I see a large IPO on the horizon. You're surrounded by journalists. Hold your position.
There's misinformation. Clearing order inbound. Let's just roll. We are surrounded by journalists. Hold your position. Come on, get up. Trust the experts here. We are experts.
I see multiple journalists on the horizon. Standby. UAV online. Blaze. Double blaze. Triple blaze. Double kill. Five is wins. Team deathmatch. We are experts. Triple blade. Let's just roll. Right.
Hockey clearing order inbound. Come on, get up. We are surrounded by journalists. Hold your position. Strike one. Strike two. Activate golden retriever mode. Market clearing order inbound. Five.
You're watching TVPN. Today is Wednesday, May 13th, 2026. We are live from the temple of technology, the fortress of finance, the capital of capital. Uh, bunch of news going on today. Huge show. Max Levchin's coming on the show. Delian's coming in person from Varda. Uh, Recursive is coming out of stealth. Brandon Hill from Boreas. Nate Tepper from Trueshort. Very excited to talk about uh vertical video. The trog, the trog is cooking. Some of the trog economy. Uh, Joubin Mirzadegan from Roadrunner and Roman Chernin from Nebius. Absolutely destroying earnings. Very excited to talk to the co-founder and chief business officer over at Nebius.
Uh, but the big news this week, and I'm sure what uh you know everyone will be focused on is of course on the cover of the Wall Street Journal. That's the Trump-Xi Ping summit that's happening in Beijing. Uh, Trump and Xi Ping, winner wins at summit. They're duking it out. And uh Trump brought, they both brought uh a big uh a big crew with them, a team. Yeah, Trump brought geopolitics, a couple couple team sport, couple political people that we won't even care about, but the business leaders are what is particularly interesting. So we got Elon Musk from Tesla, Tim Cook from Apple, Jensen from Nvidia, Kelly Ortberg from Boeing. Uh, little flex there because Boeing has been able to maintain the lead in manufacturing large planes for a very long time. Uh, DJI has not been able to scale up and make a 747. What's going on over there? Um, David Solomon from Goldman Sachs, Steven Schwarzman from Blackstone, Larry Fink from BlackRock, Jane Fraser from Citi, and Dina Powell McCormack from Meta. Uh, so we should read through the stakes of the Trump-Xi Ping summit in the Wall Street Journal from the Wall Street. Let's get let's get into the more important stuff which is...
Do you think Jensen was always coming on the trip? Right. There's been some speculation he got on uh he got on the plane in Alaska.
Was that just the most convenient place for him to get on or was he a late addition?
I don't know. Why would he have been a late addition? I don't I don't know. It seems like...
Well because there was there was reporting earlier this week he runs the largest company in the in America.
I know but but there had been statements made from what I saw that he wasn't going. I don't know.
Yes. Well, so Trump had a Truth Social post that said CNBC incorrectly reported that the great Jensen Huang of Nvidia was not invited.
But then he he corrected it. He said he didn't invite him personally.
I don't know. These schedules move around. But it's got to be really really difficult if you're the leader of the biggest company in the world to uh move things around and and commit to this. I don't know.
I know. But what's more important for Jensen and his objectives?
No, of course it's critical importance. But uh at the same time, there might have just been other moving pieces of like, okay, when can he go, where will he arrive from? Where will he get picked up from? I don't know. These things always like shift. Uh I I I don't have a good thesis. I mean other than the fact that like there that is the company that has maybe the most tension between the trade deals with China like Apple obviously does a ton of business in China but uh has been much less of a focus in the chip sanctions and so having Jensen there is a little bit more of a discussion point. What do you think?
I mean I think on China like the Nvidia stuff seems like pretty like fine now but but yeah but Meta I think has like some conflict right because I mean Manus. Yeah. Maybe it's like not that big of an acquisition on like market cap wise but...
A lot of these things you know we say in the in the context of Silicon Valley it's like a $2 billion deal or a company that's doing 40 billion in revenue. Uh and then you get over there and the stakes are like, you know, a trillion dollar oil market or like all the rare earths in the world and uh and it's your, you know, geopolitical conflict which is obviously a much bigger deal. And so um I don't know, we we we we can speculate on more. There might be more reporting that comes out about uh Jensen's goals there, what actually happened, uh how he found his way onto uh Air Force One, but that was confirmed by Elon Musk, right? You said that uh Elon, Jensen, and Trump were all on Air Force One together flying over. Um I certainly wouldn't want to get there. I would want to arrive on the cool plane. Can we play the video of the uh the arrival? Trump arrives in Beijing with Elon Musk and Jensen in tow. I guess the rest of the business leaders came separately. But uh here is their arrival with uh are those cheerleaders or something? I don't know what...
They they seem to be waving. Oh, flags. They're waving flags. That makes sense.
Um well, there they go walking down the red carpet. Uh Air Force One is truly stately. There's It's so It's so uh disappointing that the economics of the airline industry didn't just get bigger and bigger. And so you find yourself very rarely on 747s unless you're flying across the world. Uh because they do have such a presence and they're spacious inside. Uh but we we we have been ref uh you know cramped down into the 737s pretty much everything.
Where they they give a proper tour of Air Force One or is it kind of locked down?
I think they've done tours. I mean there isn't there are several Air Force Ones I believe that are uh on display at presidential libraries and so you can go and tour them. There might be one at the Smithsonian. I'm not exactly sure but uh I'm pretty sure you can go walk around old ones. I wouldn't be surprised if they keep the the the current spec like somewhat secretive. Uh there we go. This is very grainy footage. It looks like Oh, maybe it's just uh the middle.
Kind of reminds me of uh Dune.
Dune. My Iraqis. That's Taiwan though. Um anyway, let's go through Jensen's pack list. He was having trouble thinking of what was he going to bring. Of course, he landed on just a ton of leather jackets. The bag and the sneakers I feel like are pretty iconic. Yeah, like the uniform is really starting to compound where it used to just be the jacket. Soon we will know the exact brand and specs of the of the pants as well as the shoes and uh the whole Jensen uniform will be out in full force.
Yeah, the clothing industry doesn't want you to know this, but um wearing just just getting a few pieces of clothes that you like and then wearing them until they fall apart is incredibly underrated. Not saying that's what Jensen's doing, but...
Uh it's pretty funny. Uh I I tend to wear almost the exact same thing. Yep. Every day. And so sometimes people on the team will come in and be like, 'Oh, we're matching today.' Like every day you dress like this, you will match with me.
Yeah. That's right. That's right. Um there's obviously a personal brand benefit to adopting a uniform. We've seen it with Jensen. Steve Jobs with the black turtleneck. Palmer Luckey with the uh with the Hawaiian shirt. Big news from Anduril today. New round, Series, are they running out of letters? Series H. Series H, $51, $61 billion, something like that. Uh fantastic progress. But uh you adopt the uniform, you become more iconic, you become more recognizable outside of your face, and that builds your personal brand, your lore, and it helps you sort of differentiate and stand out, right? The flip side, the thing you got to watch out for is that if you are always dressed exactly the same and you never switch your wardrobe, clips of you saying something a decade ago can go viral. And if you're not, if you're aging gracefully, it will look like you said them just yesterday. And that can be a big problem if say you're an AI CEO who a decade ago was reading some sci-fi and thinking like, man, there's something to this Terminator thing. And then maybe in the past decade, you've done a lot of work, you've investigated the systems, you have a more informed position and opinion. Well, the old video of you still looks like you today. And so, uh, there might be something to leveling up the wardrobe and changing the wardrobe over time that actually creates more distance between...
Oh, well, clearly that was, um, so this is indirect way of saying you want Jensen to be in a leather sort of like, you know, sort of trench coat eventually, like flowing robes.
Well, the easiest thing to do is go bald. That's what Jeff Bezos did. If he was saying some crazy stuff back when he was building Amazon in the 90s and he's changed his opinion, uh it will be very evident from new clips that oh that's an old one because it's very it's very iconic his older look. Now his new look, he looks transformed. He looks very different.
I I don't think Jensen has to go bald, but he can do the the Bezos playbook which is just get like get really jacked.
So as he gets more jacked, you'll just see...
Oh yeah, there's I mean he's looking like 320 lean right there.
Like that must be pretty basic. 320.
Jensen at 320 would be electric. Uh well, he's...
Uh we have a video of the Air Force One Golden Palace.
Okay. Pull it up. What is the Golden Palace?
Is that what they call the interior these days?
The the Air Force One that was gifted by Qatar.
Okay. And that one is never going to be in use, but going to go straight to the library, I think. Is that the idea, or is it going to get flown around? Well, let's uh let's watch the...
