Limitless: An AI Podcast - This Week in AI: China's Lithography Heist // The Lunar Mass Driver // $15 Billion Deals

Episode Date: December 19, 2025

ASML has a clear lithography market dominance, but it faces potential competition from a Chinese prototype. We introduce challenger Substrate, highlight Elon Musk’s lunar mass driver concept, and critique Waymo's fundraising amid competitive pressures.We also analyze Amazon's $10 billion investment in OpenAI and the challenges of sustaining AI operations, along with Google’s Gemini 3 release and OpenAI’s image generation model.------🌌 LIMITLESS HQ: LISTEN & FOLLOW HERE ⬇️https://limitless.bankless.com/https://x.com/LimitlessFT------TIMESTAMPS0:00 The Invisible Monopoly2:17 China's Breakthrough4:48 Competing Technologies5:24 Dependency on ASML9:44 Lunar Mass Driver10:52 Energy and AI11:45 Waymo Fundraising14:19 Amazon and OpenAI15:14 Circular Economy16:47 Gemini 3 Flash23:08 Advances in Organ Preservation26:40 Recap of the Week------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

Imagine if Brad kept all of humanity alive and there was only one oven maker and each one of those ovens cost $200 million and they only make about 40 to 50 a year. And they're all based in the Netherlands. This is the reality of our tech industry. Your iPhone, your laptop, your smart tablets, your TVs, anything with electrical circuits is run by this mysterious company that sits at the top of everybody called ASML. We've essentially built the civilizational Jenga Tower. They don't own 99.9%. They own 100% of the lithography market. What is the lithography market?

Well, it is kind of upstream of everything else. And I think that's what we're going to start this episode of talking about today, is this invisible monopoly that no one is really aware of that even exists. Everyone thinks that the most important company in AI is Nvidia. And if you dig a little deeper, everyone says TSMC. But the truth is, this company, ASML, is the only lynch. It is the most important company that the entire stock market is currently based on. And they're based in the Netherlands and they do this really interesting thing,

which is called EUV, which stands for extreme ultraviolet lithography. It is a way to basically design very intricate patterns on chips that are used to train frontier AI models and inferencing and even on your mobile phones as well. In a really specific way, it uses a bunch of laser beams, gases and stuff like that. and it's an incredibly complex and expensive process. And this hasn't been something that they've been able to crack or any competitors been able to crack for decades until this week, where it was revealed that some ex-ASML engineers that were hired by the Chinese government

have been able to recreate a prototype of this lithography machine that they have. Now, Josh, here's some crazy stats. This single machine costs about $200 million. to create and then to maintain even more, the company itself is worth $400 billion, and they only have a couple of these machines. So if you do the math, if one of these machines get recreated, you've now taken over tens of billions of dollars worth of market cap and potentially even more if you can create AI chips that are of the same level as Nvidia, and that's what China effectively has right now.

It's funny because you hear the $400 billion number, and it doesn't even seem that high for how existential this is. It should be real. They're the total linchpin when it comes to creating any sort of electronic device. And it goes, I wonder if that's part of the downstream reason why they haven't been disrupted thus far, because the opportunity isn't in the trillions, but it's coming now. And if it's not from China, it could possibly be coming from the U.S. because a lot of people are coming and trying to tackle this all at once.

So this is all amazing stuff. Ejas, how were they actually able to pull this off? Well, this is the thing. No one really knows. And there's kind of like a gray area where we're like, maybe they stole the secret formula because Josh, this entire machine weighs 450,000 pounds. That is an incredibly heavy machine and it is very, very complex and there are tens of thousands of people that are responsible for different parts of this machine, the secret formula which collectively

makes up the makeup of this machine. China somehow have got their hands on it. So there are some rumors that are floating. One tweet here says, my sources on the ground says they say, stole the source. Like they used a bunch of other versions in the past decade, and it goes on to explain how they've probably got moles. Yeah, I mean like people that they've planted within this company throughout several years that have been able to take these secrets. It sounds very tinfoil haty, but to kind of zoom out and look at the wider picture, China, if true, has built an effective working prototype of extreme ultraviolet in less than five years. And he compares it to the American venture capital space who have been trying to crack the same problem for over a decade now, Josh,

