we tried to like, we had 23 different box sizes.

1:31:20So like we were very advanced when it came to the logistics and trying to pull costs out of the system.

1:31:27Marc Lore:yeah and um what what uh what elements of the internet boom like what are the ways in which the internet boom feels similar to the ai boom and and how does it feel different um i think it feels very i think it feels very similar um i mean certainly um you know just there's people on both sides of the fence this is good this is bad what's going to do like a lot of uncertainty. It felt the same way. We were looking back on how people, all the fears people that had about the internet are almost mirrored one-to-one with AI. Like every single one kind of matches up. You know, the job displacement, fears, Y2K, things like that.

1:32:12Marc Lore:It's pretty remarkable how humans feared the internet in the exact same way as... Humans fear everything. You go back to when the trains, okay? I don't know if you guys know this. I wasn't alive. When trains were a thing, they strongly advise people not to go on trains because at that speed, they don't know the long-term effects it'll have on your brain. Whoa. And they told pregnant women, do not go on a train if you're pregnant. Whoa. Because for the same reason. Like any industrial advancement that's ever taking place in history, there's always an incredible amount of fear. Yeah, it's fascinating.

1:32:55Can you tell me about the process of the idea for Jet.com? I remember that process happening, and it felt like, okay, at this point, with Diapers.com, there's a niche, there's a landing zone. It makes so much sense. It's complementary to Amazon. Jet.com felt much more like, okay, this is a direct competitor. Was that the correct framing? Were you thinking about that? Were you just thinking there needed to be more options in the market that you could differentiate? What was your thought process going into launching Jet.com? Yeah, I think there's just a massive market, and there was no number two to Amazon.

1:33:31And we had an angle. You know, Amazon at the time was, you know, shipping stuff from multiple warehouses. It was very inefficient from a logistics standpoint. They were burning a lot of money, but they had no competition, so they could do it. And the idea was very simple. let's empower people and teach them how to shop smarter so that they can save money and so we built this smart card technology where you know when you started adding items into your basket it would reduce the prices of the items that could be fulfilled from the same warehouse in the same box sure as an incentive to get people to save money because if you shipped an item from two different warehouses, you had to pay, you know, at least$5 per shipment.

1:34:15Yeah. And if you shipped it in the, from the same warehouse, the marginal cost of ship might be 15 cents. And so that was the idea of, of teaching people how to shop smarter, uh, to save money. And, um, yeah, what was the fundraising environment like throughout that, uh, throughout that journey?

1:34:32Marc Lore:I doubt you've had a hard fundraise at the same time. It's like, you know, you're going up against to Amazon during this time. I mean, I tried to raise money in 2001, 2008, 15, some really tough years. I've been through it all. I've been raising money now for 30 years. I've done probably as many venture capital pitches as anyone. You know, probably over a thousand pitches easily now over the career. But the fundraising has changed. In the early days of Divergence.com, if you were to raise$100 million, that was like a huge deal. Yeah. and and then today you know it's raising billions tens of billions you know yeah it's very different environment today but back then it was as much harder to get private capital what drew you to food we had we had travis kalanick on uh a few weeks ago and and uh we're sort of joking around is like you do if you've done food feels like uh the hardest possible you know massive opportunity but it's like probably the hardest food tech, probably one of the hardest categories.

1:35:38Marc Lore:And so if you've mastered, you know, at scale logistics and these, you know, incredibly, you know, capital intensive things like e-commerce or rideshare, it feels like the final frontier, an incredible challenge. But what what brought you what drew you into it? I mean, first of all, I think the margins in food is so much better than e-commerce. and I felt like the restaurant sector was ripe for technological disruption. Restaurants haven't changed fundamentally in 100 years. They're still capital intensive, labor heavy, difficult to scale and they're places. We wanted to challenge that and ask the question, what if restaurants weren't just ideas?

1:36:20What if you can build a restaurant and scale it across a network like software without any incremental capital or incremental labor with robotics in the back end. And if we're able to do that, then we're able to bring, you know, make great food more accessible. We can bring restaurants to places that currently don't have access, a ton to the day that don't have access, and at price points that are really unfathomable today. That was really the thinking there. And, you know, we managed to today in 2 ,500 square feet have 25 unique restaurants across 20 different types of cuisines, everything from a high-end Bobby Flay steak to Jose Andreas to barbecue, burgers, Chinese, Mexican, Italian, Middle Eastern fried chicken, pizza, all in one 2 ,500-square-foot kitchen with no gas, all electric, no open flames, with very lightly trained labor.

1:37:16So it's very systematized. And we see a future where we'll have 1 ,000 or more unique restaurants operating out of the same 2 ,500-square-foot kitchen. Yeah. And so wild. Have you read any of those like old stories about the automats in 1950s? Have you heard of this? It's like this it's like this restaurant. I think it was in New York. And basically there was like a wall of cubbies where the food would be prepared in the back. And then you would just sort of open the cubby and take your food. And there was no interaction. And they were like, see, they were heralded as like the future in sci fi.

1:37:55but they never really took off in like the 50s, I guess. And I've always wondered about like, how important is the online interaction, like the delivery point? Because there's been a number of attempts to make like robotic restaurants work, but it feels like if people are going to a restaurant, they want a particular experience. But if they're ordering food online, they're okay with a different experience. I'm wondering how how separate those are even though we think of them as like it's the same name It's you know this pizza place in person or this piece of place in in on delivery They're actually maybe much more different experiences and like and like experiences delivery like what the value prop is Yeah, we're I mean we're focused.

1:38:44It's primarily delivery first So 70 % of revenues delivery about 25 % is pickup. Okay, and then less than 5 % is sit down So we have 10 to 20 seats in the front. So it looks like a high-end, fast casual in the front, but it's a pretty small front of house. But being vertically integrated, so we bought the RubHub, so we own the delivery. We own, and we also own all the restaurants. So we own all the restaurants. We do the cooking. We built all the technology. And so the vertical integration combined with a very tight delivery radius allows us to offer an incredible experience, like faster, more on time, hotter food, great quality.

1:39:20And of course, everyone in the family can order from a different restaurant and it all gets cooked and delivered at the same time. And we can do that in ex-urban and even rural areas where restaurants can't typically go. And that's one of the advantages of the model. And then recently we just added drone delivery in New Jersey. And that'll be a big part of next year's. Is that flying or road going with wheels? I'm sorry, flying. Flying.

1:39:47Marc Lore:And who did you guys partner or did you fully verticalize? Uh, it's fully, no, no, we're partnered with a couple of different, uh, drone companies to do it, but, but we're alive. You can order from drone, uh, in New Jersey, uh, in this one location and, uh, what are the next year? What are the challenges with doing prepared food delivery via drone? Like I've seen some of these videos where a package will fall, you know, 10 feet out of the sky. And, and obviously that's not going to work if you're ordering, ordering a nice meal. No, it comes down with a tether. It comes down with a tether and very gently puts it on the ground.

1:40:21But the advantage is no tips. It's more on time. You can service a bigger area. And then you could also deliver to, if you're on a boat, on a lake, on a beach, a field, you could be camping. So the idea of being able to deliver it to the point as opposed to somebody's residence is really cool. And it's really at an inflection point now. I know we've been talking drones for 10 years, but when we go to Texas next year, we fully expect half of our deliveries to be done via drone.

1:40:49Marc Lore:Wow. We've had Keller from Zipline on a number of times, and we've been feeling the acceleration with every interview we have. How do you... Yeah, one of the things that... Go for it. Oh, sorry. Go ahead. um how have you approached everything on the supply chain side actually because i imagine some of that is out of your control but that you know you need uh you're you're heavily relying on those inputs no matter how good your whole technology stack and process is the food's gotta arrive uh at at each of these kitchens um what is that that is absolute most important i'm glad you called that out we have you know 40 culinary engineers on staff uh we just hired Victoria, who's the head of global supply chain at Cisco Foods.

1:41:36So we have a 700 ingredient library. We source everything from a number of different purveyors. It brings it to our distribution center. And then from a distribution center every day, we replenish every location. Today, we have 118 locations open. Next year, we'll have 400. So we're growing very fast.

1:41:56Marc Lore:Wow. But the food quality is the most important. I'm assuming it would never have worked to work with third-party food distributors. You can have them come into the central hub, but you need to actually understand what your timeline is. You want to have specialists. You have everything from pizza dough to barbecue to savit steaks to vet. Everything comes from a different supplier, in some cases a co-manufacturer. But yeah, the food quality is most important. But the 700 ingredient library is fixed. All the equipment is fixed. Think of that as like a data center. Once you have the data center piping in the ground, anybody can create a restaurant using the platform and launch it across instantaneously across all locations.