Presumably, it's going to be used.
Golden Palace. Really particular aesthetics here. It looks like a normal 747 on the outside.
I feel like, isn't there a big uh there's a big swing in in uh fuel efficiency based on the color of the plane. If you paint your plane black cuz you think it looks cool, it will attract more heat and you will burn a lot more fuel. And so most companies go with white. Uh Spirit Airlines went yellow. I don't know if that had anything to do with their downfall, but uh you would think is there a chance is there some weird economic logic where if you paint it gold, the gold will reflect the sunlight and increase...
Unfortunately, the exterior is not...
So you should wrap your plane in like a mirror like...
Yeah, maybe. But yeah, I don't know if that actually that probably reflect the most, right? Mirrored plane. There are some silver There are some silver bombers and and jets that are in service across the United States Air Force. I can't call upon it right now.
This really is a sky yacht chessboard. You think anybody's played chess on there?
Yeah, the whole plane is actually used for 4D chess, John.
4D chess. Chess is already 3D by default. It's a 2D playing board, but time is the third dimension because the turn that you're on is a dimension as well.
This goes deeper than I thought.
So 4D chess is merely a 3D representation of 2D chess that's played over time. That is the fourth dimension.
Anyway, uh AI is being put to good use visualizing Xi Ping, Donald Trump, Elon Musk, and Tim Cook having a beer and I guess a cigarette as well in uh in China somewhere. Um, and then look who pulled up. If you scroll down, Jensen coming in from the back. Looking pretty good. Looking pretty good. Maybe not 320 lean, but he's certainly up there. He's scrolling up. Uh, and people are really having fun with this. It it turned into a a little chain letter. Um, anyway, um, what uh let's go through the Wall Street Journal's report and the what the editorial board had to say about the stakes of this summit. What what's actually at stake? So the United States wants stability uh but Chinese communist leader has larger ambitions says the Wall Street Journal Journal's editorial board. Uh President Trump visits visits Chinese President Xi Jinping in Beijing on Thursday and the pre-meeting US spin is a search for stability in quotes. Uh it's a nice idea as long as Mr. Trump doesn't think personal rapport can overcome Mr. Xi Jinping's anti-American purposes says the Wall Street Journal. Uh the agenda ranges from trade to technology to Iran. Uh on trade, the best outcome may be ratifying the status quo. A truce on tariffs with a promise from Beijing not to hold the world's rare earth supply hostage again. The markets, business leaders, no one wants chaos. Everyone can sort of work around a tariff plan if it's going to...
For a long time for Trump to be coming in being like, 'Look guys, all I'm asking for is a little stability here.'
Yes. I just want stability. I want calmness. Yes. I don't I don't want any conflict.
Yes. Didn't he famously call himself a stable genius? That stability has been part of his brand for a long time.
Uh at least the one that he projects. Uh Mr. Xi Jinping on the other on the other hand uh China's weak economy is an incentive for him to cooperate and he'll hope to placate Mr. Trump with promises of buying more farm goods. Uh we talked about this previously, more aircraft and uh other things. Uh so that's maybe why Boeing is there. Although there is not, it doesn't look like there's representation from an agricultural leader, although I I can't name one off the top of my head, but maybe they should be in uh maybe they should be represented. Uh but Mr. Xi Jinping has made that promise before and US farmers have never regained their lost market share in China. Uh Rare Earth's ransom offer uh the rare earth's ransom offer may be more US advanced chip exports to China. Uh Xi Jinping views artificial intelligence as a decisive theater in Beijing's competition with the United States and he is trailing though not by much maybe six months by most people's estimates. Uh the administration wants to talk with Beijing about AI guardrails and by all means keep the phone lines open. Uh but don't expect much from AI arms control. Uh and the best deterrent is US dominance on models and computing power. Beijing will be happy to make pronouncements about responsible stewardship and then pursue its own interests with little regard for norms or laws. Uh Beijing is is engaging in quote industrial-scale theft of American AI models. The Trump administration warned this year, and don't forget the Justice Department's indictment this year of a technology executive and associates allegedly running a sophisticated operation to divert high-end chips to China. That was uh...
Super Micro. Yeah, they were doing the the the heater. What was it? The hair dryer.
Yeah, they were putting the the fake shipping labels on there.
Fake shipping labels. Yeah. Um, noticeably absent from this trip are the key leaders of the top AI labs. Um, I mean Elon Musk is there. Yes. And Jensen is there obviously deeply involved in AI as as are many of the other folks. Uh, Dina Powell McCormack at Meta, but you don't have Demis from Google. You don't have Sundar from Google uh DeepMind. Uh, you don't have Dario from Anthropic or Sam Altman from OpenAI. And each one of those leaders has their own complicated relationships with the federal government and the Trump administration. Uh so it's not shocking that they aren't there, but uh it does feel like there's a little bit of a disconnect between the conversation that's happening in Silicon Valley and the conversation that's happening on the global geopolitical stage at the highest levels. And so, uh, if you were looking for answers to some of the biggest questions, uh, posed about how superintelligence will play out in the global stage, the AI 2027 scenario, the China wakes up scenario, uh, you're going to have to keep waiting because uh, that's not really what this this uh this is about. It's about uh maybe a little bit of semiconductor supply chain details and export restrictions, but not answering the big questions about uh what a US-China relationship looks like in a post-AGI world, a post-ASI world, a post uh you know a fast takeoff scenario.
OTP says why is Meta there at all? Good question. All Meta Platform products have been banned for quite a long time. I don't think they have this man. I would be shocked if there's any time between Xi and Trump that gets dedicated to Manus because like $2 billion is really a rounding error when you look at all of these other issues. Um but uh but it is a big question for both economies. Not not only can American companies acquire Chinese companies that have relocated to Singapore and have sort of moved out like how how strict will China be around maintaining talent and restricting exports of whole companies and technologies uh there even like a single sentence on it from this talk will clearly resound throughout the industry. Um and so uh the Venus flytrap that Xi Jinping is setting for Mr. Trump is on Taiwan. Xi Jinping wants veto power over arms sales from the United States to Taiwan and he is pressing for the United States to formally oppose Taiwanese independence as opposed to the current posture of not supporting it. Uh Xi Jinping will argue the tweak is of no great consequence to Americans and stoke uh and stroke Mr. Trump's ego that he can bring peace to one more troubled region. Yet that change would disrupt decades of US policy that for all its delicate diplomatic wording has held the peace. Taiwan is not the aggressor in the Taiwan Strait. A Xi Jinping fiction that opposing independence would indulge. Mr. Trump may not care about Taiwan's freedom or its example that a prosperous Chinese democracy is possible. Uh but the president doesn't want a crisis on his watch, which would be an economic and geopolitical catastrophe. That's 100% accurate.
He would never want an economic geopolitical catastrophe.
Uh Xi Jinping will be looking to see if Mr. Trump suggests he won't defend Taiwan in the clutch. Uh Trump's diplomacy is above all personal and no one can predict what he'll do in the room. Japan and others in the region are watching with anxiety. A reminder that US support for Taipei is an interest that informs America's alliances around the world. One mistake would be not stopping in Tokyo in advance of Beijing as a signal of solidarity with Japan. Uh Trump has said he'll bring up the case of political prisoner Jimmy Lai. Uh but Xi Jinping won't move to release the 78-year-old publisher who was convicted of bogus charges in Hong Kong unless he believes Trump's request is more than a token gesture. Uh the US wants China's help on Iran and it would be an improvement if China at least stopped actively helping the enemy. Uh retired Navy Rear Admiral Mark Montgomery suggested last week that Mr. Trump should ask Xi Jinping directly if he's helping Iran with intelligence. And the Wall Street Journal goes on to say the larger context for the summit is that the CCP continues to be the main financier and industrial base for the world's bad actors from Russia to Iran to North Korea. The first Trump administration understood China as a strategic adversary, military, economic, and ideological. The second Trump admin is searching for detente. And Mr. Trump is the chief dove. Uh, this has some merit if America spends the interlude diversifying its rare earth supply chain and passing a $1.5 trillion defense budget to rearm, says the Wall Street Journal editorial board. Um anyway,
Yeah, and a lot of people are speculating that uh there's already been significant deals that have been made and that this whole thing is ceremonial because if if you bring if you make this, you know, huge trip and you bring all these leaders of industry and then you don't come away with some significant progress like it just feels like a even bigger letdown.
