Limitless Podcast - THIS WEEK IN AI: ChatGPT Built its Own Chip, Micron Soars, a24's $75m Google Deal

Episode Date: June 26, 2026

Today we're unpacking OpenAI’s plan to build its own custom AI chip—the Jalepeño—and what that means for its hardware strategy. We also cover Micron’s strong earnings, its ties to Anthropic, Meta’s next model timeline, and new developments from Google, Amazon, Anthropic, and Valar Atomics.------🌌 LIMITLESS HQ ⬇️NEWSLETTER:    https://limitlessft.substack.com/FOLLOW ON X:   https://x.com/LimitlessFTSPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQAPPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890RSS FEED:           https://limitlessft.substack.com/------TIMESTAMPS0:00 OpenAI’s AI Chip Move2:46 Compute and Hardware Strategy5:30 Micron’s Earnings Surge11:03 Meta’s Mythos Model Delay14:47 A24 Meets Google AI17:51 Sam Altman Movie Standoff19:02 Claude Tag Enters Slack21:10 Newsletter and Sponsorship Call------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

Open AI announced they're building their very own AI chip, custom designed for OpenAI models, Codex, chat GPT, but the craziest part was it was designed by their AI models themselves. Everything from the software stack to potentially even the hardware design itself was made by OpenAI's internal model as well as Codex, their coding model. It's going live in nine months, which is absolutely crazy. Typically these things take one to two years to at least design and build the first couple of prototypes. So the fact that Open AI is entering this market is the first real example of an AI lab owning the entire stack going from hardware all the way to frontier AI models. Now, in other news, we've seen a few rumors. Fable 5, the most powerful

model from Anthropic, might be coming back a lot more on that. And Micron absolutely killed their earnings report, blowing every bare expectation out of the water and expecting to make more revenue than Nvidia did in Q3 of last year. A lot of stuff to go through today. It's spicy little jalapeno. Look at that. This is exciting. It's fun to see the Open AI team move into, I guess, enterprise hardware is what we would call this. They're making their own chips. And who is the last company that made their own chips? We have Google, who has their TPUs, incredible. Amazon has Traneum, and now OpenAI has their own jalapeno chips. And it's important to note that they're going the same route as Amazon and as Google. This isn't a GPU. It's

an ASIC. It is a chip specifically designed for a very narrow task instead of something general purpose. Traditionally, when Open AI was to use Nvidia's GPUs, the bottleneck isn't actually the compute, but instead it's the memory that sits on the chip. So at any given time, they're paying for 100% of the GPU, but they're maybe using 70 to 80% of the capacity because they're limited by that memory bandwidth. What these accelerators are going to do with this ASIC chip, whatever this jalapeno chip is, the idea is that is going to be used for inference. It is going to be purpose built to supply AI tokens very, very fast and to be fully utilized. And that utilization, when vertically integrated, creates magical things. We always talk

about this on the show, but Apple and their M-series chips and how it was such a step-function improvement in every single aspect of the devices. It was like a day and night difference overnight when they released this chip. There's an opportunity for Open AI to do that now, because owning the stack allows you to squeeze at so much more efficiency than you have anywhere else. And it's very exciting to see them build a chip that looks something like this. I liked the idea that they used ChatGBT to actually accelerate the creation of this chip. I think it was nine months, which is pretty damn fast. And you have to assume that these two things are going to exist in parallel.

It's like as chat chip gets better, it is able to help with the tape out and the design of this chip. As the chip gets better, it's able to process tokens more efficiently and better. And it's this like kind of cycle that this flywheel that they're starting. And I think this is step one. I mean, it's very exciting to see them getting in the game. I think the entire story for this has got nothing to do with the chip and more so that the models were used to design the chip.

So the way I see it is the most scarce resource that every single AI lab that's fighting out for the number one position has right now is compute. And where you apply that, compute defines whether you're going to win the race or not. Now, for the last nine months or for the last year, at least, Anthropic has made it very clear that it's coding. So use the compute to train a better model

that's better at coding. If you own the best coding model, it can build pretty much the entire software stack. Now, emphasis on the software stack, what good is AGI or an AGI-like model if it runs on sub-optimal hardware? So seeing Open AIA make this move, realizing that they've used all their compute to train a model that can then build or design better hardware, that then potentially makes sure that they spend less money to run the same level of intelligence, means that they have more compute in the future to build whatever model that they want, and it runs on optimal inference or whatever hardware stack that they build.