1:42:44So at the end of this year, we're launching Wonder Create, where anybody in the world with just an AI prompt can create their own restaurant and launch it across all Wonder locations. for$10 a month. We think it's going to fundamentally change how people think about restaurant creation.

1:43:02Marc Lore:So yeah, walk me through this. I come in, I want to prompt, I start prompting or even just typing out what I want to make. And then your guys are using robotics. You can just make it on the fly. What is the lead time? Can you actually open up the floodgates here? How limited will it be? In December, you'll be able to go to Wonder Create homepage and say, create me a fast, casual Mexican concept for Gen Z. And that's it. Hit create. Then AI will brand your restaurant, name it, give you a couple options, do all the images, do all the recipes, write the descriptions, price everything, do all the health information, create your entire restaurant in under a minute.

1:43:46You can decide to publish it in all Wonder locations for$10 a month instantaneously. And now you're live to potentially 20 million people next year. You can push it on DoorDash, Uber, and Grubbub if you want as well. And you own the restaurant. You price it. You own it. It's live. And if you're an influencer, you have...

1:44:06Marc Lore:So are you doing a rev share back with the creator? Or how does that work? Yeah, you just basically, the creator would pay us for food and pay us for robotic time. So it would plug in. Let's say you created a fast casual Mexican concept. It would plug into our infinite bowl machine that we acquired from sweet cream. The infinite bowl machine would make your bowl according to your recipe. And then we would actually print your packaging on demand to match your logo. So you're fully in business. You don't need to do anything. You can be live in a matter of minutes with your new restaurant. And then we're next year launching an automated sauce machine, the infinite sauce machine, which has 130 raw ingredients.

1:44:46and can make 80 % of all sauce recipes on the internet on demand, 500 sauces an hour. So you can also create your own sauce recipe, your own dressing, your own whatever sauce you want on these bowls that are made from the infinite bowl machine. So that's sort of all. It's all going to launch at the end of this year. IBM. I mean, you're describing a very broad vision for the type of food that can be delivered and made. walk me through the level of robotic automation because you walk into a place that toasts bagels and they have a machine, it's sort of robotic that will like move the bagels down and toast it or the pizza goes through the oven and then we see pitches all the time from humanoid robotics companies where it's operating a toaster that's not automated at all but the robot is moving and it's fully humanoid.

1:45:40What is actually useful? Where are we in the deployment of this? what robotic form factors are actually moving the needle and driving value? So where we are today, we have conveyors in every location. We have, again, no open flames, all electric cooking platform. When the item gets onto the conveyor, it goes down to the expo area. It gets auto-scanned, no human, robotic arm, picks it off and puts it into the right bagging lane. So that is the extent of the robotics today. But the robotic machine, the infinite bowl machine we bought from Sweetgreen is live in 32 Sweetgreens. It makes all the salads and all the bowls without any human intervention.

1:46:23The only thing you have to do is put the ingredients in the machine. But the bowl is beautifully made. It turns the bowl, neatly puts the ingredients in the bowl, and there's no human labor. You're pulling out 25 points of labor on making those bowls. And the bowl machine can do 500 bowls an hour. It can do$13 million in revenue and 300 square feet. Wow. Incredible piece of machinery. This great team from Spice Robotics, they have 34 MIT engineers up in Boston, are now working for us, and they're building the sauce machine. The sauce machine will go live early next year. We're working on an infinite beverage machine that could make basically any drink in a coffee shop or a cold brew concept, all the foaming, layering, blending.

1:47:09That machine will also go live next year, so we're layering on the automation. And then in the middle of next year, we're launching an automatic retrieval system where all it has a cold storage, a frozen and ambient, and everything's stored in there. And so if you order wings, it'll robotically pick the wings and send them to the fryer. And then the person just puts them in the fryer. In the future, the frying will be automated. But that's kind of like where we are today, where we're going next year.

1:47:38Marc Lore:But we envision. Yeah, go for it. What we envision in the not-too-distant future, being able to generate$20 million of revenue at 2 ,500 square feet with only 15 people operating that$20 million in revenue. So your labor cost is a small percentage. Your rent is a tiny percentage. And we're able to make a 50 % four-wall margin, which we're going to take that extra margin and put a lot of it back into price. So we'd expect prices to come down and be deflationary over time. and is consistent with our mission of just making great food more accessible. Price point. Yeah, has food ever been deflationary?

1:48:15Marc Lore:It just feels like forever. It just goes up and up and up. I wanted to get your thoughts on humanoid robotics, specifically in a restaurant context. I'm guessing I already know what you think, but what's your take? Yeah, so we don't use any humanoid robots, so I think we like to think about. And now I mean more like diffusion. Do you think that do you think a humanoid form factor will ever be productive in a commercial kitchen? Or are these kind of like bespoke systems like you've created going to dominate? I mean, I think definitely the short term what we're doing is sort of like first principle thinking on this stuff.

1:48:59Like the humanoid robots going into a kitchen that exists as, oh, there's the fryer. This is what you do. You put the fries in the basket, the basket down, the basket up, the basket over. we like to rethink and think why do you need to do that that way why can't you just create a fryer that has a conveyor and the conveyor goes through the fryer and fries it along the way so you don't need to put anything in a basket and take it out of a basket like that sort of that sort of thinking i do think there's some things that can't be automated it'll be a long time before we could ever automate it like assembling a burger or rolling a burrito if a humanoid is able to roll burrito or assemble a burger, we'd be very interested in that and we would look at it.

1:49:38It doesn't seem like anything that's going to happen in the very short term, but I'm always open to know that anything you think is not possible today could be possible.

1:49:46Marc Lore:Yeah, we were asking a humanoid founder about what we're calling Diet Coke Bench. How effectively can a humanoid open a Diet Coke? It's simple, but probably not too easy. I have one last question. Yeah, we really didn't like this. Sorry. Okay, good. Just my last question is, I'm wondering about your vision for the future of like restaurant brands generally, because if like, will we get to a point where I can go and prompt or design a specific meal? And if I want buffalo wings, put on pizza and fold it into a calzone, deep fry, like But as long as I'm using the tools that are within one facility, will I be able to just create something that's essentially a one-of-one meal that is to my exact specifications and actually be direct to consumer?

1:50:41Yes, you'll be able to create your own restaurant, your own recipes, the way you want, and go direct to consumer. So we envision a future where there are not only influencers creating restaurants, but every college student can have their own cold brew concept. A high school student can have their own salad or Mexican concept. We envision in the not-too-distant future, there are tens of thousands of salad concepts in the U.S., tens of thousands of fast casual. And the market will be very long-tailed, very fragmented. It won't be consolidated the way it is today.

1:51:10Marc Lore:It's fascinating. You're basically taking franchising and bringing it online. And franchising, in a way, has been so successful because you're giving people a system and a proven product. but in this case, you're not asking for them to take on hundreds of thousands or millions of dollars of debt to set up these. $10 a month. That's all it is. It's$10 a month. Well, thank you. Incredible. What are you hiring for right now? I mean, across the board, we're going to a lot of open positions in robotics and engineering. so you know we've got over a thousand engineers on staff but we can't seem to to get enough so um yeah if you're listening reach out amazing well thank you so great to meet you mark for taking the time to come talk to us you guys yeah i've been uh followed your career yeah same here i was building an e-commerce company in 2012 and i was like yeah this guy's the lebron james so thank you for everything you've done for the industry great to see you guys we'll have to create a tbpn kitchen yeah i'd love it goodbye uh there was something that you wanted to pull up jordy what should we pull up here from rat king our friend over at the new york times yes tech reporter what does he say uh he wrote he wrote a profile on us yes he was able to get uh a screenshot actually from our contract okay with open ai uh and he says interesting as part of negotiation with OpenAI, TBPN has a commitment to editorial independence written into the contract, which states the following.

1:52:46Marc Lore:TBPN retains full control over its daily programming, editorial decisions, guest selection, and production schedule. TBPN will continue to host a broad range of voices and perspectives, which we shared earlier. Yeah, open stage to talk about whatever you're building in technology and AI. TBPN independently determines its external appearances and commentary. OpenAI will not control TBPN's planning materials or working documents. OpenAI will not provide direction on TBPN's editorial calendar. OpenAI will not influence who TBPN books or what topics it covers. Again, we just talk about so many things.

1:53:26Marc Lore:Oftentimes we're talking about topics like on the fly. There's breaking news. This was incredibly important to us. It's also just more fun to have the directors be the hosts as well. Yeah. So that we're not talking heads. We just say whatever we want. Yeah, people have always asked us, how do you guys decide who comes on the show or decide what to talk about? It's whatever we're interested in. The honest truth is it's whatever we're interested in. It's always been that way. It's super important. It's always been important because we have to sit here and talk for three hours every day. It's got to be interesting to us.