Um, and uh, and it's good to it's good to uh, it's good to see I I every time we talk to a ch- to an expert on on China like like Bill or Jordan Schneider, it's always shocking like how little like direct comms there actually is between these groups like and so um, the more the better. Mhm. Well, uh, in somewhat related news, uh, Anthropic has granted Mythos access to three major Japanese banks. Uh, Japan's three mega banks are set to gain access to Claude Mythos, the powerful artificial intelligence model developed by US startup Anthropic as soon as the end of May. Uh, the banks MUFG Bank, Sumitomo Mitsui Banking Corp, and Mizuho Bank were likely informed of the move by the US Treasury Secretary Scott Bessent in a meeting in Japan on Tuesday. This will mark the first time that a company from the East Asian nation has been granted access to Mythos. The AI model, which can discover and exploit software security flaws far faster than earlier tech, had been restricted to just 50 or so corporates and organizations worldwide, including US firms, US banks, and UK government organizations. So, this isn't the first international expansion since the UK did have access, but now Japan does. Japanese Prime Minister Sanae Takaichi had instructed cabinet members to ramp up efforts to find cyber security weaknesses in Japan's infrastructure and minimize risks posed by cutting-edge AI models like Mythos. Access to Mythos was likely mentioned in her meeting with Bessent on Tuesday. And so, uh, Chris Maguire says, 'Tonight, the uh secretary of the Treasury is personally vetting and approving uh each company that gets access to the most advanced US AI model because the risks of the model being misused to hurt US national security are so high. Also tonight, Jensen Huang is flying on Air Force One with President Trump to Beijing to sell China the AI chips it will use to develop its own Mythos-level AI model as soon as possible. The administration's AI policy remains inconsistent, incoherent. It is impossible to justify these two approaches simultaneously.' And um yeah, I I'm I'm still sympathetic to the the the export control take. At the same time, I feel like there's a larger negotiation that's going on, so I can sort of see both sides. But uh the interesting question that I've been noodling on that I haven't fully gotten to the bottom of is uh you know, Dario has said that, uh you know, the the open-source community, China, is maybe six months behind Mythos. Uh and that seems to fall in line with the progress that we're seeing from all the American labs. And so the interesting question is what would the uh Mythos moment look like in China? Because in America there was a showdown with the Department of War and Dario going back and forth and Amodei, Michael, and there was threat of a supply chain designation which it doesn't feel like ever really went anywhere uh because the business is still cooking and it's available on all the hyperscaler clouds and uh the the model's being granted to Japan and so uh it feels like the the the administration has backed off of that. Um but it was a really tense moment and it did call into question all these all these debates around what rules and what authority should rest with the private sector versus the government. Uh and America handled it in sort of a democratic way loosely uh sort of evaluating the technology uh having the option to sort of nationalize and there's this bigger discussion around that. Um, but in China, we've seen many times where tech companies have gotten really big and it feels like the the CCP is applying pressure to the CEOs and High-Flyer is uh the the company behind DeepSeek is in a very unique position where the founder has an immense amount of control and uh you know no matter what you think of...
Two billion of his own money, really low dilution round.
Yeah. And I'm pretty sure he's a solo founder who like owns probably like majority control, at least majority voting control, if not majority economic control. And that's wildly different than Anthropic where Dario had like 15 co-founders, tons of dilution from uh from different uh funding rounds. And you know, we had talked about the SPV thing yesterday. Like there are lots of people on the cap table like Anthropic is going the way of like a typical American company where there are lots of different stakeholders. They might be a public company soon, which means the government has more oversight as the SEC controls a lot of what they will do. And so you're in this odd scenario where you you could have a similar level of technology deployed in China from a very closely held private company that is up against a much more aggressive uh government. And so what that looks like...
Yeah. What that looks like functionally. Yeah. Like if you can't do e-commerce without disappearing for a week or something like...
Do you think you can do Mythos-level AI cyber security bug finding at the same time while having full control and not being a public company? Like like there are like it feels like the stakes are much higher uh for when DeepSeek if DeepSeek is able to catch up. You know, there's some there's some some evidence that like China's on a on a slightly uh shallower growth trajectory, but still, you know, if you take Dario's claim of six months, even if you extend that to a year, like something is going to happen there where China is going to have their own Mythos moment and what that looks like and what we learn about it, I I think is going to be dramatic and interesting. And also it's just it's just odd that like this moment there's this backdrop of all this stuff that's happening in AI in America and then there's going to be you know talks over like you guys want to buy more apples basically or like like you know what uh I mean obviously there's going to be more high-stakes discussion around geopolitical conflicts and rare earth but uh we are talking about like tariff rates at a time when there like just a couple months ago in America we were talking about like you know is AI going to like be able to take over for the government and like what does that mean and how does the government control the AI technology as it diffuses. So, uh very very interesting things coming at some point. A lot of this will probably be pre-negotiated behind the scenes since the Mythos moment happened in America and Xi Jinping obviously saw it and can get ahead of it. But you do see this consolidation of control at DeepSeek and there's a question about yeah well where does that go? Uh Tyler, do you have any uh any thoughts on on this? Am I directionally correct on the timeline or you know this concept like there's the graph that shows that uh open source is increasingly getting like further and further away from frontier models. Um but yeah I don't know it's just like so hard to tell like what's actually going on in China like what what is it actually yeah like what what happens when the CCP starts to to have you know more control over a company like what does it actually look like? Yeah. In in instantiation.
Yeah. I mean it it's it's interesting like so much of the way the US government exercises control over the development of AI is on like the FDA regulatory we want to review it. We want to take a look at it. We want to take it for a spin and throw some prompts in there before we let you send it to Japan for example. Um and in China it's a very it's a much more industrial process where you know if the if the chips are coming in and it's a whole negotiation at the government level there's just the question of like okay can you even get the data center to and it's much earlier stage whereas that's not happening in America at all like there hasn't been any movement on the government saying okay well we're going to the way we're going to control AI is that we will be the we will decide if you can have a gigawatt we will decide if you can have 10 gigawatts Whereas in China, that's sort of the de facto status quo.
Yeah, they they still are very pro competition.
At least until you get these, you know, breakout winners. So, yeah.
And and can they continue to be pro competition if they need to consolidate all of the Blackwells that are coming into the country into one data center, this network together that can actually train a Mythos-level model, right? That's uh that's that's a big question. Uh, but people are having huge FOMO over the uh over the summit. Uh, dude, I wish I was in Beijing with Elon Musk, Trump, Xi Ping, talking business, discussing AI, real boss moves. Must be a movie, says someone who we're not going to read because it's a bad name in the thing. Anyway, uh, speaking of Elon Musk, uh, SpaceX and Google are in talks to launch data centers in orbit. Deal between the two tech titans would give a boost to SpaceX business ahead of a historic public listing. Uh Google is in talks for a rocket launch deal as a search giant expands its own efforts to put orbital data centers in space. Launch deal would put the two companies in partnership as they gear up to compete on orbital data centers, an unproven technology that SpaceX CEO Elon Musk has said is the next frontier for his rocket company. Google is also in discussions about a potential deal with other rocket launch companies. Um, they got to at least have a stalking horse.
How are we not talking about blimp data centers yet?
I know. Sergey Brin has a blimp company. You know, Bezos has a rocket launch company. So, there's a very logical line from AWS to space data centers through Blue Origin. But, we got to put some GPUs on the blimps. We got to start mass manufacturing the blimps. This could be the thing. I mean, no, no, there's no NIMBYism in the middle of the Pacific Ocean if there's just a bunch of blimp data centers floating around in a circle.
Well, I think if you're already in the ocean, you should go down as well. You know, Atlantis map. Yeah. Like I feel like none of the ocean guys that are working on ocean robotics are thinking about like putting at the bottom of the ocean.
Yeah. I mean, it feels easier than than uh space data center infrastructure. The space is free. The land is free. You don't run into the regulatory stuff. Cooling potentially free. Energy a little bit tricky down there.
Geothermal maybe. I don't know.
No, there's there's there's oil and natural gas reserves.
On the oxygen there because how do you burn the oil in the natural gas water?
Diving bell tube. Yeah. Scuba scuba tanks or something. Anyway, let's continue. Where are we in this? SpaceX and Google. Speculative technology has been at the center of SpaceX's pitch to investors ahead of its planned IPO this summer, which is anticipated to be the largest IPO of all time. Last year, Google announced its own plans to launch prototype satellites by 2027 as part of a moonshot initiative called Project Suncatcher. It's working with another company, Planet Labs, to build those satellites. We're working on getting Will Marshall from Planet Labs on the show in the next week or so. Uh, we'll send tiny racks of machines and have them in satellites, test them out, and then start scaling from there, says Sundar. There's no doubt to me that a decade or so away we'll be viewing it as a more normal way to build data centers. Uh, a little bit of an interesting timeline there.