and no one's been able to crack that 100% monopoly. So the fact that we've finally made this breakthrough is awesome, but the fact that it's also a Chinese venture that's been able to do this is kind of bearish. Yeah, well, I would say there is another competitor in the ring here. It's not just China that's trying to take ASML's market. It is also a company called Substrate, which recently raised a bunch of money. They were kind of announcing what they're doing. And Their take on it is trying to reinvent, if you think of these fabs as printing presses, they're trying to reinvent the printing press. And their bet is that they are going to take an advanced X-ray lithography, like we mentioned

earlier, but turn it to particle accelerators. This is this really crazy, technically complicated thing. But I do want everyone to understand that they are working on it too. And it appears as if this monopoly is kind of being broken into two. One, of course, is China. One is the United States. We each have a leading company working on this technology. technology to dismantle the singular monopoly. So who is going to win that race? I'm not sure,

but it certainly seems right that you should be diversifying if we are so reliant on this singular company for every single chip that we make across the board. Like you said earlier, Nvidia is not the bottleneck. The TSMC is not the bottleneck. It is this weird company in the middle of nowhere that makes these chips. That is the actual bottleneck. It just blows my mind how dependent the entire stock market it is, Josh, how your and eyes investments are. The world is dependent on these two companies, or in this case, this one singular company. And just to highlight how important of an opportunity this is if a company or country is able to crack the secret formula, you have so many other companies that are focused on this.

Samsung, in fact, has been at this for over a decade with no luck. They made some recent announcements this week where they're creating a new fab that is going to focus on building two nanometer CPUs. Remember, that's CPU, not GPU. So they're still about five years behind getting to the same level that this company, ASML, is. So if China has in fact cracked this, it would be a miracle. We still don't know how they've done it, but crazy to see. Moving on, Josh, I saw this tweet by Elon Musk this week, and I had no idea about it. And you said, we have to talk about this on the show. I have to. I love it. Walk me through it. Okay, this is called a mass driver.

And a mass driver is not even a new idea. This was proposed in the 1970s at MIT. And one thing I found about sci-fi is that oftentimes, I mean, it could be wrong, but when it's right, it's very right. It's just right on the wrong time scale. It seems as if that scale and that time window is finally getting close to where we are now. So on-screen you're seeing a visual of this crazy thing called a lunar mass driver. It was so fascinating.

And just when you were first reading it, you were like, whoa, this is real. and I think that was the reaction that I had too. Basically what is, it's a giant solar-powered electromagnetic catapult on the moon. So you can fling cargo into space without rockets or fuel. So if you imagine building this for like really long track on the moon with superconductor coils, you load it with chunks of lunar material, and then it uses electricity to yeat them into space at 5,300 miles an hour. With no rocket engines, no fuel, it's fascinating.

And why can we talk about this? Again, all of the most interesting conversations that we're having recently, are downstream of the fact that starship has been able to get actual mass to orbit for a low cost or the assumption is that this will happen in the next year. So the idea is that you can go put this on the moon. Why the moon? Well, because it has low gravity. It has one-sixth the gravity of Earth, so you need to spend a lot less energy to throw stuff off. There's no atmosphere. It has a ton of sunlight for energy. And naturally, the question I asked myself is, okay, well, why should I care about this? Why does it matter? And the reality is the economics of making this a reality are pretty

staggering. Yeah, I was looking into the to math before we came on the show and the Falcon 9, the old rockets from SpaceX, costs about, it costs about $3,000 per kilogram to get mass into lower but Earth. Starship targets $100 per KG, which is a massive, massive reduction. And so if you were to use a lunar mass driver powered by solar and running continuously with zero propellant, well, an early study in 1979 estimated that it would cost one, So the point here is that the cost of bringing mass back and forth to Earth into low orbit space would come down to minimal, would come down to zero, which is just insane. So then the question becomes, well, what does that timeline look like? And the post goes on to describe that we're roughly going to target the mid-2030s to have something like this functioning. So, Josh, this is something that's going to be pretty much realistic and achievable in our lifetimes and not far off when we're old men.

