Limitless: An AI Podcast - Elon's New Chip Could Actually Dethrone Nvidia (Tesla AI5)

Episode Date: September 16, 2025

In this episode, we discuss Tesla's groundbreaking AI5 chip, touted by Elon Musk to be 40 times more powerful than its predecessor, and its potential to revolutionize products like Robotaxis and the Optimus robots.Additionally, we highlight the Megablock energy system, designed to unlock 50% of unused energy and revolutionize the energy industry. Join us as we examine Musk’s ambitious vision for a synergistic Tesla ecosystem.------🌌 LIMITLESS HQ: LISTEN & FOLLOW HERE ⬇️https://limitless.bankless.com/https://x.com/LimitlessFT------TIMESTAMPS0:00 The Rise of Tesla's AI Chips1:33 The AI Chip Industry9:32 Batteries of the Future16:15 Powering the Future with Megapacks20:44 Revolutionizing Energy Distribution22:16 Community Engagement and Feedback------RESOURCESJosh: https://x.com/Josh_KaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

Elon is playing Game of Thrones yet again, this time releasing a new AI chip that is going to power his Robotaxi, his new Tesla Optimus Primes, and all of his devices and cars that he launches right now. And the reason why this is so important is this is going to contend directly with Nvidia. He's coming for Jensen Huang's throat. But in other news, he also released this new product called the Mega Block, which consists of megapacks, which can power everything from homes to massive. towns. But the reason why this is so important is it's going to unlock 50% of the energy and power that the US and the world currently does not have access to. And that is so, so important. But starting off with these new AI chips called Tesla's AI 5 chip, which is supposedly meant to be 40 times better than their current AI 4 chips. But don't take my word for it. Take it from the man

himself. And we're finalizing the design of AI5, which will be an immense jump from AI4. By some metrics, the improvement in AI5 will be 40 times better than AI4. Wow. So 40% 40 times. And this is because we work so closely at a very fine-grade level

on the AI software and the AI hardware. Josh, I know that you watch the entire presentation for the AI5 chips and for the Transformer stuff. I want to hear your take before we go any further. Yeah, we'll start one thing at a time. We'll do the AI chips first because those are freaking awesome. So AI chips have been a big part of Tesla since the beginning of the company. When you require autopilot in cars, it means you need a lot of processing, a lot of inference power on these chips to run locally on the vehicles.

AI5 is the natural extension of what they've been working on. So if you've owned a Tesla, you're familiar with the Tesla. They started with AI1. They went up to 2.5, 3, 3.5. Now we're at 4, which is currently shipping on all. vehicles. Hardware 5 is a slight change in the strategy because previously, Tesla had these two domains of experts in the world of AI and ship fabrication. They had their dojo team, which was building this like commercial infrastructure arm. And then they had the AI chip infrastructure team,

that was building the ships that actually went in the vehicles. What they did recently is they merged the two together into a single AI fabrication team. And they are working solely on AI5 and AI6. Now what makes AI5 so special? Well, aside from it being much, faster than AI4, which is currently running on vehicles like Elon just described, AI5 also does this interesting thing where they're starting to actually use it for real-time inference in data centers as well. So an interesting thing that happens with AI5 that I don't think a lot of people are going to recognize. And I want to kind of draw a tie to Apple is that they now own the vertical integration of this hardware chip. So AJA, if you remember previously, Apple used to use

Intel chips for their MacBooks. When you had a MacBook Pro, you used an Intel chip. And it worked really well until Apple rolled out the M-series chips. And when they put those chips in a MacBook, suddenly I no longer needed my 65-pound PC to render a video. All I needed was my tiny little MacBook. And my battery life doubled, the processing power quadrupled. Everything about the device got so much better because it was vertically integrated into the system.