Like, I thought we doubt to me a decade or so away, we'll be viewing it as a more normal way.
Yeah. A decade or so away.
He's being a little cautious in his language on his timelines, but...
Google was an early investor in SpaceX and own 6.1% of the company.
Um, the other the other funny dynamic from SpaceX's deal making recently is given that the OpenAI Startup Fund was uh an early investor in I believe Cursor's seed round. True. The $8 million round. Um, OpenAI, at least the Startup Fund, will have some meaningful exposure to SpaceX.
Yeah, SpaceX, an OpenAI-backed company. That's the way people will introduce it. OpenAI-backed SpaceX IPO.
Yeah, I thought it was it was quite interesting having that happen while the whole trial was going on.
Everyone owns everything. Everyone has diversified into every every company at this point basically.
Uh, let's see. Uh the company last week announced a deal to sell earthbound computing resources to the AI company Anthropic. As part of the agreement, Anthropic expressed interest in working with SpaceX on orbital data centers. Earlier this year, SpaceX filed an application with federal regulator to launch up to a million satellites for its orbital data center ambitions. That's crazy. I remember, didn't they get what what was the original Starlink approval? I thought it was like tens of thousands, maybe a hundred thousand satellites. I'm starting to think like like looking ahead there's going to be a backlash to like they're blocking out the sun. Like we've seen the Matrix. This happened in the Matrix but for like chemical reasons, but like if there's like no like light pollution from this at all. It's going to be a absolutely massive...
Oh yeah. Pantanalosa ocean data centers. You don't need to sink them to the bottom of the of the of the Mariana Trench. Just do them on the on the ocean.
Uh do did you look up how many uh satellites SpaceX currently has in orbit? Because I think I mean it's a huge number. It's over 10,000 I believe. But uh a million is a huge jump. Um but that should be that actually should math out I think we did the math. That should math out to gigawatts in space which is definitely meaningful.
So right now they have uh like just over 10,000. Is that...
Just over 10,000? Okay. But I think they're approved for maybe up to a hundred thousand or something. And so they go and get the huge the huge allocation and then they start launching. Uh very very excited to see uh the progress on Starship. What isn't there a launch that happened or is happening this week? Like what?
Yeah. When's the next time when the next launches? I'll keep reading.
The other the other story, this was back in March. Uh Blue Origin has a play in in space data centers as well. So I would kind of you could imagine them and Amazon teaming up in some capacity. But you could also imagine Amazon saying like, 'Hey, if we're going to be a real player here, we need to we'll probably need to work with StarCloud finally probably finds a dance partner, if not multiple.' Uh, many industry leaders see orbital computing as a solution to the limitations of Earthbound data centers. Earthbound is capitalized, it's a proper term now, uh, which require swaths of land and power. These data centers are designed to be powered by solar panels, emitting uh, eliminating the major power constraint faced by data centers on Earth and one of the major ecological issues with emerging technologies. Uh but there are also major engineering challenges. Uh SpaceX made its name as the world's leading pro- private launch provider, sending NASA astronauts to the space station and launching thousands of satellites as part of its Starlink internet constellation over 10,000 apparently. Uh as it prepares to go public, SpaceX has struck a number of deals that have rewritten its balance sheet, helped Musk consolidate his companies, including its acquisition of xAI. Uh and SpaceX also announced a deal with Cursor as we have talked about on the show. Um Star Starship, what's the update?
So the next launch, the 12th one is next Tuesday, the 19th.
That's when they're targeting. So...
Okay. Yeah. Yeah. And what's the goal?
But they were like moving it down the runway. Sort of getting ready to...
Do you have any idea what success looks like?
I think this is the first orbital test of the V3.
Okay. Um, so they got to get it to orbit and then get it back and probably land it. But are they going to land on land or water? Because I it the last didn't the last one make it up and back but exploded in the water or was that the Blue Origin New Glenn? I don't know. Anyway, uh there's a whole deep dive on space data centers in the Wall Street Journal. They say, 'Is it a pipe dream or AI's next big thing?' They're pointing out some of the challenges. Uh, you're going to need thermal louvers, opening and closing shutters to block or allow heat, thermal contact, special bolts to separate items with different heat requirements, deployable radiators. Heat is transferred to special surfaces that unfold in orbit. It looks like a big...
Pull up this graphic showing sun-synchronous orbit.
Yeah, this is cool. So when if if you're orbiting the earth vertically essentially uh the the sun is constantly hitting the satellites so you you're not out of power for half the day. Um but they say these radiator the radiators for data centers will likely be very large. I like that phase change materials. Materials that can absorb uh that can change phase to absorb heat like wax or salicylic acid usually from solid to liquid when returned to normal when temperature is reduced. Sun-synchronous orbit 373 miles to 497 miles up. So about 400 miles up is when you get into sun sun-synchronous orbit. That's on the low end of low earth orbit. Low earth orbit around 1,000 miles, mid orbit 1 to 10,000 miles, high orbit over 22,000 miles. Um, and so, uh, we we we we covered a decent amount of this, so we can we can move on. Um, do you want to talk about Shanu Matthew?
Was adding to the conversation we had with Doomberg.
Doomberg was fine. Doomberg was saying the only way the data center buildout will be successful is by either breaking all the rules or going off-grid. And Shanu said the tech crowd keeps leaning into this off-grid is the only way thesis and I disagree with this notion. The exhibit A example of xAI Colossus even overstates what's actually happening at Colossus.
xAI did not build a permanently off-grid AI campus because the grid was unusable. Colossus already has 300 megawatts of TVA interconnect approved across two substations xAI funded plus contractual...
obligations to curtail during periods of grid stress. That is a grid-tied architecture with supplemental on-site generation, not an islanded power system. I think Doomberg highlighted the other side of this, that Colossus is the exception that proves the rule because it moved so quickly and took advantage of so many different grids, and there was power generation from multiple states at one point. And now there is pushback, like there wasn't pushback when Colossus was built originally because data centers were not a hot topic. And it's the same thing with launching rockets. We were talking to someone who lives up in Santa Barbara. They're launching a lot of rockets from Vandenberg. It's getting to be a lot of traffic. It hasn't risen to the level of, you know, protest, but you can imagine that building 10 more Vanderbergs would be difficult because you go to the local community and you say, 'Hey, we're going to build a space launch station, a spaceport.' And they say, 'Oh, that sounds good.' Like, you mean because we're going to go to the moon once a decade? And you're like, no, we're going to launch rockets 10 times a day or a hundred times a day. And people, if they have an example that they can draw from and say, 'Oh, well, that community doesn't like it, so I shouldn't like it.' That's some of the pushback. But anyway, we can keep reading Shanu Matthews' post.
Turbines solved a time-to-power problem. So it accelerated the deployment while substations and transmission infrastructure were being built. That's very different from doing all of this off-grid and economically. Most large-scale AI load will still want to be grid-tied long term. Interesting. I have made this point several times, but running your own generation fleet at hyperscale is expensive and operationally complex. Fuel logistics, maintenance, N+1 redundancy, permitting, emissions compliance, balancing load and supply continuously, multi-year equipment lead times. You don't just figure that out or want to do that overnight. Amin, the chief technologist at Google AI — does that mean DeepMind or straight up — confirmed this on a podcast and they literally bought a hybrid island power developer. Behind-the-meter generation absolutely matters. It's becoming a critical bridge solution during a period where utility interconnect timelines are badly lagging demand growth. But I'd strongly push back on the idea that the steady-state future is fully off-grid hyperscale campuses. The industry is converging towards hybrid architectures: grid interconnect plus captive generation plus storage plus demand response. They're throwing the whole kitchen sink at it. Anyway,
Um, related — yes, More Perfect Union had a post earlier and they're turning off — nearly 50,000 people in the Lake Tahoe area have been told their utility will stop providing power to them because it's redirecting that power to data centers. NV Energy, the Nevada utility that supplied most of Lake Tahoe's electricity for decades, says that next year it will stop servicing homes in the area and instead direct that electricity to the growing demand from Nevada data centers. Northern Nevada is one of the fastest-growing data center corridors in the country. I read this and I was like, this is completely insane.
It's so stupid. Like, you read it and you're like, this is such an L in the making for the data center industry. What were they thinking?
And then it's immediately getting community noticed. Okay.
And a quote from the president of Liberty Utilities in Lake Tahoe says, 'This does not mean the power is shutting off. Energy companies, utilities, and large customers change energy supply frequently.' So there's a supplier of energy that is saying like, 'Hey, we're not going to renew our contract. We're going to divert the energy elsewhere.' But there's a bunch of suppliers of energy.