rather than something quite young within like the decade, which is just insane to think about. Yeah, we filmed this episode about SpaceX on their IPO last week. And I would highly recommend watching it because the idea is that SpaceX going public and getting the resources required to go and build in outer space really changes the fabric of reality that we live in today. Like these sci-fi concepts that have been ideated on for the last 60 years almost are able to be put in reality from at least a first principles perspective where if you're able to get this amount of mass payload to orbit on the moon, ideally you can turn that into one of these mass drivers that then offsets the global economy in a way that is the thesis for this universal high income

where if you can get this abundance of resources back to Earth cost effectively and efficiently, well, that really disrupts a lot of the financial systems we have today and it gives us this level of abundance that is almost inconceivable to someone who doesn't read sci-fi on a regular basis. So to me, this is exciting because, one, the math checks out. And it's really just a matter of making the right set of decisions over the next decade or two to turn this into a reality where we do have a moon base and then eventually a Martian base and we have AI supercomputers in space. And this world of space is just starting to be unlocked. And it's going to unlock so many unbelievable ways that we can take it. I like your point on abundant energy being the key to unlocking

future progress, Josh. The key to everything. It's the only thing that matters. Yeah, I was listening to this interview with Demis Hibis on the Google Deep Mind podcast this morning. And one of his key takeaways was, one of his learnings, rather, was he plays AI models like a game. And this game is to solve certain subroutes, is what they describe, or what he describes. And he describes a subrout as something like solving AlphaGo. And if he solves that game, then the AI model is able to solve other games. And he said, he was then asked, what's the most important?

important subroute to solve right now, Demis. Any answers with one word, energy. If you can create abundant energy, you can create an infinite type of progress and world. And on the topic of Google and on the topic of energy, we have an update. Yeah, bring it back to Earth. There we go, all the puns. We have some new news from a Google-owned company. Waymo, the self-driving car company, is in discussions to raise more than $15 billion at a valuation near $100 billion, which seems pretty cheap to me, but we'll get into that in a second. But the most important part is being led in a round by its parent company Alphabet, aka Google. I don't know what to think about this, because I thought Google was loaded, Josh. You would think so, right? That a company with hundreds

a billion on their balance sheet, who owns the entirety of the company would want to just double down on their investment, but they have not. They are looking to raise another $15 billion. That signals to me, perhaps a lack of confidence. EJAS, do you have a different take? Because to me, I'm like, well, we just recorded an episode about this. Waymo is the clear loser here. Does Google recognize that and are they hedging their bet? Or is there something else going on behind the scenes? Okay, so I'm trying to put my head into the head of Waymo. product division and his pitch deck. What's he saying to justify a $15 billion raise? Well, I'm guessing he's going to say, let's spend $10 billion of these dollars to purchase more Waymo cars

so we can expand our network. The issue with that is each Waymo car costs $150,000 versus a Tesla which will cost $30,000, maybe even less because everyone else that buys a Tesla can effectively add to the Robotaxi network. So I don't see a justification for the $10 billion. Now, Now, if I put my tinfoil hat on, Josh, I'm thinking the only reason why they're taking external capital is to, one, de-risk the venture itself and two, suck the capital out of competitors. So instead of the venture capital money going to competitors, it would go to Waymo to fund their expansion. I don't really see how this is going to be sustained. And, you know, listen, it's a great 2X for anyone that got involved in the 2024 round. But I'm not really kind of impressed by it, I guess.

boring. I mean, Waymo's got a long way to go if they want to compete with Tesla. And that's just the harsh reality of it, where there will still be a business but is it capable of growing. I do not know. But that is not the only big fundraise we have this week. We also had news in our circular economy section of the show. It's circling. It is circling like a bunch of vultures. Vultures ready to feast on an infinitely expanding bubble that has grown by another $10 billion this week as it relates to open AI and Amazon. Ijaz, you have been very bullish on Amazon. So please explain the bull case for this deal, at least.