1:54:01It's got to be interesting to us. Or else it's brutal.

1:54:04Marc Lore:and yeah, OpenAI will not materially alter, discontinue or rebrand TBPN. OpenAI does not have rights to TBPN host likeness. Yeah, we specifically have it set up where they can't even train models on TBPN. Can't train it on the chat. They can't train on anything TBPN related. I'm sure other companies train on TBPN already in one way or another. But yeah, this was incredibly important to both teams And I think we got it to a really good place. And thanks to Mike Isaac for covering it. Will, let's bring in our next guest, Adam Myers from CrowdStrike. He is the head of counter adversary operations at CrowdStrike and the host of the Adversary Universe podcast.

1:54:47Adam, welcome to the show. How are you doing?

1:54:49Marc Lore:Adversary's worst nightmare. The adversary's worst nightmare. Yeah, thanks for having me. Congrats on the acquisition, too. I was reading about that earlier. Amazing. Awesome. Yeah. Yeah. And thanks for hopping on a crazy week. I imagine it's an extremely crazy week for you. Can you sort of level set for us? Is this a really, really weird week generally? Yeah, it's interesting because when we had last time we had George on or maybe the time before I was talking, it's like we haven't had big AI related security breaches to date. And not everything that's happened this week has been directly tied to AI, but it feels like everything is heating up.

1:55:29We're seeing more of it. There's been a couple in the last two weeks of these supply chain attacks. And I think what's clear is that, especially with the use of cloud code and some of the different agentic solutions that are out there, people don't know necessarily what these things are downloading, what they're loading on their systems. I know you guys have talked about OpenClaud before and things like that. And it's just the supply chain for software has been an issue for a while. And I think it's just escalating. We've seen so many of these over the last week. We've seen Team PCP doing a few of these, and then this most recent one, which was a big one, tied back to North Korea.

1:56:09North Korea.

1:56:10Marc Lore:Yeah, two things are happening. People aren't reading code. They don't understand the ways in which they're actually building products. And then there's also, it feels like 100 or 1 ,000 times more software just being created, too. And the combination of those things is what's driving this. Maybe I've been doing a plan with it a lot and it's great. You know, you can't even see it as it's running and it's just pulling down NPM libraries and God knows what else from pipey and stuff like that. So it's just, you know, who knows what's out there. And the other thing is you install these libraries one day and then they become evil overnight.

1:56:44And, you know, you have to go back and review everything that you're using. Yeah. Can you give us a supply chain attack 101? Some people might be familiar with SolarWinds. That was the last one that really broke through to the mainstream. But just how does this work? What's the anatomy of a supply chain attack? And then we'll go into how to fight it and why they're on the rise in particular. Sure, yeah. So generally, supply chain attacks are going to impact the software supply chain. And so in SolarWinds, I actually worked that one. So I can tell you that one was pretty special because they got into the CICD environment and were able to actually affect the building of the software so that it was done in a way that nobody really could tell that that had happened.

1:57:26What we're seeing in these new supply chain attacks is that they're targeting the developers themselves, many of whom probably listen to the show. And so it's important to pay attention that they, you know, they go after your credentials, they fish you, they get access to your logins to things like NPM and pipey and then abuse that. And there's a number of ways they can do this with Git tags where they can actually, you know, hide the code and then redirect Git to a Git tag. But effectively what happens for the layman is that you have software that is depended on by enterprise applications, by open source stacks.

1:58:08And Axios was downloaded something like 100 ,000 times per week. So that gives you a sense that these things are being used extremely frequently. And people don't look at them. They don't analyze that code. They don't analyze that library. They trust it came from NPM. It must be good. The communities looked at it. But those things can be updated at a moment's notice. And you don't know what that latest update is unless you review the code yourself. Yeah. Okay. So on reviewing the code yourself, we were debating this. The two sides that we were kicking around was maybe the reaction to this is that, sure, we're not going to write as much code going forward.

1:58:46We know that the models are good at writing code. But there will be a lot more code reading that goes on. and people will just demand that if you're hired as a software engineer that you read every line, you work through the dependencies, you are in charge of the code that you ship, regardless of whether or not it was generated by an LLM or not. The other side of it was sort of like maybe the answer to bad AI code is more good AI code and that we will be doing more AI code reviews and that we can actually throw more AI at this problem. I imagine the solution is somewhere in the middle, but how are you wrestling with those two paths forward?

1:59:26Well, I think what we're looking at is how do we secure it, absent of is somebody reviewing the code or not? And so there's things that you can look for at the endpoint, at the security fabric. We kind of see ourselves as the operating system of security at CrowdStrike. So being able to see that the AI agent is writing code and it loads a library, and inside that library, you know, use this Axios example, there's a chunk of it that's base64 encoded, and it's being, you know, decoded and instantiated. That's suspicious behavior, right? Just if I say that out loud, I'm like, hey, you guys are going to pull down a bunch of base64 code and then decode it and run it.

2:00:03You'd be like, no, we're not. So, you know, I think those things that we can see at the endpoint at the build time, we can actually look for that and say, okay, this is suspicious behavior. Let's flag this and pay attention to what's happening with this particular thing. Because to your point, you could demand everybody read all the AI slop that they're generating, but the reality is that even if they do that, they're not looking at all the libraries that they're using. How many libraries does a standard build have? I've been vibe coding for a while, and who knows what libraries? You run it, and that's the beauty of it.

2:00:39You run it, it spits out a bunch of stuff, and then it works. The reality is nobody's looking at those code bases until something like this happens. Yeah. What does the future of security products, CrowdStrike, I can just imagine a world where more people need security products for open source projects, for their personal projects, for their personal life. Like if you have someone whose day job is important, but then they're vibe coding their, you know, their their family calendar at home. But all of a sudden that creates a vulnerability where you can get their phone number and then, you know, SIM swap or something like that.

2:01:19There's just so many different people.

2:01:21Marc Lore:People have had like relatively high trust with software for the last 10 years. right it's like oh it's a google login and yeah only only this year have i started to be like really question yeah different apps and things like that that people are making because there's a tendency if you're working in tech to like want to try the new thing totally and now when something pops up it's actually kind of a headwind yeah for startups because i think we're probably headed into a lower trust era yeah yeah i think that's part of it i also would point to the people that are building these libraries. We've seen this with Chrome plugins, too, and browser plugins.

2:02:00You need to have MFA, multi-factor authentication, on all of your authentication for pushing code out to these library bases because somebody comes along, phishes the developer, convinces them to enter password for drive or some sort of cloud storage or something like that. They take those credentials, and then they use that to log in and change the code base. So the developers really need to be focused on securing their own accounts and identity. And we've seen, you know, we released this global threat report a few weeks ago. Identity, you know, for a long time, endpoint. And when CrowdStrike, I've been with CrowdStrike since we started back in 2011.

2:02:41Endpoint was kind of where the bad stuff happened. It really is. What a good run.

2:02:49Marc Lore:what what did what advice are you giving to developers today that are you know i think everyone that my advice would not be freak out and you know slam your computer into the table we're going back to pen and paper yeah we're going back to pen and paper but like what what are the you know two two factors an obvious one any any other like advice that somebody should basically implement immediately? If everybody secured their identities, my job would be a lot easier. The number one thing that we've seen is identity attacks. I mean, two years ago, we saw the voice-based phishing. So threat actors calling up the help desk and being like, hey, this is Jordy and I can't log into my account.

2:03:34Sorry, Jordy, I'm just picking on you there. But hey, I can't log into my account. And they're like, all right, well, who's your supervisor? What's your work location? You answer those two questions and they reset the cred for you. And that happens to the point, hundreds of percentages of increase that we've seen that occurring from threat actors. So, you know, one of the ways to get around that from a defense perspective, multi-factor authentication. Don't use SMS, by the way, for multi-factor authentication, because that's a whole other thing. because I can sim swap you or I can, you know, and, you know, if you have a backup for your multi-factor authentication as your Gmail, well, I'll target your Gmail and then I'll intercept that key.

2:04:15So having smart secondary factors and using pass key and things like that really important. And then using different, you know, I tell this to everybody and, you know, even if I'm talking to a bunch of school kids, like, Hey, use different passwords for everything. And I love nothing more than going into a room with people. I'm like, all right, raise your hand if you don't use the same password for every account. And they all raise their hands. And I'm like, look around. If your hand is up, you're a liar because everybody does password reuse. Yeah, of course. Give us the pitch for the podcast.