And so Andy Masley says, 'Everything MPU posts about data centers is complete garbage. They have zero respect for their audience. Literally, no one here is losing power. This tweet is a complete lie.' What's actually happening is that a supply contract between two utilities is ending and the small one is just buying power from elsewhere. And this was all expected to happen since 2009.
No way. Wow. You were going to say like two years ago.
The company that serves homes on the California side of Lake Tahoe is a small utility called Liberty. Liberty buys about 75% of its electricity from a much larger utility, NV Energy in Nevada, and generates the other 25% itself from solar farms it owns.
Liberty then sells that to 49,000 customers.
NV Energy has told Liberty it will stop selling them wholesale power after May 2027. It's kind of like Liberty is a coffee shop that buys beans and sells coffee to customers. The customers are the homes, and the beans are the electricity it buys from NV Energy or makes itself. This is like your local coffee shop ending a contract with a specific bean company and starting to buy beans from somewhere else. It doesn't stop you from buying coffee. Why is their contract with NV Energy ending? NV Energy selling to Liberty was understood as transitional since it started in 2009. Long story short, NV Energy was basically Liberty's only wholesale option, but a new transmission line opening in May 2027 gives Liberty access to a much wider western market and, among other things, a much larger share of solar and wind and hydro. That's the whole story here. Ending the contract with NV Energy and opening up this much wider pool with much more renewable energy was the plan here, completely separate from data center demand.
Interesting. Well, anyways,
it would be remarkable if all of Tahoe lost energy and Mark Zuckerberg couldn't charge his electric wakeboard because I somehow —
He would hit the streets. He'd be protesting.
I think he'd be protesting. I think he would get something done if power was lost to Tahoe. I mean, a lot of influential people are out there. So anyway, all right, we got to set the record straight on that.
We got to talk about Unitree G1.
They build a new robot and you're going to love it. It's not scary at all and it looks very fun and safe. This is from Avatar. They just built the mech from Avatar. This has been pictured many times in science fiction and will strike fear into many people. Coincidence they use a racing bucket seat.
It is cool. It does feel very hacker culture, like it feels like this was a crack team, a little skunk works project. It doesn't look as polished or ready for sale, but remarkable results. They're like, 'You got to put this out so that the Americans cancel all of their technology ambition and just go back to the dark age.'
It is intimidating. Is it a coincidence that it was launched the same week that the summit is happening in China? I don't know. It is amazing that it turns into a horse of some sort, a four-legged creature. Although, how do you drive it at that point if you're laying backwards? It needs some sort of gyroscopic seat. That one hasn't really been considered because if you're laying back while you're driving it in horse mode, maybe you'll just nap. Maybe you're like, 'Oh, I want to commute.' You know, Henry Ford said if he asked people what they wanted, they would ask for a better horse. This looks like a better horse to me.
You get to work, you pop up into vertical mode. I'm fascinated by how they're controlling it. Is it just like a video game controller? Because some of this seems like it would require you to telegraph, like in the Avatar films, the operator of the mech has full — the input is the actual arms. So if they want to swing a sword, they swing the sword or they swing the machete physically with their arm and then the robot does the bigger version of that.
Timeline until these mechs replace horses for police use cases.
I don't know. There's going to be pretty crazy pushback if that's rolling around during a ride.
I know. But the benefit here is that I don't think the mech will, you know, go to the bathroom on the ground.
Obviously this is going to happen at some point. A lot of these developments are very early. I bet you if you give that thing a strong kick, it falls over and sort of gets stuck at this point. SA over on X had pulled out the funniest part.
Which is that Unitree unveils the world's first mech, show it breaking down a brick wall. And then in the launch announcement, they say, 'Please everyone be sure to use the robot in a friendly and safe manner.'
Yeah. Wild. Egocentric data collector versus the mecha-mounted teapot Chad. Absolutely wild. Is there another video of this? Oh, yes. This is an AI video, but we have to play the version of this getting hot-dropped into what you could imagine as the battlefield with four FC-2004 drone formation. Max payload of 600 kg. Unit's new transformer mecha, 500 kg of weight. That's probably without a human inside. So you might need six drones. But this is something that feels not that far away. Not that far away. Although, you drop that thing in the ocean — I feel like there are so many steps on the humanoid robot path where, you know, the deployment on a flat surface in a warehouse makes so much more sense than standing on a ladder straddling a chimney in the rain and there's a little bit of mud and you use a power tool and dust kicks up. Like actually ruggedizing all of this is probably a whole second step in this process where you're adding more weight, that means more battery, you need better battery technology. I don't think you'll be dropping this into the Taiwan Strait anytime soon, but it certainly strikes fear into the hearts of many. Well,
there's more humanoid news. Brett Adcock has a livestream going.
I do wonder how they're — again, I'm sure the comments are all, 'How do we know this is actually live, right?' So is this the actual livestream right now? It's continuing going because I pulled this up an hour ago and it looked — I mean, I guess exactly the same because it's the robot and they've been sorting for an hour. Battery pack or plugged into the wall? What do we think?
He says watch a team of humanoid robots running a full eight-hour shift.
Oh, okay. So they might have swapped it out at some point for a different — interesting. It looks like a lot of progress. It's extremely fast. It is fast. Like the movement's very natural. I wonder what is the goal here exactly? Because is the robot sorting these? What does it do with the blue one?
I think it's putting it so the shipping label is visible.
Or the shipping label's down. Oh, okay. Okay. Yeah, that makes sense. That feels like — because some of the robot applications are things that you could do with a two-axis gantry or a sorter. Like if you watch How It's Made, there are industrial machines, not robots, that can do all sorts of sorting. Like you watch how Diet Coke gets filled and the cans roll down and at some point they all need to be facing upwards. And so there's a machine that sorts them and tilts them all upwards. Obviously the humanoid use case is much more unstructured and that is beneficial for so many things. I don't know if this is your —
Apparently they're just going to have the robot go until it taps out.
Okay. I like that. That's great.
Very cool. Well, in other video news that we need to play, General Catalyst launched an ad on X that might be making it into TV ads or ads at some point. What?
GCAT. What's GCAT? Oh, General Catalyst. Maybe that's the new ad campaign, but the campaign is called Meet GC. And it's a clear reference.
Who's your friend here, VC?
This is Woof AI, an AI-native companion platform that combines robotics and machine learning. You'll never want a real dog after this.
Well, I think people like dogs as they already are, though, VC.
You don't need to walk it. You never need to tell the kids you sent Milo to the farm. We're leading the seed and could probably make room for you.
Well, I'd love to hear more, but we actually have a really high bar around responsibility for these things.
Stay. Stay. No. No. Stop. Stop. Stop. Stop.
Interesting symmetry to the podcast set that Reggie's doing, right?
Yeah. But it's cool that like the brand — like you could wind up doing two very different designs, but there's a through line there even though it's obviously a nod to the Mac versus PC ads from the 2000s, which we can pull up to watch one of those. I think we've watched these before. And they have a very similar, you know, stiff PC versus Mac look. But I think that was a good message of like, okay, like the other —
that, but also like let's not fund the stuff that is pointless and useless, and if you already have a dog —
this is a big moment because you have been wanting a VC firm to just commit to paid advertising for a long time.
I wouldn't say this is a commitment yet. We got to actually spend behind it.
Yeah, of course. Of course. Yeah. No, no. Magazine ads.
Yeah. In the right — I mean, certainly in like the Wall Street Journal and Forbes and like the places that entrepreneurs go for information, it makes so much sense to actually spend on advertising to build a brand, especially when you're at the scale of General Catalyst. Can we pull up the Mac versus PC ads? We're not going to watch all of them. There's 66 of them. Talk about a generational run.
A lot of the same kinds of programs.
Yeah, like Microsoft Office.
But we retain a lot of what makes us us.
You should see what this guy can do with a spreadsheet. It's insane.
Yeah. And he knows that I'm better at life stuff like music, pictures, movies, stuff like that.
Whoa. Whoa. What? What exactly do you mean by better?
By better, I mean making a website or photo book is easy for me and for you it's not.
Oh. Oh, that kind of better.
Yeah. I was thinking of the other kind.
What other kind? Um, hello. We can switch away because you've seen all these. Mark Heaney quote-tweeted it, dunked. He said, 'Stay tuned for our upcoming ad campaign. We're the VC who doesn't sneer at your idea.' Interesting. Getting spicy. Um, yeah, red alert over there. Red alert. Um, it's a lot of red alert going on. Well, both companies are invested in robotics company Anduril.