They own the entire stack from the hardware to software and everything in between. This is Tesla's attempt at the M1 chip, but for robotics. So Tesla currently has two arms of real compute, real world. compute, and that's their autonomy in the cars and autonomy in the robots, AI5, and shortly soon after AI6, will be the actual chip that runs in both of those things. It's a universal chip that exists to run local inference on any sort of autonomous hardware device. And it's going to be really freaking fast, really impressive, and it'll be small enough that it could fit into all of these smaller devices. So I think that's why this is a big deal, is this is very much Apple's

M-series chip moment, but for Tesla, which I think is really cool and super, super exciting. Yeah, I was reading up on a few of like the important stats about this new chip and I read that it can basically service up to 250 billion parameter models, which isn't the largest models we've ever quoted on this show, Josh, we've been talking about like trillion parameter models coming from like the likes of, I don't know, Kimmy K2 out of China and stuff like that. But I think the important part here is the efficiency that comes from this chips. I believe this is like the cheapest silicon per power watt chip that would exist theoretically. once they start mass producing it, I think it's slated for 2026 out of TSM, which is close to another big AI hardware giant, Josh, right? And that is Nvidia. And it got me thinking about Nvidia as a whole, which basically has had the linchpin on any kind of AI GPU or chip

that is needed from some of the biggest players, right? OpenAI, Anthropic, Google, they all use and rely on Amazon. Invita, this tiny company which is producing all these kinds of things, not tiny company, but is manufacturing all these kinds of things on a tiny island called Taiwan via TSM. And so there's never really been some kind of major competitor. There have been a few attempts at this out of China because they haven't had access to Nvidia stuff, but they've never really gotten anywhere.

And now we're seeing a real challenger to the throne, which hasn't materialized just yet, but has plans to do that. And I was checking out this video, which. really kind of like summarizes what Elon is trying to pull off here. Maybe we can play it super quickly. What constellation of all these little components fits? And then he figures out, oh, this Optimus teams is this and Robo takes you this and this and this. Oh, we actually don't need this invidio thing.

We can build this into one chip. And then it matches the Samsung constellation. And then we tweak it a little bit more. Boom. And that's what I mean with reality engineering. So what he's referencing here is the current setup for Tesla and all the devices and machines that they build, is it all requires different chips with slightly different architectures, right?

And as you mentioned earlier, Josh, since kind of like August, at least that's been public knowledge, he being Elon, has started merging all of his teams that are building these different kind of chip architectures into one team to focus on creating one megachip. And call me a low IQ person, but the thing that immediately pops into my head is the Iron Man kind of like sparkling glass that's in the center of,

his chess. And I'm thinking of this all-powerful core that basically can power anything from his robotaxies to his Optimus Prime robots, which are being teased and stuff like that. And I think that that is super powerful. Building that in-house is going to be a huge investment and probably a very risky endeavor, but probably there's no one better for the job than Elon. I mean, Josh, do you think similarly or am I kind of like being too much in my fantasy brain right now? No, no, no, that's right. That's right. They're going to do it. They're going to do it. It's going to be remarkable. AI5 is going to be pretty great in terms of raw compute about eight times more powerful than AI4. AI6, I believe, is what has been teased as being the real game changer. That's where we're

really going to see the world start to shift. That's probably going to be the one at Masscale with Optimus. That's going to be the one where you have full robot taxes all around the world. I do think that I might disagree in one point in the sense that I don't know if it's an Nvidia killer or if it's going to take any sort of market share from Nvidia outside of just the sales that Tesla gives to Nvidia. Because if you're a company, all you really want is GPUs. And all you really want is compute and intelligence. So if Tesla's able to create an infinite amount, or if they're able to scale this vertically and create more than they're able to buy from Nvidia, they'll probably just keep them for themselves. So maybe Nvidia loses the purchase order from Tesla. But I don't

see any reason for Tesla to actually sell them to anybody else because it's such a benefit to have that much compute power under one roof. And for it to be maximally efficient. So not only are they going to have a ton more chips, but the chips are going to be a lot more energy efficient, power. All of the things will just be much, much better. So I see your point of view. I think why I would argue against that is Elon seems to exhibit a behavior where if something isn't built to his kind of perfect specification, he gets super annoyed and just builds it in-house, right? If he can't outsource that in any way, we saw that last week when we covered Starlink's, uh, acquisition of a company which basically gives them access to crazy spectrum broadband, which

will now allow all his satellites, which he's eventually going to be releasing in space, to beam down lasers to our cell phones and get access to premium 5G and internet access wherever the hell you are, right? If you haven't seen that episode, definitely go and check that out. So I have a feeling that he might get annoyed enough to just go hull hog and go for NVIDIA's throat. But I agree. If there's no necessary business means, he probably won't. Yeah, if I'm Tesla and I have infinite compute power that I can generate myself by making it, I'm taking all of them. I want all the chips. I'm not giving that to anybody else.