2:04:47Give us a summary. Tell people where to find it. Yeah, it's the Adversary Universe podcast. It's myself and Christian Rodriguez, who's our field CTO of the Americas. And just like you guys, we like to have a good time, just kind of back and forth banter, bringing guests occasionally, but we really dive into the adversaries. So in this case, with the Axios incident, we're attributing that to Stardust Chalima, which is a North Korean group. We've been tracking them since 2015. And one of the things that they...

2:05:14Marc Lore:So, and then Lapsus, I don't know if I'm saying that correctly, was connected to the Light LLM attack, but do I have that correct? That was a different group. Well, yeah, that's that's kind of team PCP. I think claim claim credit for that. But, you know, there's they're they're actually fighting. You can watch this in some of their channels. The two groups are kind of, you know, talking. They're like, I did it. No, I did it. Well, I think the lapses guys or shiny hunters was like, this is why we hack them. Right. So, you know, they hacked team PCP. So there's a lot of back and forth there. But, you know, this one was North Korea, who has been doing, you know, they stole last year,$1.46 billion in just one attack against a cryptocurrency company.

2:06:00So they've been stealing cryptocurrency since 2016 at billions of dollars is kind of the current total. And when they do a supply chain attack like this, they want to go after developers who are doing blockchain and kind of DAOs and stuff like that because they know that if they can get into there, then they can actually steal cryptocurrency and use that to buy things for their weapons program.

2:06:24Marc Lore:Wow. Yeah, a lot more straightforward than acquiring data and then trying to auction it off, which we've seen with the Rancor incident. If you get the coins. Well, yeah, some of those are interesting because they actually, they were deploying stealers, right, to steal credentials to then go after other targets. And there's a whole ecosystem on the underground where, you know, you could go in, you could buy creds to an organization. And if you have creds to an organization and you log in, you can go straight into their single sign-on, you can get into Microsoft SharePoint or any of their data stores, and you're off and running.

2:06:58And there's really very little that organizations can do to prevent that. It's a legitimate user logging in with a legitimate credential, so it becomes very difficult for them. Well, thank you so much for taking the time to come break it down for us. Have a great weekend.

2:07:11Marc Lore:Yeah, this was insightful. Come on more often. Yeah, this would be great. Anytime, guys. My pleasure. Have a good one. Good to have you. Goodbye. Hi. Our next guest is Jeremy Allaire from Circle. He's live in person in the TVP and Ultradome with us. We're very excited to talk to him. Jeremy, good to see you. How are you doing? I'm good. Take us through the announcement. We were covering it over the past few days around maybe reset on where the business is today, and then we'll go into the quantum story. Yeah, absolutely. I mean, look, obviously, we went public last year. Yeah. And, you know, as we went public, we talked.

2:07:49Thank you. There's a soundboard in person, too. Don't worry. Can I hit that now? Yeah. And, you know, at that time, like, we were talking a lot about, like, our stablecoin network. Yeah. Which is, like, rooted in USDC. And, but really, you know, over the last year, we've really widened out. Yeah. And so we are a broader platform company now, and we have an operating system that we've been building called Arc, and that's getting ready to go main net, as we say. And then we've been kind of building out abstractions around kind of payments and sort of making payments more simple and mainstream, and then just building a lot of infrastructure and tools for developers and stuff.

2:08:30So the platform has really broadened. But, I mean, if you look at, you know, where we are now, like, USDC has become the most widely transacted digital currency in the world, bigger than anything else, basically, and has overtaken Tether in terms of transaction volume. Interesting. And, you know, we're seeing new uptake. We're seeing new uptake from, you know, major companies now, you know, whether it's like a fintech like Ramp that just launched like whole treasury management with it. Or it's, you know, a global bank like JP Morgan that's now has used it to like sell digital bonds. And so it's all over the place, which is pretty cool.

2:09:11And then, you know, one of the things that we saw really start to pick up and we'll talk about, I think, hopefully today, too, is effectively like this explosion in people building agents. Yeah. And agents consuming services and needing ways to kind of pay for all of that and the technology and standards to kind of do agent-to-agent economic transactions, financial transactions. And USDC has played a big role in that. And that's emerging and to me is like one of the most interesting things happening in the agentic economy. And yeah, USDC and our stablecoin network is like 99 % of the transactions happening now at that layer.

2:09:53Marc Lore:Yeah, so how are you tracking adoption? What kinds of use cases are you expecting to be the first to really kind of go more mainstream? This is something that people have been talking about for years now. And as agents have become massively more popular, it feels like now could really be the moment. But what are you tracking most closely? Yeah, I mean, look, there are a bunch of things. I think if you think about the agent stack as it's emerging, you're clearly seeing a lot of companies that have services, have software, have other things. And they're basically like, OK, I need to kind of convert what I do into an agent consumable system.

2:10:36So I want to have an MCP server. I want to have, you know, a CLI interface. So all these kinds of things and this sort of discoverability. And then you need ways to basically enable agents to like onboard and transact really, really quickly and not go through the kind of onboarding that you would typically deal with. And so that's where we see this first wave is basically agents consuming services that are now kind of agent ready and have been kind of structured that way. And so there's like a registry of now, you know, thousands of services that are basically becoming what's called X402 enabled, which is a key standard we're involved with.

2:11:17It's a big announcement actually today about that. And so that's like the first is sort of like getting that connected. But to me, the really interesting and more explosive kind of growth potential is really, you know, agents that are providing services to other agents themselves. And so if you think about, like, I'm building an agent that's, you know, extremely good at a particular form of, let's say, medical analysis. Or I'm building an agent that's extremely good at reviewing resumes. or whatever that is, this sort of agent as a service is, again, the way that people are talking about this is, like in those models, you're basically paying for, you know, intelligence.

2:12:01And intelligence is going to be priced in dollars, but also as consumable tokens. And you need, at that point, like if you look at like what's the cost of intelligence for X number of tokens with a given model, it could be$0.05, it could be$1, it could be$0.35, it could be more, it could be something long-running, it could be$100 or whatever. And so once you start having agents contracting with each other and needing that medium of exchange to settle that, that's where I see this as explosive because you get this compounding effect of agents.

2:12:35Marc Lore:Yeah, we were talking earlier, there's a guy in the New York Times today that built$1.8 billion. Yeah, yeah, I know. I mean, I've been reading I watched the show before I came in here. And what's notable is he's using a bunch of different intelligence providers. But if you're just hopping between different LLMs all the time, it's not great. You're copy and pasting, and it's not unified. And so you would imagine that over time you get to actual aggregators, and you're like, hey, we need to tap into a different model for this. 100%. Yeah. And an agent marketplace and these registries of agents, And obviously, because this is all machine to machine, it's super discoverable.

2:13:16And what's behind any given agent, which is just kind of consumed through some kind of underlying CLI or whatever API, right? Like, it doesn't matter what the underlying model is behind it. It's just like, there's a function I need, and it's here. And so, yeah, I think it's also the kind of compensation, fees, transactions around agent-to-agent stuff is obviously has that like huge exponential potential. I think what gets more interesting is when you really start to think about structured economic relationships. Like most economic relationships are not just like you have a thing, I'm paying you for the thing, right?

2:13:59It's like a labor contract. There might be a lot involved in it or a service contract. There may be many, many stipulations that are part of it. And like a trade agreement between two companies, there's like all kinds of stuff. And so you can imagine like more elaborate forms of contracts. And those need to be machine written contracts, machine enforced contracts and machine validated and audited and kind of have dispute resolution against all of those. and like economic operating systems that are built on blockchain tech stacks are like the way that will happen. And that's more of a view of like, what is the agentic economy?

2:14:39What is the economic system? And actually these, these, these new companies that are, um, you know, the one person billion dollar startups, et cetera, like all of that, um, phenomenon in some ways, like these are, these are emergent corporate forms, right, that are using a lot of AI labor. But over time, like, I think even the nature of corporate forms will change and we'll have more digitally native corporate forms that most of the substance is actually just software execution. And, like, there's humans involved, and maybe the legal system will always require there's a person that's there. But I think we're going to face a lot of very hard legal questions around sort of entity substance.

2:15:27Like what is the substance of the entity? What is the entity? But I think a lot of this, there's a lot of material here that we have to work with to ultimately create these sort of fully digitally native forms that are composing and interacting with contracts and economic systems.