So this to me is like that kind of like mind-blowing moment that I had, whatever, a year ago when Claudecote went live, where I was like, oh my God, coding is the future. Open Air has kind of like strategically made a very cool decision here where they're like, no, we want to own the hardware as well and we don't want to rely on Ambidia. And we're going to partner up with Broadcom. We're going to partner up with Media Tech, which by the way, like, these are companies, Broadcom is publicly traded. So these are companies that like could potentially rival Nvidia in the future to build their own chip. And the fact that they're releasing it by the end of this year is just a sign of the times that Open AI that doesn't have any scaling hardware experience has been able to accelerate this using their own model. So it's the first real proof that you can do this not just on the software scale, but on the hardware scale. It's just very cool.

Yeah. You know what other chip? Broadcom makes Google's TPUs. It's the same company that makes the TPUs that are actively working today. And Broadcom CEO actually said these are on par with the NVIDIA Blackwell chips. This is really exciting. I mean, it's cool to see OpenAI move into. more hardware. It feels like they're becoming this well-rounded entity. They have the consumer

hardware angle that they're working on with the Love From team and Johnny Ive. They have this enterprise hardware that they're using for their vertically integrated stack of actually generating tokens. And then they have this large software pillar that exists both in enterprise and consumer. And it's like, all right, open AI, you could like start to see them planting the seeds for these pillars to they grow up. You remember what our number one critic was for the consumer device, right, Josh? We were like, we're excited about this, but like, will they be able to scale this at all? and they've been able to do this for like the hardest technical chip architecture ever. It's just it's impressive, man.

I'm very excited about it. We also have other news that I must ask you about EJS, our resident investing consultant in the world of AI and memory, because I remember talking about Micron, like, I don't know, maybe a couple months ago, six months ago, maybe 12 months ago, and it was like a small fraction of what it is today. It seems like these memory stocks continue to just go up. Every day I'm looking at them plus 15%, plus 15%,

with no end in sight. And Micron just report our earnings, and they continue that trend. I think they're up 20% this morning as we're recording this. Yeah, that's unbelievable, 17.5%. What's 2.5%. Like, tell me, Micron, can you just actually zoom out for a little bit on six month maybe? Just we could see a little bit of the trajectory. Yeah, like that's crazy.

A 10x in six months. Yes, exactly. Yeah, exactly. 10x and six months. Oh, wait, no, sorry. That's $3 in six months. A thousand dollars of game. That's probably that correct. Either way, they're ahead of ourselves.

They're up a lot. So the question then becomes, what's going on with Micron? What's changed? Has something novel broken through in the AI memory trade? We've spoken about this a lot on the show before. It goes something like this. If you have an AI chip, if you're using AI models, if you want it to understand you, if you wanted to remember stuff about you, guess what? You need memory for it to store all the information about you. Now, there's two ways to store it. You can have memory on the chip itself, or you can have memory that's live in the session that you basically need a lot of memory. Micron is one of the key suppliers. They're actually number three, believe it or not, but they're the number one American supplier of memory chips, and they had their earnings report go live after market trading hours yesterday. And I'm not kidding, they blew it out of the water. Forty-one billion dollars in quarterly revenue. For context, that is around 15% more revenue than Nvidia, the largest, most valuable company in the world made this time last year. Their projected revenue for the next quarter is going to be around $55 billion, making $38 billion like profit off of that, which is, again, record beating way more than Nvidia made this time last year. To provide

context, they make one component for the entire GPU or chip architecture. So the reason why the stock is rallying is not just because it beat its earnings, believe it or not. It's because it disproved a very important theory that a lot of bears have on the AI memory trade, which is the memory trade has topped. There's too many people in it. This week, literally two days ago, the Korean Stock Exchange had to shut down because too many people were over-levered.