Why are they making this deal? I am going to give you the bull case, and then I'm going to give you the slight bare case. So the bull case here is open AI needs a load of compute to train their AI models to compete with the likes of Google Anthropic, which, by the way, they are still losing to. No surprise is there. If you were to guess or predict in January who would win,

you would say Open AI and now everyone is shocked. So one way to achieve that is, okay, what if I told you this, Josh? What if I told you that you could get the same frontier level intelligence for 50% off? Would you take it or would you leave it? I'm taking that in a heartbeat. Okay, that is effectively what Trinium 3 chips, Amazon's latest AI chips, can afford you if you were to kind of enable that via Amazon. But there's an issue. Open AI is losing $12 billion per year.

Now it's probably going to ramp up even more next year. And so they need to figure out a creative way to get their hands on their chips without spending the necessary money. Insert this idea, Amazon invests $10 billion in Open AI, which they're probably going to spend that money on Trinium 3 chips. So this, which brings you to my bad case, the circular economy of these deals is insane. Why? Because it's going to boost their valuations even more. Josh, I don't know if you saw, but rumors broke this morning that Open AI is going to get valued at $800 billion. they're raising another $100 billion. So I'm just like what is going on here?

Everyone wants in. And the Mario Ground continues to spin the magical, the musical chairs continue to go around and we will see where everybody stops. Yes, cue Michael Barry meme. This is a good post for those not watching. It's Michael Berry and it's a quote with the Amazon investment would help OpenAI afford commitments, including from Amazon Web Services, which is ironic because Amazon Web Services is just GPUs and cloud servers.

and work compute. But, EJA, as you did mention, a 50% improvement in your model. And for users of Google, whether you're a builder or a consumer, your model has just gotten 40% better and 50% cheaper overnight with Gemini 3 Flash. The best model in the world just got better. This one appears to be the best pound for pound model on Earth by a fairly large margin. And you have pulled up on screen the ARC AGI prize chart, which I love to look at, because it's just clear.

It's such a good chart. Because it shows like the Pareto Frontier. And we talk about this previously where there is these tradeoffs, right, between cost and quality per token. And what you're seeing with that green line that goes straight up is that Gemini 3 Flash is really on the most vertical trajectory out of any of these. So you'll see it falls right in line with GPT 5.2 high. But it costs less than, what is that,

a full order of magnitude less per token. Yes, it does. It's crazy cheap. Crazy. Yeah. I think the cool story about all of this, Josh, is the process of distillation, which is, okay, you spend tens of billions of dollars to train a frontier model. And then what you can do is you can distill all the intelligent components of that model

into a cheaper, more, like less costly model and then scale that out to whatever product that you have. And who has the most product of distribution? Google does. So if you can surface that to a bunch of different people, let's say through, what's their number one product? Hmm. I'm racking my brain. Oh, of course, it's Google Search with 5 billion users.

That would be, you know, that would be an attractive proposition. And guess what? Like with this update or like a story that kind of went under the waves, but we're going to highlight here is this domain called Google. dot AI, which now effectively embeds Gemini 3 Flash into Google Search. Now, there's a few steps involved here, but the simple fact that you can now go to Google search accessible to 5 billion different users and get access to AI mode with frontier level intelligence at a fraction of the cost, I just don't see how Google loses this at all. They have the

they have the distribution. We've said this multiple times before. We've done a Google Bullcase episode. you definitely go check that out. But now they've combined the frontier model with this, Josh. Like, they're like cannibalizing themselves in a way. Yeah, we're starting to get the answer to what do you do in the case of the founder's dilemma where what happens when innovation eats your core product. And that's what's happening. And Google has decided to double down and fully lean into it. And we are going to see how that plays out. I mean, this is Google.a.i. This is not Google.com. So their core search product is still intact. But I imagine this is a transitionary. period where they are testing. There is currently the AI mode button in Google.com's homepage,

and it's only a matter of time until they're able to, until they commit to fully switching over. And then you have to ask the question, well, how are they going to support this business? And is it just going to be an effort to bleed out the results from all these other companies? Because Open AI needs to make revenue. They must make a ton of money. In order to do that, they're going to need to start implementing advertising and sponsorships into their products. Google has a very different business model. They are a behemate. They have a tremendous amount of cash on the balance sheet that they can use to subsidize

all of this compute and all of these AI needs that people need. So they have TPUs that are hyper-efficient. They have the best, most efficient model in the world, and they have no need to monetize urgently because they have such a huge cash balance sheet. So OpenAI and other companies that need to compete on this run are going to start to have to make some trade-offs here. You might be thinking as well, why doesn't Google just switch search completely to AI if it's so smart? And the simple answer is they haven't been able to embed advertising and search rankings into the model effectively yet. So the same kind of experience where you see sponsored posts at the top of your Google search results page, you can't quite do effectively with AI models.