2:15:46Marc Lore:How are you guys working with companies today? The beauty of USCC is anybody can kind of get the power of it by just permissionless. Just tap in. But I know you guys have invested in companies in the past, partner with a ton of them directly, like power people. I'm sure you're getting an insane amount of real pitches and then also people pitching you with open claw, things like that. Oh, my God, yeah. I mean, look, I think like so a couple things. I think within this AI space, we're making investments, like venture investments in different types of startups that are doing interesting things. We have builder programs with funding and stuff like that.

2:16:31So we have builder programs in this space. But a lot of what we're doing is just basically making sure everything that we have, like all of the infrastructure for wallets, the infrastructure for money, the infrastructure for creating smart contracts, like all the stuff that goes into this, that all of that is just consumable and discoverable by either AI development tools or by AI agents themselves. And and then letting letting things happen. But like stepping back and a little bit away from the AI specific side of it. But, yeah, I mean, the range of companies now getting involved and plugging into this is really broad.

2:17:10And that's exciting. This is definitely kind of going into a mainstream phase. I imagine AI companies are ahead of the curve there in terms of adoption and turning on agentic payments because you go to an agent and it needs to use some resource. That's the logical pattern versus…

2:17:31Marc Lore:There was a post, I forget who put it out earlier this week, something to the effect of people, AI people are more excited about crypto right now than crypto people. Yeah, yeah, yeah. Well, you know, I've always thought of these as sort of highly complimentary. And even going back 13 years when we were kind of starting Circle, the thing that got me really excited was this idea of programmable money, that we were going to have self-running software machines that would intermediate economic stuff on the internet. And that was the promise of the tech 13 years ago. You couldn't do it. It wasn't possible.

2:18:09It didn't exist. But it was like, as an internet technologist, that was really exciting to me. I'm like, wow. If you can actually have these sort of self-running software machines that are auditable and can run. Now, I didn't have a concept of generative AI. Yeah. But definitely it was like, okay, programmable money that is going to be like a completely new form of utility for money that's never existed. And so that was very exciting. And so I think in many ways, like, you know, a crypto infrastructure, crypto as a tech stack is very, very purpose built for the agentic world. Like they really are hand in glove and will work very, very closely together.

2:18:52And so, you know, trustless intermediation is like what happens in AI, right? You know, whether it's of data or of...

2:19:03Marc Lore:Yeah, it's interesting to think about how many business transactions, like even if there's a contract, they're so trust dependent. Yes. And stables can play a role in helping that evolve because you can say, you know, basically if, you know, you can escrow funds. Yeah. You said you can make it programmable. You can create contracts. Yeah. And these can be very, very elaborate. Yeah, that's always been the thing. Will people, like people like to buy from people, but part of that is because you want to know, okay, who's my counterparty? Yeah. Do they have a good reputation? Yeah. Do I trust that even if we have a contract, they'll follow it and I don't have to get into some, you know, legal.

2:19:40Marc Lore:Yeah. Yeah. I remember the pitch for machine-to-machine payments. Yes. I think it was around like Ethereum, maybe 2015. Yeah, yeah. So this was like an early thing. Yeah, and it sort of made sense. But it was very hard to imagine writing all the code to do all that. But now the code is just running itself. Exactly. Yeah, it makes a ton of sense. I'm wondering what you think holds about the current economic philosophies or the economic rules of the road just in business and what might be different as we move towards a world of decentralized, agentic payments. Like, I imagine network effects still hold.

2:20:18I imagine that if you create value, you create the network, you can capture value through some transaction percentage and everyone will be happy in the ecosystem. How do you think about the economics? Yeah, I mean, look, I think there are a bunch of different pieces here, right? So Circle is a money issuer. Like we issue digital dollars. We issue digital euros. EURC is the largest digital euro stable coin. We issue digital treasury products. We have the largest tokenized money market, basically. So we're a money issuer. And so if we can build networks that have wide utility and protocols for those monies that we issue that are super, super widely used and provide a great utility, you can permissionlessly plug in, et cetera, and we can grow that, the monetary base alone can be massive.

2:21:09So we can imagine that these kinds of digital dollars running on these networks, there'll be trillions of dollars of these as opposed to hundreds of billions of them now. And so that itself is significant. And the economics are very significant from that. But I think, ultimately, I think as the infrastructure becomes kind of more widely utilized for real-world transactions, I mean, whatever that means, right? Yeah, we absolutely have an opportunity to participate in the velocity of those transactions and the economics around that. But, you know, we had this philosophy when we started Circle that, you know, over the long run, that the marginal cost of storing and moving value would go to zero and that the business model of charging fees for payments would collapse.

2:22:05Interesting. And, you know, I think over the long run, that is still going to be true. Actually zero or like 0.00001 % which is so big? Yeah, I mean, actually zero to most of the consumers of it because the costs underneath it are going to be so low. It's like, you know, you don't get charged for your WhatsApp audio call, right? They're like, you know, we're going to deal with it. We'll make it up other ways. Yeah, it's too cheap to meet up. But like we just announced something called Circle Nano Payments. It's a module for agents to basically be able to have a stored balance of digital dollars, of tokenized dollars, and be able to transact them to different wallets on different blockchains.

2:22:49And we've gotten it to the point where we can actually have transactions priced at one one millionth of a penny per transaction. And you're like, well, who would ever need that? Well, if there's like, yeah.

2:23:06Marc Lore:Well, here's an example. So yesterday we had Eddie Q on from Apple. He was talking about the early days of the App Store and how when you bought a song, they would effectively kind of open a cart and they would hope that you would buy multiple songs. Oh, yeah, I remember. They would hope you'd buy multiple songs in a 24-hour period before they actually like process the payment because otherwise you're giving a quarter. Because it's so expensive for its action. Yeah, a quarter of its action. $0.25 to$0.30, yeah. Yeah. Yeah. And so like that right there is insane. Like the fact that the Apple, one of the biggest consumer tech companies in the world had to use this kind of weird workaround just to not give 25 % of every transaction, taking money away from, of course, like the creators and the record labels and Apple, just like pure kind of rent extraction.

2:23:50Yeah. It's like payment systems, like we're never designed, we're not designed for a machine and an AI economy. They weren't. And we, We even see this now where because the cost to transact USDC has gone to sort of fractions of a cent in general, you can do that. Like the velocity of transactions has exploded. And you see that in the growth and the year-over-year growth. It's like tens of trillions of dollars of transactions on a monetary base of around, you know. We like trillions. Congrats on saying the biggest number. On scale, what are you tracking right now to understand where we are in the adoption of agentic payments?

2:24:33Because I imagine it is happening. It's small. It is. Are you doing a percentage of GDP thing? No. Percentage of your transaction base? Just growth month over month? What's the correct internal KPI? There's a couple ways to look at it right now. So one is, there are specifically standards for agentic payments. And so we are a co-developer of the X402 standard. Actually, the X402 Foundation just got merged into the Linux Foundation. A lot of great companies, Stripe, Coinbase, who is a pioneer in this, Circle. This has been a proposal in HTTP for like 25 years, right? It's this old thing. It's so crazy.

2:25:18It's like payment request model. But now, if you look at X402 version 2, it's building out into a pretty rich vernacular that's needed for the needs of agentic payments. So one way to look at it is, what's X402 traffic? And you can look at X402 traffic. You can track it. You can look at the payloads. You can see what the transactions are. And that's in the, like, it's been growing. It's in the hundreds of millions. Okay. Okay, so quite small from a starting base. But I think if you look at a leading indicator, which is the number of these agents or services or endpoints that are X402 enabled has been growing very rapidly.

2:25:57And we, as well as companies like Stripe and Coinbase and others, we're basically shipping stuff so that it's super butt simple to have X402 in front of whatever it is you're doing or in front and behind. So you could be a consumer and a service provider. And so I think that, like, 2026, we're going to see this huge growth in that. And then the consumption models. The leaked anthropic code actually had all of these references to X-402 in it. So, you know, it's clearly, like, even in foundation models, becoming something that they're going, like, hey, this is going to be, like, a common pattern.

2:26:34But, you know, so that's one way to look. And then the other is like, it's hard to measure, which is like people are using AI to create bots, like these polymarket trading bots or these hyperliquid trading bots or like all kinds of stuff. And those aren't using X402. And so that's like other traffic, but it's AI driven transaction volume. And so it's hard to measure.

2:26:54Marc Lore:We got to talk quantum. Quantum. Yeah. Quantum. I mean, I'm assuming you were aware of the risk of quantum to encryption overall. No, it didn't. No, I mean, so very aware. And I'm saying like aware predating starting circle, right? Like, because this is just a risk for overall encryption. Yeah, and this has been, you know, crypto is cryptography, which is math. And, you know, all this is about who can compute these, you know, keys and ciphers and all that stuff. So, of course, it's always been a background debate, always been a theoretical background debate. as we've been building up our own operating system, our own crypto-based operating system arc, we have been saying, okay, we're doing this kind of clean room from the start.