Oh, yeah. I mean, I'm sure there's a ton of portfolio overlap.
Oh, no. I just think in another world that's an Anduril product, that dog thing.
I'm sure they've experimented with that form factor and some fresh funds to build a mecha.
I mean, if there's ever someone to build the mecha, it would be Palmer Luckey. He has talked about this stuff. I even asked him on stage once, like, walk me through the Pacific Rim vision, like how do we build that? And we had a fun back and forth about it. Well, we — do you remember much about his response?
Yes. So his response to stuff like that — pick a form factor — is just like, hey, I'm a sitting duck on a platform.
Yeah. Exactly. So his point is always like the technology to build the mecha is way behind the technology to like remote control something or put an AI on board or something like that. So you actually don't want to build the Iron Man suit because you can just build the drone, the submarine, the —
Or he'll use the example of like the Batmobile. It's like if you need a vehicle for the city, like do you really need this like massive vehicle, right? No. Clearly Bruce Wayne just — the parking. He's obsessed with building Batmobiles. Very funny.
I don't know. That is a very cool demonstration though. But yeah, Andrew always says, 'Don't build the Batmobile and focus on like is there an actual urgent need from the military? Is there support from Congress to actually appropriate the funds?' And Unitree, I don't actually know. That feels like just an R&D that's like a marketing project at this point.
Yeah. I mean, I think their IPO is like very soon, right? Because I think the original application was like late March.
That would probably pump it. And then also, you know, they're showing it like break down a brick wall, but like when I was thinking about what that's actually valuable for, we were talking about like could you deploy that and like sneak through a forest and then attack the enemy or something. It's so big. It's so visible. Like a single drone with a, you know, fiber cable coming out of the back is going to blow that up. It doesn't really seem that practical. Maybe if you're going over really rough terrain, you want the legs instead of a tank tread, but you can sort of just put a huge tank tread on something and just crush everything in your way. But if you're building and you need to get a bunch of 2x4s onto the roof, like there's a theory that that might be a little bit better. You were saying you're forklift maxing and you think the forklift will always defeat that. But there's got to be certain areas where you can't necessarily bring a forklift because there's a narrow — it's uneven ground, a narrow entryway into the back of a house. You have all the 2x4s. You need to get them on the roof. You would otherwise be going up and down a ladder. And so mecha to the rescue.
Yeah. I mean, it's just like pretty hard to justify this. Like I think yes, for certain like emergency scenarios, it makes sense to have some like super weird form factor.
Emergency. I need a really cool piece of tech. I need something that looks really cool before my IPO emergency. Well, we can flip over to the other financial news that's making headlines on the cover of the Wall Street Journal. Inflation accelerated 3.8% last month. Not good news. Joe Weisenthal has a couple posts, but one is just — man — and he says the headline: US April producer prices rose 6% year-over-year. The estimate was 4.8%. So obviously the closure of the Strait of Hormuz is having an effect. We were talking about Diet Cokes. You would think that the caffeine doesn't come from the Middle East. The delicious caramel flavor, whatever else they put in this, does not come from the Middle East. But aluminum actually — there are smelters in the Middle East. Aluminum is in shorter supply because of the closure of the Strait of Hormuz and so Diet Coke is rapidly going out of stock and so you need to load up if you're prepared.
How many you got? Four or five pallets?
I have not actually gotten any. I'm not stockpiling Diet Coke yet, but who knows? Maybe prices go up. That's what inflation represents and Wall Street is getting more anxious about it. So we can read through this quickly before we are joined by Max Levchin from Affirm. So surging energy prices have pushed investors' inflation expectations to multi-year highs. Tuesday's hot readout on consumer prices is intensifying Wall Street's anxiety about inflation. Even before the latest release of the consumer price index, inflation expectations had been climbing, a potential trouble spot in the market where stocks have largely shrugged off the US-Iran conflict and the resulting energy shock. And so investors bet on inflation by buying and selling both ordinary US Treasuries and those that hedge the risk of inflation called TIPS, Treasury Inflation-Protected Securities. The gaps between the yields of those two types of bonds, known as the break-even rate, recently hit the highest level since October of 2022, according to Federal Reserve data, suggesting that investors expect annual inflation to average around 2.7% over the next five years. Now, that's not entirely doom and gloom, but the risk here is that the AI boom — the GDP numbers are sort of concentrated in AI capex and data centers. It's not evenly distributed. The real economy is only growing at 0.1% while the real economy is suffering from 2.7% inflation over the next five years. That's a recipe for stagflation. Stagnation plus inflation. A very rough thing for any economy to go through and something that the market and the economy and the politicians should be focused on. So the 10-year break-even rate has also climbed, hitting 2.5% this month, its highest level since 2023, driven largely by the war-fueled surge in oil prices. The rise in inflation expectations worries investors because of the challenges it poses for the Fed's interest rate policy and major indexes' latest run to records. And so for a while everyone was thinking about maybe interest rates will come down, maybe mortgages will become more affordable. But there was even the whole debate or the fight with Powell and lawsuits which were eventually dropped. But Powell and Trump were at each other's necks and issuing statements about their vision for the future of the American economy. And on the flip side, you have Kevin Warsh coming in, maybe more friendly, maybe more receptive to a rate cut, but the data makes that very, very hard to justify. And so we are in —
Yeah. And in some ways, you know, who knows how history will view Powell, but like he landed the plane.
And then, you know, we get war in the Middle East and now Kevin Warsh in some ways looks like he'll have just as tough of a job.
Potentially. Potentially.
So well, our next guest, Max Levchin from Affirm, he's the co-founder and CEO. He's returning to the show and we've been keeping him waiting far too long. Max, how are you doing?
I'm great. Thanks for having me again. Third time's the charm.
Third time can't be the charm. I want the 25th time to be. Let's keep it going. I love these.
What have you learned about making coffee since the last important update? Because I'm assuming you're never like, 'Oh, I'm pretty good at this.' Or are you resting on your laurels? Are you the final boss?
I'm absolutely not. Okay. I took apart an E61 group, which is the classic 1961 sort of — that's the granddaddy of all the espresso group heads. And I wanted to learn how the mushroom valve works because I thought mine was clogged and it turned out not to be clogged, but I went down an extremely deep rabbit hole of taking apart an E61, which I've actually never done before. So I'm now really boned up on how espresso machines worked in 1961 and since. But yeah.
Yeah. People always talk about like when you get into coffee, you effectively vertically integrate or go deeper in the supply chain. You start roasting your own beans. You got to grow your own coffee. At a certain point you have to be smelting the metals that go into the machines, understanding the alloys, coming up with new chemical processes. You have to set up your own mine.
You're putting all kinds of wrong ideas into my mind. Check. Check. Ooh, not yet.
Yeah. Yeah. Yeah. Yeah. Yeah. Did you smelt the metal that went into the espresso machine?
I've not smelt anything lately. I'm a novice. A real novice.
Next time I'll tell you about my smelting experiences.
Fantastic. I imagine that the smelting process is as intricate and as rewarding as the coffee-making process in some ways. I expect you to be able to try a coffee and understand the alloy of the machine.
Yes, possibly. But I'm very, very long depth. Anything that you can go very deep into like this is the time in general, but like depth as a competitive advantage is like a profound strength. So the reason I'm so into whatever things I'm into is I found over the years that if you out-depth your competitors, they just can't beat you. So I'm very pro-smelting. I'm very pro going ultra-deep.
Yeah. Yeah. That's interesting. Is there a translation or if you transpose that to sort of like advice for young people who might have anxiety about traditional career paths? I was sort of, you know, we look at like job numbers all the time and layoffs and at the same time I think about when I was a kid if I said, 'Oh, when you grow up you will be a live streamer at an AI lab.' It's like none of those words existed back in the '90s and here I am. Also, there's so much advice out there, it's like you want to be a generalist. This is the age of being, you know, strong, maybe it's the opposite. And when I look at people that are maybe sort of midway through their career, the highest earning, the most respected are almost always the ones with extreme depth where they can simply out-compete anyone else in that role because they know more about it, they put in more hours. But yeah, what are you thinking for young people?