They were borrowing money and buying into this stock and buying into Korean memory stocks, SK Hinex and Samsung. And basically they said, like, this is too much. The trade is overcrowded. What the earnings report from Micron

revealed yesterday is that the demand hasn't even started yet. This week, they signed a very important partnership with, you might have heard of this company, Anthropic, they signed a pretty huge deal with Anthropic, basically saying that they're going to be the prime memory provider for all Anthropics' GPU co-partnerships going forward, and maybe if they build their own ASIC in the same way that Open Air just released or announced Halapeno, they will also use Micron. So the idea here is memory demand is going up way, way more exponentially than memory

optimization can happen in models. So let's say you create a new model that demands less memory, it doesn't matter because the demand for more people using more agents, etc, that require even more memory is there. And Micron is going to be the primary provider of that. That's why the stock is up almost 20% this morning. It would disprove in the best. In the case that even the AI train stops, there are other use cases for memory. It's like everything is requiring memory. And that next thing is that physical AI, the robotic demand. It's like all the physical infrastructure that's going to be built with AI, all of these things need memory. All of the autonomous robotaxis that are on the road need

memory. All of the future optimists and humanoid robots that are going to be deployed. They all need memory. Everything needs a tremendous amount of memory to process. Josh, you have a Tesla, right? Yes, I do. Guess how much more memory of Tesla requires than the average card that has a smart system? Oh, that's a great question. I have no idea. Do you know the number? Five to eight X, depending on the model. And guess what? The later models need more memory. So the trend basically is, as these autonomous vehicles go live and improve version over version, you need way, way, way, more memory. There's only three suppliers. Micron is number three. You do the math. That's unbelievable. What a great opportunity for Micron. And they're also doing stuff outside of just publishing great earnings.

They're signing deals with pretty large companies, the most recent one being Anthropic. You may have heard of them. It sounds like they are now officially working together. What is the capacity of this deal? What does the deal actually consist up? Do you know? Yeah. So basically, Anthropic has a ton of investments in data centers in general. And a lot of it is to do with GPUs, but in order to kind of make sure these GPUs work well together, and in the future where Anthropic wants to create their own inference chip in the same way that Open Air has created Halapeno, they need to design the perfect memory architecture and the custom racks to be able to suit the memory implementation into all these different chips. What better way to do that

than not hiring a staff team on your own, but partnering up with the number one memory or the number three memory company, but the number one memory company in the US. That is like completely beating earnings. So now what Micron's going to do is they're going to form a team. They're going to place it in Anthropic and they're going to work very closely with the Anthropic team on any chip architectures that they build in the future. And you might be wondering, well, isn't this a conflict of interest? Well, actually, Micron put in a pretty large check into Anthropics' recent series H round, I believe. So they're also part owners of Anthropic. So it's all kind of like this symbiotic kind of like thing that they're in right now.

Well, there's also another fun news update that we have to share because there is a mythos class model that is coming. And it's not from where you would expect. It's from meta, who claims they are on the way to their mythos class model. This sounds amazing. This is very exciting on paper. The problem is the timeline of the mythos class model. It seems like there is no timeline at all to what it's going to be going.

Is it nine months? It's not months. Okay. So in the time it takes to birth a huge. human child, meta will somehow figure out how to create a mythos class model. And it begs the question. It's like, well, if you're getting to mythos in nine months from now, then what is the future mythos class model going to look like in nine months? And that seems like a bit of a problem for them. I mean, that essentially puts them behind the open source labs, right? Because the open

source labs are six months behind. It's going to look worse than an open source model. So the irony is, you have meta that went from building and releasing models in an open source fashion to then becoming close source because they had this super special model that they were training, only to get out competed by all the Chinese AI labs that have a better open source model definitely in nine months' time. This is crazy. It's crazy time. Yeah. So, I mean, is meta-cooked? I don't know. We do have some good news at least on the meta front in the sense that they revealed some glasses. They have some glasses now, which I thought were actually great. The bar is pretty low, Josh. Come on. So here on the right, we have Evan Spiegel's glasses from Snapchat that were released.

this week, last week, I think. And they were horrendous. They were nothing short of just an abomination. Zuck tried his own version of this, which we know that it has been working on these smart glasses. And you know what? I think it's a home run. For what this is, I think they did a great job.

They partnered up with Kylie Jenner, who were seeing on screen to do the, I guess, promotion to be the face of it. She has a voice when you talk to the AI. And what I like about these is the lack of ambition within them. And I think they have kind of grounded themselves in reality here where a lot of people are trying to do the heads-up display, like the virtual display that requires a lot more technology baked into it.