2:27:40It's a new infrastructure. We have a chance to do post-quantum readiness from the start. Whereas existing networks and there's this Google paper and all that jazz and so on, there's a migration and there's risk and all that kind of stuff. And so we basically said, okay. Yeah, just a massive coordination problem. Huge coordination problem. So we basically said, look, we have phenomenal PhD crypto people in the company. And so we made this a key priority. And so we actually just announced today our whole post-quantum roadmap. And one of the most powerful things is that when ARK goes mainnet in the not too distant future, it will have post-quantum signatures there from the start.

2:28:26So if you're issuing assets and you're doing transactions, you can actually use post-quantum methods from the start. I think we'll be the very first blockchain network in the world to have that there, which is key. Now, there's multiple layers to this. And so the roadmap also details, like, okay, how do the validators on the network get themselves to be quantum resilient and resistant and so on? So there's multiple layers that have to happen over time. The base layer, which is the most important, which is the fundamental mechanism for how you sign transactions, how transactions are published to the network, and how assets get defined.

2:29:07Like, that's post-quantum. There's a lot of detail. We put it out in our documentation today. But, you know, we've been thinking about it a long time. And, you know, it's sort of interesting because we were gearing up to launch this. And then this Google DeepMind or whatever it was came out. And we're like, okay, well, time to launch.

2:29:25Marc Lore:Well, it's an interesting challenge for the industry because everyone needs to get their house in order and like create a plan. But then it's also like the entire industry needs to rally because even if you guys have your. Oh, yeah. No, it cascades across all these ecosystems and like USDC runs on 32 blockchain networks. Right. And so like, you know, we are part of the whole ecosystem, obviously. Yeah. And then, you know, of course, like if from a de novo perspective, people building new stuff. I mean, one thing to remember is like basically like crypto infrastructure blockchains have been an early adopter phenomenon until very, very recently.

2:29:59And like the actual usage at a global scale in the real economy, like, you know, 98 percent of that's still ahead of us. So we're still very early in all this.

2:30:08Marc Lore:Last question for me. How did you process the 2028 intelligence crisis? That paper, Petrini, because some of it was basically making the argument. There's a lot of kind of rent seeking in the economy and AI can potentially make a lot of transactions a lot more efficient. That's kind of what you're betting the company on. They had a bunch of specific stuff about stable coins and USDC and all that. I mean, look, we are – so I read that and there's a lot of really interesting things in there and broadly like even Circle itself. Yeah, the interesting criticism too is like the criticism was like, okay, this is like sci-fi.

2:30:49Marc Lore:But when you look back at sci-fi throughout history, there's a lot of sci-fi that is kind of come to life. I mean, look, I mean, the fundamentals of completely restructuring what transaction costs are and basically models where much of the work that's conducted in the real economy is work conducted by AI or by intelligences. and that that that work is economically going to be measured out and metered out and executed using these new digital currency forms. Like I believe that very, very deeply. And I don't think anyone knows how big that will be. But it's certainly, you know, again, like total processed volume, which is like this number that like the card networks and others use, like these numbers are going to be look so small in comparison.

2:31:40Sure. The unit economics are going to be very, very different. So I agree with a number of those conclusions. I'm not a hedge fund. I don't have any positions in any of these companies and all that. But I think it's real and it's happening. And yeah, we're trying to build for that future. Amazing. I'm glad we have you working on it. Thank you so much for taking the time. Thank you, guys. To come on down to the TV. Thank you. I really appreciate you. Our next guest is in the waiting room. and we will bring in Justin Levine from Shepard into the TBP and Ultradome. Kicks off our lightning round. Jeremy, how you doing?

2:32:20Oh, sorry. What's going on? Justin, I'm still updating the Chiron. Congrats, big day.

2:32:25Marc Lore:Thank you. Thank you. Honor to have you here on this day. Sorry we couldn't give you a heads up, but it's all good. Well, we are not here to talk about us. We're here to talk about you. Please introduce yourself and the company. yeah uh justin levine co-founder and ceo of shepherd we're a ai native insurance business we focus on large-scale industries like construction and and renewable energy yeah uh how how are you thinking about the growth of the construction data center industry broadly like what's the headline number that you're quoting because i imagine you have a tam and it's growing but how do you actually sense make about the what to expect over the next couple years as you build the business?

2:33:06I mean, we're in a massive construction super cycle specific to the AI infrastructure layer. I think this year alone, there's something like$400 billion of infrastructure spend just related to either data center assets or the energy assets that need to support those data centers. There's not enough energy on the grid, so we now have to build the energy assets next to the data centers in order to make it all operational. So, you know, from a TAM perspective, We are we're seeing a massive inflection in terms of construction this year alone. 80 billion dollars of data centers were started. So 80 billion dollars of new data center starts just happened in the last 12 months.

2:33:45So we're just seeing a massive wave. And it's kind of ironic for us. We we built Shepard on this thesis that AI was really ripe to productize this service of underwriting and doing underwriting differently. And then ironically, a lot of these companies became our customers. Because this industry is really the driving force of construction right now.

2:34:07Marc Lore:Okay, so who are all the different parties that you guys are helping ensure? It's AI companies, Neolabs, private credit funds, who calls you? Yeah, it's a mix of all the above. So everything from hyperscalers to the labs companies to contractors that are building, obviously, these assets. Our portfolio is actually more diverse than just AI-specific companies. So our typical business is a large-scale developer of assets that are either energy assets or healthcare assets, hospitals, all sorts of different, I'd say, large commercial physical assets. And then we insure the contractors as well. So a big portion of what we do is identifying the best contractors who are using the best technologies out in the field and then rewarding them with innovative pricing.

2:34:59Marc Lore:Yeah. So what what what are what are they insuring? Right. Is it is it weather, inclement weather, fire damage? There's so many different. What can you what can you insure versus like what is just to what could you insure? But it's just too hard to price. You have to go to, you know, too high to make it feasible. There's a lot of different hazards, obviously, related to the construction cycle. So there's obviously just the cost of the building during the construction phase. So the property risk that is associated with, again, a large data center campus might be$20 billion of both construction material as well as equipment that's being installed.

2:35:40So there's an enormous property component and just insuring the actual property during the construction phase is a huge strain on the insurance ecosystem. Then there's a liability perspective. So, you know, the builders that are creating these assets are they are on the hook for mistakes, for defects, for potential injuries to people or buildings and other other assets that are around these job sites. So depending on sort of like the product offering, you're obviously focused on a specific hazard. We really focus on like the driving hazard. So casualty is our key product offering. And again, that's going to be like property damage or bodily injury to people in and around the building of these large scale assets.

2:36:28But then we also do property. So we also do the materials and the cost of construction while it's being built.

2:36:36Marc Lore:What are your relationships like on the actual insurance side? Who are you actually, who's putting up all of the capital and taking on the risk? Yeah, so we partner with large scale insurance balance sheets, almost similar to like a neobank that's going to have, you know, sort of a traditional bank behind it, you know, providing all the financial stability and rating and the necessary requirements to fulfill whatever the contractual risk that is there between a lender and a builder. so we kind of we stand up under what's called an mga model which is that shepherd really owns all of the upfront marketing sales underwriting policy servicing everything from a submission coming in the door the underwriting that happens the making the decision making around what risks we want to we want to ensure and then ultimately servicing those accounts once they become policy holders underneath our brand uh take us through the rays we got a 42 million dollar rounds

2:37:40and and yeah what came in special about this uh like the actual investors on this round yeah so so this is our series b um it was led uh by uh one of the largest insurance companies in the world so intact uh private capital red led the round that's the venture arm of intact insurance um intact's a 30 billion dollar global insurer um i i would say what's special about it is the signal You know, I think a lot of the things that we're trying to do differently, specific to the automation of underwriting, making underwriting significantly faster and more efficient. A large insurer, a large commercial insurer like Intact pairing with us and partnering on this, I think signals that they see that the world of underwriting is about to change dramatically, just like other large industries, whether it's legal or accounting or all these other areas where we've seen a huge impact from AI.

2:38:31And brokerage and underwriting are kind of, you know, right in that wheelhouse of where services businesses can change dramatically. So Intact had partnered with us at the A. They were a capacity provider for us. So they are one of those balance sheets that sits behind us. They've seen what we've been doing. They see the innovation that we're bringing to the industry. And so they actually preempted the round and doubled down into the B. Last question for me. How human is the process of underwriting? because I feel like you could probably build a model, square footage, what are the assets, where is it?