It's a completely different beast. And both companies, Google and Open AI, will need to figure this out eventually. I think all roads lead to ads, but the question is, at what time does that make sense? If Google can hold out to your point and Open AI kind of comes in first, Google wins the moat because people just want to use an ad-free product. It is crazy.

But Josh, on the topic of Open AI, did you see this news? They've introduced GPT, chat GPT image 1.5, which is their answer to Google's Nanobanana Pro. And as everybody knows, we, of myself, I'll speak for myself, absolutely adore Nanabanana Pro. It is one of my favorite models of the year. It is an excellent image generator. And as I was looking through this announcement, I was hopeful that this would be a better replacement.

It's built into the chat GPT app that we know and love. And I think the reality of it is it is just not that good. I mean, in the video, you just have just a... Oh, we start right here with this example. So for those who are watching, close your eyes. If you're not, what we're looking at on screen, and I don't even know if you should scroll down any further than this, is a post from Mr. Sam Altman himself, shirtless as a firefighter, in front of a Christmas tree. And if you scroll on even further, it's actually a holiday calendar. And it's an example to reflect the quality, I guess, of the new 1.5 image model. It's good. It's not great. It would appear as if nanobanana is better after going through a bunch of examples. Nenomana Pro still is good. But what's cool about this is if you don't use Google, if you're you're not familiar with their work flow if you don't use Gemini, if you only use ChatGPT, you have a pretty good imaging model because this is built right into the app.

And what you'll notice is EGIS is it's really good at faces. So we're seeing on screen a picture of Sam and Ilya kind of a little sad at a party. It's pretty good. It shows faces in a world where they didn't show faces before, and they're fairly accurate. So it's an attempt. It is by no means a Smackdown on the King Nana Banana Pro. But hey, it's pretty cool. And if you use chat GPT, your image gen model just got much better.

Listen, call me a Google Bowl, but like I don't think it's that impressive. Like, we're looking at 1.5, GPD 1.5 on the left, it just looks so much grainier. Like the example on the ride by Nano Banana looks way more accurate. Like you can see the physical expressions on Ilya's face as well. It looks more like Ilya as this post claims. I don't know if this is a zinger. And what's important about this is Sam initially. code read a week and a half ago.

And since then, he's done something really impressive, which is divert all resources towards number one, creating the best model, and number two, creating the best image model. Now, he succeeded on number one, releasing GPD 5.2 last week, which broke all frontier benchmarks, again, surprise, surprise. But this image model, he's kind of flopped.

So I don't know if we've reached a point where Open AI still has the edge that they once had at the start of this year. I don't know. It still leaves me uncomfortable. I don't believe that open AI has kind of nailed it just yet. Yeah, they haven't. But if you are, again, if you're a user, go try it out. It's better than it was before.

And if you're being too stubborn to go use Gemini, well, this is the best image model on the block for you. Now, I guess we can head over to a little more frontier science-y stuff. Josh, how am I going to live forever? Okay. Can you tell me? I cannot tell you the answer to you living forever, but I can give you an answer to your organ surviving longer if they ever leave your body for whatever circumstances that might require. There's a brain computer interface startup called Science Corporation, which we love, founded by