I'm definitely big on depth. I think these are like entirely non-contrarian opinions, but maybe they're contrarian right now, but I think it's a great time to get a computer science degree. I think if you're like ultra-deep into really understanding how software is made, you are like — everyone's going to be a 10x engineer. If you were a 1x engineer yesterday, you better be a 10x engineer tomorrow. That's the new baseline. But if you really, really get it, if you're smelting your own code — I'm just going to go with the smelting analogy forever — but if you're that good, you're going to be a 10,000x engineer and like you will be worth your weight in smelted gold, who knows? But it's very, very powerful to be a deep expert. Like you are the one AI wants to learn from and that is the unattainable Mount Olympus of value. And so I would strongly suggest that being deep is great. Majoring in computer science is great if you love computer science. Like this is the time to major in whatever it is you see yourself being as deep as possible because then you will become absolutely valuable in the world.
What about industry-specific depth? Like if a founder comes to you and they're like, 'Oh, I've been building e-commerce and I've done quite well, but really I'm interested in space. I want to work on that.' You've done —
You're interested in space and you're not working on it? Like what are you doing? This is the very best time in history to do that.
Yeah. But generally, like you've done quite well by just focusing on something, a category that you know better than maybe there's three or four other people in the world that understand the intersection of like money and technology the way that you do. And I feel like a lot of founders oftentimes will do one thing, feel like they kind of climb the mountain, and then some will just go back and climb a very similar mountain and do quite well again. Others want to jump ship and do something completely different.
So I tried both. My last startup in between the other payment startup was not in payments and it turns out that I'm very good at climbing this one mountain. I'm just going to keep climbing it because — but also the reason I'm good at it is because I love depth and I love getting deeper and deeper and deeper into payments, into how to make payments more expedient, more transparent in the case of Affirm. And payments are a lot of things and access to credit is a totally new idea. Like I've never touched credit until Affirm and now I, you know, maybe without false humility, I think we are definitely one of the very best credit underwriters in the world and are just getting better and better. Just announced yesterday, we've built an entire new model family loosely inspired by the attention mechanism that is powering all of your LLMs out there, including your parent company. Now, some ideas we borrowed from all these really amazing discoveries in sort of, you know, language AI, and yet it feels like we're just getting started. Like every time I look at the sort of corpus of things that we can try to do, things we can build, the products we can launch, it's like, oh my god, like there's so much to do. Like there's absolutely no chance I'll get to rest. You know, I'm not sure I'll get to after all. It's the greatest time to build stuff because everything is faster and more exciting and easier to start.
But the depth reveals itself the deeper you dig.
Okay. So yeah, we sort of mentioned, we referred to it a little bit, the 2026 Investor Forum. Who attended? What was the goal of that? Take us through that event, and then I have a whole bunch of follow-up questions about how you're communicating at the current time.
Yeah. It was amazing. I'm still kind of slightly high on the whole thing, but the thing that was really fascinating — so we did this event three years ago, or a similar event three years ago, and it's like a blink of an eye. Like I remember it was in the same building. A lot of the same people attended. So these are mostly buy-side and sell-side analysts, a bunch of our shareholders, a couple of our board members actually flew in for it, which was like a nice surprise to me. And the entire management team gets on stage and like, 'Here's what I work on. Here's what the company is.' But like the reason analysts are there, they're like, 'Okay, tell us about your vision, Max.' Yada yada yada. Really, what will the next three years' worth of growth targets look like? And so the main act — I was the warm-up act, I try to be funny — but the main act was our CFO, Rob, who got on stage and said, 'All right, so we're going to grow to $100 billion of GMV. We're just in the range of touching 50. We're going to double and we're going to do this by compounding at 25% every year. And we're going to move our profitability target from 3 to 4%, which is our current range, to 3.75 up.' And so we're going to up the floor. We're going to be more profitable and grow faster. And it's like, faster? We're a bigger company now. Like what are we doing? Wait a second. Last time we did this three years ago, we said we're going to compound at 20%.
But instead, you did it at 30-plus.
And so just like this out-of-body experience of like we're getting bigger. The room was significantly larger. Same building, different floor, different room, many more people, many more shareholders. And yet we're telling people like, hey, we're going to compound that much faster than the last time.
It's like I guess it's like a flywheel. This network you've built that keeps on spinning and spinning faster.
And so it was amazing in a sense that like I knew all these numbers, like I obviously approved and worked with the team to make sure we feel great about them, but when you sort of say that out loud, like compare that to the last time we told you the same thing. We're going to accelerate.
Like wow, that is like so cool. So anyway, I'm still very high from like being able to say that out loud. It's very fun.
So there's a flywheel. I imagine with a lot of financial companies there's economies of scale and so that seems doable. Are you also talking about TAM at this point because the company's so large or is it still early enough that that's not something that investors are coming to you asking to sort of justify, well, you know, the global economy is only this big and if you take 50% of it, well, there's nothing else.
The good news is that $100 billion of AMT in this country is but a drop in the bucket.
Okay, that's of an even larger bucket.
We're in no danger. The US revolving credit, which is obviously a fraction of total economic rotation, is like $1.3 trillion right now and $100 billion is like — then we're taking share. So I like to compare to that number. People are not using revolving credit, they're using Affirm, which is good for them, non-revolving, no fees, all that. And so we're not yet at the sort of TAM. By the way, these are all US numbers and we're already live in the UK and Canada and we've announced a bunch of other countries.
Okay. So speaking of the investors, obviously it's very important to communicate with investors right now, both from your growth plans and also what's going on in the market. As you think back to the March time period, the SaaS apocalypse, every company got sort of beaten up. You came back really quickly. How much of that was show versus tell? How did you think about communicating to investors through that? How important was the investor communication versus just, you know, continuing to deliver on the actual metrics? Like what as you're confronted with one of these like, oh, there's a new narrative, what's your playbook for actually working with investors to get them comfortable?
You know, to be completely honest, I don't know. I think the — well, and I'm hugely honest, so I might as well tell the truth every time. So this particular time we said nothing. We basically said, 'Okay, I guess the sky is falling. Doesn't feel like it's falling on us. We're printing a really, really good quarter. Yeah, we're about to report it. We could like wave our hands, 'No, like you got it all wrong. We're fine.' Or just wait a few more weeks and be like, 'Hey, here's what we printed. How about that?' And by the way, we're going to guide. So we're going to reveal that we're having a pretty good quarter too. And so that's what we did this time. And we just didn't spend too much time communicating how the story isn't true. I don't know if that's like the best response. Like we've definitely in the past we would read, you know, the most recent time when I thought we needed to speak was when the rates were rising very quickly. We were screaming into the void of like, no, no, this business is fantastic at ZERP. It is fantastic at a 5% Fed funds rate. It is just fine at a number that's higher than that. We love our value creation opportunities more or less at any rate. Like obviously, you know, it breaks at some point, but the US economy breaks at some point too for all possible values of the federal funds rate. We're going to be just fine. And a long explanation why, and I don't think anybody heard it. Like, 'Yeah, yeah, but the rates are up, so you're obviously.' And so the dip and return of our stock is like, 'Oh, you had a great turnaround,' but it basically amounts to people saying, wait a second, this thing is in just as much demand, does just as well, grows just as fast, just as profitably through whatever rate cycle.
Yeah. I think we've talked about the history of the effect of interest rates on Affirm. But we're now in a new regime and the fear is not higher rates, it's higher inflation. And so do you already have an intuitive sense of how a different inflation regime or different inflation environment will affect Affirm? Are you confident? How do you think about the relationship? It's like I can imagine it's somewhat of a double-edged sword in that like, you know, higher prices means people are more likely to want to use credit, but maybe there's a drop in some purchasing activity of some items, you know, also just like if you're getting $10 in two years and it's worth less, that should have an effect. But how is it actually playing out?
So a little too early to tell, but you are exactly right in a sense that we saw this game play out in the last inflation spike just a few years ago. We absolutely saw an increase in demand because as things became a little bit more expensive, people felt that it would be better if they weren't paying for them upfront. People who initially would say, 'Yeah, it doesn't really make sense for me to finance this thing. I have the cash.' Like, 'Ah, I kind of want my cash to go a bit longer because I don't know exactly which way the prices are going to go.' So we expect more demand. We have to be — we're always very careful. The number one line I give at every investor event is credit is job number zero. As a computer science major, I count from zero and it is the most important job. We grow as fast as credit performance will allow us to grow. Full stop. We have to print consistent credit returns, otherwise we lose credibility with our debt investors and that's the most important thing. So we will grow fast, no faster than credit results will allow us to. We expect more demand. We don't yet see literally anything related to change in credit performance. We just printed our results and they were just as we expected them. Yeah. And so TBD whether we see some deterioration of consumer credit due to prices, we really didn't see much of it at all through the last inflation cycle, which was quite dramatic if you remember. And so we feel pretty great about our ability to underwrite kind of all eventualities. We were obviously not blind to the macro reality the minute before I got on stage and promised people 25% growth with an increased profit margin. So you know we must think we're going to be okay. And that is in fact the case. Yeah.