2:39:06You know, you probably have underwriting models, but there's probably still, like, who are you hiring? How will the human capital side of the business scale? Yeah, totally. So, you know, commercial insurance, you know, differently from personal lines is definitely more complex. It's more data heavy, more intensive, and there's a lot more sort of nuance and decision-making that happens at the underwriter level. so we actually we equate what we're doing on the underwriting side to a little bit of a similar view of the self-driving kind of autonomous vehicles type of roadmap so we've mapped our own level one through level five and the idea is like today most commercial insurance companies have underwriters with two hands on the wheel and they're you know literally doing every single thing what we're trying to do is get to a place where the underwriter at shepherd is much more of an orchestrator.

2:39:56They are managing a portfolio of 200 accounts rather than just 20 or 15 on a monthly basis. And so underwriters do play a significant role. We hire underwriters who think of themselves as entrepreneurs who want to build a portfolio. They want to orchestrate and manage that portfolio. What they don't want to do is data entry and some of the, I'd say, like process heavy and high friction elements of traditional insurance underwriting today. Well, thank you so much for coming on and breaking it down for us. Congratulations on the round and good luck. Thank you guys. We'll talk to you soon.

2:40:30Marc Lore:Ripping. Have a good one. Congrats. Congrats. See you guys. Talk soon. Our next guest is Gaurav Mishra from Mirage, formerly Captions App. I found out about the company through at Randall in the previous round. We'll bring in Gaurav to the TPPN Ultra Dome. Gaurav, how are you doing? What's happening? Good. Thanks for having me. Excited to be here. Yeah. Thank you. Yeah, it's been almost a year. Crazy day for both of us. Yeah, you must have been in the first how many guests? I'm a huge user of this app. I love this app. I've used it for every meme that I send you. Every time there's a video with captions over it, I'm always downloading it on my phone, screen recording, putting it in captions, and then I'll even use it to transcribe videos.

2:41:11Marc Lore:John is a magician. He can make basically feature-length memes that usually just get sent to the team chat. But yeah, for those who don't know the apps, but also just where the business is broadly and where you're expanding to, sort of give us the lay of the land. Yeah. I mean, as you mentioned, you know, we transitioned our name. So we kind of went from Captions to Mirage. But, you know, the app remains to be Captions, the company that we renamed basically. And part of that is like we're building out a product suite. So, you know, it makes sense to sort of have a different company name for that.

2:41:48Now, obviously, we raised about$75 million recently. We announced that, which is exciting. Congratulations. Thank you. I love it. Love that. Yay! So, with that funding, a lot of what we're going to do is expansion, market expansion. So, we are very excited about the magic moment the product already has. and you talked about it a little bit, right? And when we can bring the product in front of a new customer, they're very likely to convert, very likely to pay, and very likely to retain, right? And that's kind of what our investors saw. That's kind of why this new investment came in. Now, we want to bring this to as many people as possible around the world as quickly as possible.

2:42:33You know, historically, we've been sort of marketing in like five to seven countries. That's kind of where we've been, US-centric, like a little bit in Europe and stuff. But now we want to go really broad, you know, especially Asia is a big focus region. So that's kind of the plan of where we're going. How do you think about just feature requests and the scope of what you can do with video plus AI? I don't want to narrow you down to mobile, but I think you've executed very well on mobile. But there's so many different features, and a lot of them get broken up into different models, different APIs.

2:43:08I feel like there's so much opportunity to bring all that together. At the same time, you're going up against meta and edits and CapCut and Divide Dance. There are some big players, but there's also a lot of opportunity. So how do you see the landscape of things that you can do? I'll scroll reels all the time and see some crazy edit and be like, I know that they jumped to Blender for that, or they used a particular AI model for that, or they're in After Effects for that. And I want to recreate it, but I don't want to necessarily sit down for five hours and remember how to use Blender. I mean, you actually hit the nail on the head because this is exactly the way we're thinking about how our product roadmap goes, right?

2:43:49I think if you look at the AI video space, there's tons of companies, there's a lot of activity happening right now, right? But our approach is a bit different than everybody else. Like we're not only about generating video, even though that's a big component of what we do, right? And we use all kinds of models out there. We have our own models as well, right? For audio and video generation. But it's really about the assembly of the video into something that actually makes sense, right? Something that, a narrative that, you know, ultimately can work for you, whether you're a small business, you're selling something, whatever it is, you can tell your story, sell your product, market yourself or your business using our product because we can assemble the right video for you.

2:44:25Now, here's the interesting part. What we've noticed is that a lot of our customers, especially the ones that are retaining for long periods of time, it's not just about fully generated video. Like that's like, it's the thing. If you're making a Super Bowl commercial short, like go and generate the entire thing, like it's going to look awesome right but there's actually a much much larger market for like a mix of content like think about like our average customer like they're like a plumber right or maybe an electrician or like a nail salon right yeah they want to actually show like their actual nail salon right what does it actually look like as opposed to a generated nail salon of some sort right and they want to merge that with some generated footage some cool shots like maybe you know a founder video like talking uh you know perfectly about whatever whatever they want to do and put it all together into something that they can give to customers, market on their website, wherever they're putting it out there, social media.

2:45:11So it's really about the connection of generated and non-generated together and then putting the narrative together. That's the intelligence layer. So that's really our focus. How do you think about entry points into video creativity? There's something that can be sort of intimidating about an empty text box. at the same time, it is like the universal interface. It's the most broad possible UI. When do you want a user to start with an existing video or an image versus start with a blank text box? Yeah. So I actually, so even today, over 50 % of our users start with some sort of media, an image, a video, you know, a set of images, a set of videos, audio, music, all kinds of things, Right.

2:45:58So over 50 percent of people start with media, which actually is like very different than most other like a lot of like pure video generators. People start with text. They start with a prompt and ours is flipped. It's like media first prompt is attached to it. Right. Now, I think the way we think that's going to evolve for us is that, you know, it's going to go more and more media first. That's the crazy part, right? I think people expect it to go more prompt first, but I think our angle is a little bit more media first using real footage. I actually think that there's another extreme here. Think about like, let's say you want to make a documentary, right?

2:46:35You probably want like zero generated footage in a documentary, right? It's like all documented reality, right?

2:46:41Marc Lore:Yeah, I've been seeing that. I've been lately, like I'll try to pull up a video on YouTube with my son of like a cheetah. But I have to find videos that are older than like four years because I don't want a bunch of AI cheetahs. I want the real thing. I'm trying to see like documentary footage. So you got to find like the vintage National Geographic stuff. Wow. Yeah. Soon that's going to be lost. We may never find it again. No, we will always have the archives. We'll be going deep into the vintage. Exactly. Yeah. Yeah. It's real. What do you, what are the plans to scale the business over the next year or two with the new fundraising?

2:47:24Yeah, I mean, so a lot of it is going to be market expansion, right? We talked about that. I think secondarily, it's use case expansion, right? So like you talked about like we started off with something really simple, adding captions to your videos. That was the first thing we ever did, right? Straightforward, but quite useful, right? And then from there, you realize it's really about like automatically editing your video, right? like automatically adding edits, automatically doing stuff that people do manually normally, right? And that's kind of how we built our roadmap, right? So the future for us is expanding use cases on what can we automatically edit, right?

2:47:55Today, it might be, okay, it's talking head videos, it's real estate videos, things like that, right? Tomorrow, it could be product launch videos, right? Tomorrow could be like all kinds of other things that we don't do today, maybe perfectly, right? So that's kind of where things are going. And with that comes like new sets of users, you know, expanded use cases, obviously, it's like expanding the talent for product, So that's kind of where we're going, both on product expansion, geo expansion. That's really the focus. Well, thank you so much for coming on and breaking it down for us. Congratulations.

2:48:25And we'll talk to you soon.

2:48:26Marc Lore:Great to have you back on. Let's not let it be a year. Always a great time. Yes. And by the way, congrats to you guys. Thank you. Have a great weekend. Yeah, great to see you. We'll talk to you soon. Goodbye. All right. And without further ado, we have our last guest at the show, Philip from StarCloud. putting more and more data centers in space. We're very excited to talk to Philip. He is, I believe, in the waiting room. We'll bring him into the TV panel at UltraDome when he's ready. Google released a feature called Inbox Zero. They say Inbox Zero is a thing of the past. Introducing AI Inbox.

2:48:59Marc Lore:Cut through your email clutter with smart prioritization and daily personalized briefings. Isaac says, Brands sending five emails a day to drive higher and higher. ROI won't work anymore. RIP email. Yeah. It'll be interesting to see how this channel does. We had a friend that sent us a bunch of emails that he's been getting, and they look like spam recruiter emails, but they all come from newly set up Gmail accounts, and they're not quite human sounding, so we can clock them, but basically he's getting an ever-increasing amount of poaching emails. Usually, you'll need another filter on the other side.