These glasses, they look reasonable. They don't look any different than normal glasses, which is great. They have a microphone built in. They have the speakers built in, and they have two cameras on the sides. And it's enough to be that intermediary interface with AI that isn't quite your phone, but it still exists on your face and you could still engage with it. So I think, like, from a product standpoint, this is the most exciting offering to date as it relates to smart glasses, how useful are they? Not super useful. Maybe they're just a fun gimmick now, but I thought they did a really nice shop. You know what? I would buy this product if Apple released it, because, okay, if you're not going to give me a visual display on my lens, if it's just going to be basically

speakers and a microphone, it should feed into some sort of device. I'm not pulling up the meta app on my iPhone. Like, that, ew. Like, I'll just use whatever Apple creates in the future. Now, is it sleeker and is it cheaper than their Rayband display, which was, ironically, their flagship announcement from their hardware conference less than probably nine months ago last year? Like, yeah, cool, that's better. But is it actually going to be practical and useful to me? I'm not entirely convinced by it, but is it better than Snap Scarses and will I more likely wear it?

Yes. I heard a hilarious rumor that Can Leons is happening right now. It's like a big kind of advertising influencer type thing in France. And apparently Evan Spiegel is over there trying to convince Robert Downey Jr. To become one of his ambassadors for SnapSpex, he's offered him $100 million. And the fact that there hasn't been a confirmed headline out there just tells you how bad some of the snap glasses is. So the bar is low. But one of my major critiques of Zuck in the past for his hardware specifically is he has not got the supply chain infrastructure to build it himself.

And he's proven me wrong. So, you know, I raise my hands. I hope they could pull it off. I really do. There's another unlikely combination of, or just collaboration, we could call it, that is a little bit even more out of left field than Kyla Jenner and Meta. And that is A24 and Google. And yes, A24, the film production agency and Google the technology company, they have announced a partnership to now work together. And it seems like they're going to work together on using these AI tools to further develop the storytelling ability of A24. And when I first heard the news, I was like, that doesn't really make sense. What does

Google AI and Google DeepMind have to do with A24? And then I started thinking through like, okay, well, they have the world models. We know that they are excellent at world models. We know that they have their VO class models, the video generation models. They have the music generation models that generate pretty good songs and generate lyrics. They have nano banana, which generates the visual images. And then I'm like, oh, okay, well, actually, yeah, like all of the core pillars of storytelling can be made by these Google deep mind models. And they're actually like all pretty much sitting along the frontier in their respective capabilities. So I think this is a pretty exciting news update that 824, the people that make a lot of the movies that I think we all enjoy, they are now getting

deep mind models at their disposal. And I think that's, that seems pretty cool. It's important to point out that this investment deal doesn't mean that Google gets access to all the video content or copyright IP that 824 has. In fact, there's a strict line that's been drawn that, you You can't train any future Google VEO or video models on A24 content. That's not the goal of this deal. The goal of this deal is we want to be thoughtful in Hollywood about not replacing human producers and directors, et cetera. How do we build tools that amplify the work and taste that they have, the vision, the ideas that they have? That's the whole vision behind the tweet that you're seeing from Scott Belski, who is a partner and I think founder of,

of 824 labs, which is their kind of like research division. And they decided that Google was the best partner to work with. Google has the capital. They have all the data to kind of like feed into like what the best tool might look like. And they have the time. They have the patience, right? You know, they have YouTube. They have a wealth of resources that they can pull from that might help and aid A24.

Now, A24 has a pretty impressive roster of investors at this point, Josh. They have thrive capital on their cap table. and now they have Google, they're valued after this recent investment from Google at around $3.4 billion, making them probably one of the most valuable producing Hollywood Studios. There is out there. Pretty crazy. Speaking of Hollywood Studios, did you know that there is a film that was made already and completed about Sam Alman? He has a biopic named Artificial.

And I had no idea. Like, today I learned that there has not only been a movie about him, but it's already been made, filmed, produce, edited, and it is ready to be viewed by the world. The problem is that the company that made it is not quite ready to reveal it to the world, because I assume it does not show Sam under the greatest light, perhaps. We saw this happen once before with Zuck and the social network movie. I assume this is kind of generally along those lines. And it appears as if they've hit a roadblock.

Why is there a roadblock, E, jazz? Why is this deal getting stopped? Why can I not watch this movie? Well, that little company that produced this film goes by the name of Amazon. They spent hundreds of millions. They spent hundreds of millions of dollars. You might have heard of Amazon Prime. They distribute to millions and millions of people all over the world.