How is AI changing how you think about international expansion? Does it make teams more efficient in terms of adapting the product for different markets? Everything from language to just simply being able to ship more code? Or is your international GTM just like much more like we know what it takes to launch a new market. If AI makes it slightly more efficient, great. But it's not going to change the strategy.
I think the one item on every list of things that have really changed thanks to AI is shipping code. Like I think this is the best time to be a CEO with a computer science degree because —
Like you feel what this means. Like you, you know what it's like to make code. You've done it in my case for the last 40 years. And I know how much easier it is, how much more effective teams can be, how you can have three people in a room build a product over the weekend, which is kind of an amazing thing that didn't exist six months ago. And so shipping code is absolutely—we're seeing that. In my latest shareholder letter, I actually showed an illustration of the percentage of code written by AI at Affirm. And it's not like a slow rise like, okay, we're trying—we're convinced, 60% to start, 75% last month. So it's just, you know, rocketing. It's not like people are writing less code or reviewing less code. It's just that much more productive. So that helps us internationally, helps us domestically, helps us with everything.
I think other parts of the business are much better with AI. So translation obviously, services are fantastically effective with AI. You can do a lot of interesting things with legal. Anything that's text-heavy, you have just, you know, a great helping hand. But the sort of night-and-day moment is writing code like that. That's at the top of the list of things that are just stunningly effective.
Are you expecting token cost to be like an important line item in the P&L going forward? Like we've seen because there's so many stories of a lot of code written by AI. Sometimes that moves the needle. Sometimes you're just sort of doing what you needed to do and doing it more efficiently or faster. But at the same time, if you get hit with like a billion-dollar token bill in a year, that might be material. And so you start having to do—and it feels like we're at the early innings of ROI calculations. We're more in like the feeling of like, oh, everyone loves this. We're seeing good results.
Max doesn't strike me as someone who's just going to tell your team to use a lot of AI because obviously there's stories coming out of Amazon and Meta where teams are basically just like, I want the optics to be that I'm using more AI than anyone else on my team.
Going to run things overnight for no reason.
Yeah, we are an extremely, extremely metrics-driven company. Like, some might take issue with just how completely obsessed we are with—like, you improve what you can measure. Like, if you're going on feel, you're going to, you know, feel great until you don't.
So we measure everything obsessively. We have, at the moment at least, fantastic ROI on our token spend. We are—I would say, you know, it's real numbers, so it's not like, oh, you know, don't care, don't look. We have a weekly report that's generated thanks to AI and some very smart humans who are monitoring our token use and tell us how each team is producing value and what it's costing us in tokens and all the sort of various conclusions downstream. So we are already very mindful of what it means in terms of return on investment on tokens and return to shareholders at the limit right now. It's like undoubtedly a very, very accretive thing, but we're also not assuming that, you know, like infinite budgets are just great. And we also expect, by the way, prices of tokens will probably increase over time. Like, I think there's plenty of subsidies taking place. You're now inside one of those places, so you probably know the true price.
If we can circle back to specialization from AI creating effectively more generalists, I'm interested in sort of computer science history, your career history. How have you processed—like, there was a point where front-end and back-end engineering were two very separate disciplines. At the point where you could do JavaScript on the front end and the back end, you see more full-stack engineers. Now you have engineers that are designers, designers that are engineers, designers that might be pushing server code and stuff. So I'm wondering how you're thinking about maintaining that idea of—you want the specialized genius, the artisan, the true expert in your organization versus you also want to empower someone to move really fast with a small team. And so you might not need to staff a hundred experts to do one thing. And there's got to be a tension there. So I'm wondering how you're grappling with that.
It's a great, great question and you're totally right. So, there are definitely pockets of today's software engineering that accrete profoundly to specialists. So if you are an extraordinary infrastructure engineer, you're far from being replaced by some mindless drone because it's as much art as it is science—intuition, experience. Like, I've seen this kind of a failure before. And AI is good at it, but AI is a great tool. Like, you're in no danger of being supplanted just because so much depth goes into that. And by the way, AI making a slight mistake could cost you a massive outage in the case of—and so the cost of error is very high. Like, an even better example actually in my world would be an underwriting model. So somebody who designs underwriting models—like, hey AI, go make me a great underwriting model. Come back when you're done. Like, here it is, I hallucinated one for you. Like, that's like real money lost that we will never see again if it in fact didn't hallucinate the right thing. And so many of these sort of ultra-specialist roles, they are absolutely benefiting from AI, but that human is not just in control. They're not just steering, they're like reviewing every line of code. They have another human making sure that there's absolutely nothing that could go wrong. So that's sort of one thing.
On the completely other spectrum of this, which is I think what you're alluding to—so, you know, fun examples, we just had a product hackathon. So only product managers can attend. A lot of our product managers have CS degrees, so we're cheating here a little bit. But these people were like, 'Hey, I love writing code, but I want to make it into things as opposed to code, and like, I want to ship a full product.' So most of those folks haven't written code professionally in quite some time, maybe since college. They're like really good with Figma, they know how to design things, they have a sense for taste, but they like—they don't remember anymore what it's like to sort of go deep into some, you know, SQL query debugging. We had 100 people, 37 projects in 27 hours. Went from a whiteboard plan to a shippable feature that was presented to the entire company like, hey, we have this thing. And we graded it not just on—so the winners and the whole grading, like who won first through whatever tenth, were all people that had to report not just like, here's my idea, here's what it looks like, here's my prototype, how close are we to shipping it, like, are you ready? And the winner was absolutely probably the best idea, but also the ones that said we're ready, like if we're allowed to ship it tomorrow, we will. So that is like an extreme generalist who like has a vision and a few tokens and a feature in 27 hours with like two of their co-conspirators and like a pizza box. So that is new and different and it's totally a new phenomenon.
This AI thing is going to be terrible for the pizza industry because famously, two-pizza team. You only need one-pizza teams now.
50% reduction. Short Domino's. This is a disaster. I hadn't thought about this.
We sprung for a somewhat higher quality.
Oh, yes. Okay. So, revenue is flat, margins potentially going up in the pizza industry. Okay.
Last question for you. How are you—what's your framework for evaluating all the different AI products being pitched to you across the rest of the organization? You have the benefit of being a technical, you know, CEO. A lot of other CEOs in your role are just like, I just want to buy AI everywhere. Just give me every part of my company. Just give me, give me, give me three pilots. I'll pilot anything. Whereas I think you're probably looking at some of this more non-deterministic work saying like, hey, you know, you can play around with it, but maybe you don't have budget.
So I have a great heuristic. This will be potentially worth the rest of my drill. If you know how to describe the evaluation criteria being used by the maker of the tool you're buying, it's probably worth piloting. If you can't, it is slop and you're being sold a story. So, we know codegen works and is amazing and it's only getting more amazing because we know how to validate code and you can use another AI to validate code, but you can also just like look at, does it work? Does it pass a unit test? So the feedback loop of, I gave you a thing—like, Codex was okay and then it was better and then it was great and now it's really pretty damn good. It's not because someone was coding faster, it's because it has really objective eval criteria and it just loops on making itself better with some degree of human influence. The same can be said for a bunch of industries. It cannot be said for others.
Like, I'm not going to dig into examples too quickly given we're short on time. But if you can say, oh yeah, I know exactly what these guys are doing—customer service centers. Is the consumer happy after they hung up the phone with an agent? Can we ask them? Yes, we can. Is there a way to say that was good, that was bad? AI, do more of the good kind. Fast feedback loop, high-quality eval criteria. So we use a ton of agentic customer service because we know it'll just get better and better and better. That's a great thing to buy. We're not a specialist. We are excited to buy it from the specialists.
Somebody tells you, 'Oh, it's AI. It's going to be a tool that's going to make you a better writer of fiction.' Yeah. Or even pitch deck. Pitch decks are an example because like you could have the worst formatted pitch deck and have tremendous success with it, but it didn't have anything to do with the deck. It had to do with your delivery and who you are.
I would not buy a pitch deck AI assistant even if somebody paid me to do it because it's 95% the talk track so far and the company and the idea and the market and all those things. And so, but most importantly, 100 companies pitch their decks, good, old, and made by AI. Seven out of 100 get funded. 93% failure rate. Sucks. Should never raise money. Like, is that the conclusion? So I would stay away from the non-objective or very slow feedback loop systems before I consider buying tools there.
I love it. Well, thank you so much for taking the time to come chat with us.
Have a great week. On the progress from the guide.