2:49:41recruiters are going to have to start going standing outside offices and just waiting for people to leave yeah yeah uh well we have uh i believe we have philip from star cloud in the waiting room uh we got some vertical video we'll make it work we are gonna have a quick chat how you doing what's going on hey guys apologies for the vertical video we're all good it's great we got back-to-back sounds okay so i can't turn this off second time on the show uh kick us off with the fundraising. Tell us what happened. Yeah, so we just raised $170 million at a$1.1 billion valuation. Thank you so much. That is a massive round.

2:50:24Marc Lore:The video is vertical because the valuation is going vertical. Yes, never. But yeah, this works great. So yeah, talk to us about what you'll be building with that new round. Is this hiring? Are you going to be in the R &D phase? Do you have manufacturing partners lined up? Are you going to be buying heavy equipment to make stuff? What does the next couple of years look like for you? Yeah, so we're building the third satellite that's very similar to it's going to fit on the Starship PES Dispenser form factor. So it's this constellation we've just filed with the FCC for 88 ,000 satellites. And so what this allows us to do is build a manufacturing line for that.

2:51:0788 ,000 satellites, by the way, would allow us to deploy about 20 gigawatts of compute capacity, which is on the order of$100 billion worth of CapEx spend. So, yeah, we're ramping up. Too low, too low. And that was a wee bit swinging.

2:51:24Marc Lore:The height. Yeah, absolutely wild. What kind of partners are you working with? I can imagine basically every chip company wants to have some exposure to what will be an emerging category. How are you approaching that side? Yeah, 100%. So we're working with NVIDIA on, we just announced, and GTC Jensen announced it, this Space Rubin chip. So it's like a slightly mass-optimized, slightly radiation-tolerant, and then slightly thermally-optimized chip, which can be used for space inference. We've been working with them on that for a while because we've got an enormous amount of data now from the StarCloud One satellite, which has the first NVIDIA H100 on board.

2:52:10And so we're using all of that telemetry to inform the design choices, both on the Rubin ship and also on our second and third satellite.

2:52:17Marc Lore:Are you running workloads, like mini workloads on the first satellite? What does testing look like? Yeah, exactly. So we trained a model. So we trained Andre Carpathie's Nano GPT on there. we did the first type out inference so we're running a version of Gemini called Gemma on the spacecraft and then we're now also doing some more like military kind of useful workloads which is running inference on SAR data so synthetic aperture radar which is basically where companies, where satellites collect a 3D image of the world underneath them and then they can use that to detect things like a vessel entering you know, entering your territory or other things uh how is it how is the discussion going how is the research and development going around uh heat dissipation that that that felt like the key uh will it work what are the economics are we are we expecting to be like on track for some solution do you feel like the solutions in sight on some curve or maybe even already here how is that going yeah yeah and by the way this I want to give you guys credit for shifting the eye-written window on this.

2:53:28Because I think you guys were the first to do a segment where you were like, do you remember the bit where you were like, imagine before they had cameras on spacecraft and people would be like, wow, that's so crazy. Oh, yeah. From then on, people were like, okay, maybe, yeah, this probably actually could actually be a thing. Yeah. But, yeah, so we've been working on this radiator for two years now. It's a solved problem in the sense of the physics of it. Okay. What remains to be done is the manufacturing challenge, so making this thing cheap and light. And also small enough to fit into that, what do you call it, like the PEZ dispenser?

2:54:04Exactly, the PEZ dispenser form factor. We've got a design that we've mocked up, so we fabricated it, and it's going through thermal and vacuum testing now. It's about 100 times less cost per watt of dissipation than the International Space Station radiator, and about 10 times less mass per watt of dissipation than the ISS radiator. And the main reason is we're not doing 3D additive manufacturing. We have a much simpler manufacturing technique, which, yeah, that's kind of our core IP. Sure. And then just, I mean, can you give me like a mental picture of what I'm visualizing? Because I feel like it can't just be a sail that just like there's no wind, so the wind doesn't expand the sail.

2:54:43Am I thinking about like hinges and little tiny motors to sort of like expand and unfurl? Yeah, exactly. So the deployment mechanism is what's called a pantograph. So it's like this scissor mechanism that if you've ever seen, like when you walk past a building site, you sometimes see those like platforms that will raise up and down. They're lifted by this pantograph scissor mechanism that goes like this. It's exactly the same deployment mechanism, which by the way, all Starlink solar is deployed using a pantograph. So what we've done is we've just added, you might have say 20 sections of solar panel that unfold in this Z-fold.

2:55:21And then we just added five segments of radiator at the beginning, which also unfold in this Z-fold, and then they're connected with tubing. And then we're pumping coolant past the chips directly and then out to this Z-fold five-segment radiator. Oh, interesting. You're at one GPU in space. You filed for... Five GPUs. Oh, you're at five. Okay. I saw you short. So you have five, but you're going to 88 ,000 or something like that. What will the curve look like? Are you trying to do 50 and then 500 and then 5 ,000 and then 50 ,000? Are we going by orders of magnitude or is it like a lot of work and then they all start flying up really fast?

2:55:59So we are very heavily dependent on Starship before we will be flying frequently. So we're expecting Starship to deploy the first Starlink V3 payloads either end of this year or early next year. And then from there, we're expecting 18 to 24 months before the first commercial payloads, so customers like us flying. So we're looking at the end of 2028 before we're launching on Starship. But our own deployment schedule is later this year we'll be launching StarCard 2, and that will be a 10 kilowatt spacecraft, about 100 times the power generation of the first one we flew. And that has this, the first one to have this deployable radiator will be by far the largest commercial radiator in space.

2:56:32And then next year we launch a very similar version of that. And then we launch on Starship this PEZ dispenser. Actually, we may even launch it on Falcon 9, the first one, just to test it, which will be about 200 kilowatts. So 20 times, again, the StarCloud 2. And then we can launch 50 of those per Starship. So then you're talking about 10 megawatts of compute per Starship launch. And once Starship is flying frequently, we're expecting to be launching hundreds of times per month. So that's 100 Starship launches about a gigawatt of new capacity. So we're talking on the order of tens of gigawatts of new capacity per year being deployed at that point.

2:57:08Do you have any insight into how space compute might in the short term be better for a particular AI workload? It's going to be better at diffusion or worse at diffusion. We've seen some bifurcation just in the data center around ASICs, GROC chips, Cerebris, etc. Is there a world where we're like, oh, there's this one small use case that's going to really, really work well. And then maybe over time it'll do everything. but in the short term, we're confident about this? So at a high level, it's not going to be any training workload. And at a low level, it won't be anything that requires sub-50 millisecond latency.

2:57:49So there'll be a sweet spot in the middle, which in my mind is basically all back office business processing tasks and all code generation tasks. You could even do voice agents for customer service or anything like that, video generation as well. but it won't be things that require extremely low latency or things that require very highly internally networked constelate cluster like like a training workload you could not do yeah that makes a lot of sense uh well thank you so much congratulations on the round i'm very excited

2:58:16Marc Lore:yeah massive yeah and you you uh you seem like uh you've you've even evolved yourself since the last time you're on the show i'm sure i'm sure you've just you and the whole team have been uh pushing it to the absolute limit but um yeah it's it's it's fantastic to see someone like call they're shot and then it becomes just one domino falls after another and then it's being mentioned at gtc and elon's talking about it and it becomes uh you know much more like science fiction fully moves towards like science fact which is very exciting so thank you i appreciate that no thank you have a great rest of your day have a great weekend and we'll talk to you soon hopefully congrats on open ai it's amazing thank you we'll talk to you soon thank you goodbye uh i what else I texted my dad the news.

2:59:03Marc Lore:He says, congratulations. That's so exciting. Thanks for letting me know. Talk to you soon. Have a great day. Thank you. Thank you, Dad.

2:59:14It's amazing. Well, if you've texted me or you've called me in the last three hours, there's a good chance that I might respond to you in the next couple hours because we are winding down the show. Please leave us five stars in Apple Podcasts and Spotify. Subscribe to our newsletter, tbp.com. everything is the same we will see you on the fun week next week five shows 15 hours let's be honest

2:59:38Marc Lore:it'll probably be more like 17 or 18 or 19 we'll see who knows the world is our oyster and thank you for being with us along the journey we appreciate let's get one more gong hit john more gong hit it's been an honor

2:59:57Goodbye everyone

2:59:59Marc Lore:We'll see you tomorrow