So this was a big investment for them. They hired Andrew Garfield. They put in a ton of money to make this film. And then they signed a $50 billion enterprise relationship, cloud computing specifically with open air to serve their models through Petrog. And so they were thinking, if we release this movie that shows Sam Altman in a bad light, that might damage our enterprise relationship.

So what we're going to do is we're going to cancel the release of the show and we are going to start shopping it out to other name labels, other producers, one of which was A24, and they said, you know what, we can't do this because Thrives on our cap table and Thrive is one of the biggest investors in open air. So it's interesting to see the politics over here.

These AI labs are getting involved in a very meaningful way with these Hollywood producers and it restricts what they can actually make. Okay, and to round out this week, there is a new feature from the anthropic team named Claude Tag. Now, Cloud Tag is a brand new way of working with the AI in a way that I think surprises a lot of people. We have this traditional LLM infrastructure and interface that you can work with it.

We have this claw similar to like an open claw operating system. And this is now a new third way of engaging with AI where it's built right into the place that you do your work, into Slack channels. And what's funny about this is it creates this multiplayer way of engaging with the AI and actually writing code and accomplishing tasks as a coherent workforce. A cool thing that I picked up from the launch video was that 65% of all of the code that Claude Teams are producing is actually made through this new tool that was previously internal and is not made public, named Claude Tags. So this is something that I think a lot of people who are in the workforce are going to be excited to try.

It's a very novel way of kind of multiplayer tag teaming tasks and goals that are required through just general day-to-day work in Cloud. And I think it's really fun. It's like you tag in Claude as if I were to tag in you, EJAS. hey, we need to get this agenda done for the show today. You just tag in Claude instead, and it has all the context, it knows everything from within the internal company database, and is able to actually go off and complete tasks.

It's pretty cool. Yeah, just to be clear, like, this isn't a Slack update. This is a new way to work with Claude Code and Claude itself. What I like most about it is it treats Claude like an actual person. Like, we go from like this LLM chat bot to now like a person that you can trust to independently do work. That stat that you just quoted the 64. This is real product people with ideas of things that they want to build, tagging an LLM,

and the LLM gets what they want to build. And they're like, okay, I'm going to go away and build this POV and I'll bring you back the mock up. I'll bring you back the kind of functioning model. And you let me know if you like this and then I'll go away and do it again. So completely hands off. Like most of the software engineers that are getting paid hundreds of thousands of dollars, now don't actually handwrite code anymore.

It's just very impressive. And I look forward to seeing the best ways to kind of like use this specific tool. But that brings us to the end of the docket today. We have one final announcement, which is equally as important as every other thing that we're spoken about on today's episode, which is we have a newsletter. It goes out to 100,000 of you. And Josh just wrote a banger of an essay that's going to go out. And we also include the five weekly hires, the top news items that you need to hear about every single week. That goes out on Friday.

There's also one other final thing, which is Josh, myself and producer Luke are in the market for, a sponsor. So far, we've been keeping the lights on ourselves, funding it ourselves, and we've been trying to figure out another way to kind of work with brands and partners that can, you know, kind of have a symbiotic relationship with us and keep the lights on and produce the content that we have every single week, four times a week. So if you are out there, and you'll listen to this and you think you can help us, or if you know of anyone that might be able to help us, please reach out on our emails or via X. We'd love to hear from you. Yeah, feel free to reach out to us on X. We'll leave an email linked in the description that you

find, leave a comment, we can reach out to you, whatever it may be. But we are very much in the market and very much appreciate your support. If you have made it all the way through to the end of this video, congratulations. You are now fully caught up. Thank you. One last update. I just want to shout out the Valoratomac's guys real quick. Isaiah Taylor, he was on the show. He was an early guest the show. Very bullish on them. Just earlier today or yesterday, they actually got their nuclear reactor up and running for 24 hours straight. And it works. The damn thing works. So congratulations to them. We were early on this guy. If you were an early day one limitless supporter, you knew all about this company over a year ago probably and now they're at the frontier of energy creation

so it's very exciting it's cool to hear thank you all so much for watching as always if you enjoyed share it with your friends give us the five-star rating all the good things and we'll see you guys next week thank you so much see you guys bye