Max Hodak, who is the former Neurrelink co-founder and president. And they decided this week that they were going to announce they're moving into organ preservation and life support tech. Basically, it's like the hardcore biomedical engineering mindset used for brain implants is now being pointed at a different problem, which is keeping organs alive outside of the body for way longer. And I had to learn a lot before talking about this because right now, I didn't realize that organ transplants are in a pretty brutal race against the clock. They're, normally they last like a day, sometimes less if you have lungs or kidneys. It's very expensive to transfer these from place to place. A lot of time, it was done via like private jet or really like whatever fastest way you can

get it from one place to another because the person receiving it is normally in very bad shape and the person giving it. The organs are just not capable of living that long outside of it. So what they'd shared today is this breakthrough in how they're able to use this fancy new device to actually preserve organs for longer to make them last for days and hopefully soon weeks instead of just hours. And I found it to be really fascinating just on the scientific frontier front because, again, they're using AI in a industry that has not really used AI very much in the past to create these really cool new contraptions. And you could kind of see an idea of what the photo looks like in this post here that can actually help save people.

people's lives and make a big difference. I mean, traditionally, it cost $250,000 for a machine that does this plus tens of thousands of dollars per use. And what they're planning to do is get that cost down to $10,000 and even less. And I think it's just a really fascinating exploration into what's possible in the world of science now that we're applying AI to a lot of these difficult problems. Josh, to give you a kind of crazy example to demonstrate how important this thing is, in a past life, I studied biology at university. That's right. You're a pro on this. Well, I had a crazy opportunity to perform heart transplants between murine mice.

So these are like test lab mice. It was a pretty crazy opportunity. I signed up for it. Somehow got approved because I was studying biology at the time. I loved genetics and specifically regeneration of cells. So I qualified. And the TLDR is I performed four of these things and it failed 75% of the time. The craziest part.

the mouse I was transplanting the heart into was right next to me. It was literally five inches away. And the fact that you had that failure rate just goes to show you how crazily complex it is to keep these organs alive and sustainable. Just a fun anecdote. It's really challenging.

If you scroll down a little bit, we could kind of see what the device looks like in full. It's interesting. The thesis behind this was that, like, why can't you ship organs long distance, like luggage would be? And what they wanted to do is create this backpack-sized system.

So their prototype that we see on screen, it has integrated sensors for oxygenation and flow pressure or temperature. And it's this closed-loop system that automatically adjusts depending on its outside environment. And it's really cool because traditionally, it required something much large than this. That was much more expensive.

That was single use. That didn't allow these to survive for much longer. So just an interesting thing worth checking out. We'll add a link to it in the description if you want to read more. Just something worth paying attention to that health and sciences, thanks to AI, are becoming a lot more.

exciting. It has been a crazy week for science in general, or rather sci-fi. Sci-fi's had a big win. Big week for the sci-fis. Yeah, yeah, we've got the mass drivers. Elon's launching satellites to train AIs. We're harnessing the energy of the sun. We have got this crazy device that can extend the lifespan of organs. And on top of it all, down on Earth, we have one singular company that has been discovered, which is upholding the entire world right now. Pretty, pretty insane. That is the end of another roundup. on the limitless show. Another huge week. We have some good weeks coming up too. We have some great weeks coming up. As is normal of the end of the year, we've got to do some reflection. Josh, you and I have been doing a bit of reflection recently. And we have filmed a banger of an episode which will tell you about the top wins and most importantly the biggest losses in AI this year. And then, of course, you know, Josh and I aren't short-term kind of thinking types of people. We like to put in predictions as well.

And we have a lot of big and bold predictions, which I'm pretty sure 99.9% of you guys won't guess at all. For 2026, that's coming out next week as well. We're still going to be shipping episodes during the holiday week. Listen, you can do your holidays, but you still need to tune in. You still owe us at least 65 minutes per week to listen to these episodes. Josh, do you know what I am going to ask of people's New Year's resolutions? I'm going to hope it's to be to share limitless podcast with at least 100 of your closest friends each.

I am going to say no more. Josh read my mind. There you have it. It means a lot. It's been a good year. We have some fun and different and interesting content like EJaz was mentioning earlier that's coming in the coming week. So you're definitely going to want to tune for that. It's fun. It's different. EJaz actually wore a tuxedo in one of them. So I promise you you're going to want to watch this to see why. But it's really been fun. It's been a great journey. And it's all thanks to the people who are so generously leaving comments and sharing with their friends and we read almost almost everyone i try to there's been a lot which is a really good problem to have but trying to keep up with everything um the support really means the world so if you did enjoy this please share it with

someone rate the episode five stars and thank you as always for watching and we'll see you guys