That's my new competitive advantage. And similar to Apple, like now you see it in all of their mobile. Like the new iPhone, a single threaded processor from the new iPhone is equivalent to the M4 Max in a phone, which is just unbelievable, like efficiency and compute power in such a small device. So for Tesla to give that up seems unlikely, but oh my God, what a huge unlock it's going to be when they actually make it. And this isn't even the only Tesla news this week. I mean, you teased the megablock, right? The new energy thing that they have going on. Do you want to just to kind of explain what's going on with this new technology?

Because this, to me, this is even more exciting than the chips. This is pretty cool. Well, Josh, I kind of want you to explain it for me. I mean, look at these images that are like staring at me in the face, right? Okay, okay. So let me give you my left curve take, right? Cool. These look like some 1970s transformer type

gizmos here. It looks like an old school data center. It looks boring as hell, Josh. What am I, what am I staring at? What's a megawatt per hour? And why is it important? Yeah, okay. So we can maybe define a few things that are going to be important for this segment, which is a kilowatt, which is shown kW. And that's the rate. It's kind of like you can think of EJ's like miles per hour, but for power. And then we have kilowatt hours, which are KWH. And that's the amount of energy, which is kind of like gallons of fuel for a car. And then we have megawatts and megawatt hours, which are just a thousand times multiple on those. So kilowatts, megawatts, kilowatt hours, megawatt hours.

And the easy anchors are kind of like a phone battery has 10 watt hours, which is 0.01 kilowatt hours, a battery on an electric car, like a Tesla, has 75 kilowatt hours, and so on and so forth. So the average home uses about 855 of these kilowatt hours per month. That's about 28.5 kilowatt hours per day, about 1.2 kilowatts of continuous draw. So that's kind of how we can frame all of this. Megablock did a really cool thing where they modularized these battery packs. So within a megablock is a megapack.

A megapack has a bunch of energy stored inside and it is able to be deployed at scale to subsidize the grid. So what the megablock is is really just you could think of like a small nuclear reactor almost, except it doesn't actually generate its own energy. It just distributes the energy. And a lot of the problem with the grid that we have today is that we're generating a lot of energy. we're just not using a lot of it. And there's actually a chart that I wanted to show in one of these videos, if you don't mind pulling it up,

which shows the amount of energy that we use relative to the amount that we produce. So you can see, EGES, in the chart, the amount of energy that we're actually using relative to the amount that we have, it's like 30 to 50%. We're using such a low amount of energy, and that's just because we don't have good efficiency in the grid. The way it works, and what the person in this presentation is describing right now,

is you could see how it shifts up a little bit. the grid has to accommodate for the highest point of energy consumption, no matter where you are in the world. So if there's a huge blip in energy demand, they need to meet that energy demand. So there's no blackouts, which means they have to pour a lot of resources into one place. And it's not necessarily the most efficient place. It's just the place that has a lot of demand draw at one specific time. What these megapacks do and what these megablocks do is they subsidize that demand. So I'm getting all my megablocks and megapacks confused. But basically what it is is you can drop one of these things into an area and it could subsidize the grid in a way that will smooth out that bumpy line that we just

saw. So one of the big things and the big problems that we have with current AI data training centers, for example, is getting the amount of energy that is required to these places is nearly impossible. It's so hard. And it's not because the energy doesn't exist. It's because we don't have an efficient way to get it to the place. So these megablocks exist as a way to store that energy and then release it onto the grid in a way that smooths out those curves and allows you to optimize for more energy consumption on a regular basis. That's super cool. I was thinking of other ways that I can kind of like understand the magnitude of power

that's going to be kind of like fuel through one of these. And you mentioned that these megablocks are basically going to be composed of these things called megapacks, right? And then I was like, well, okay, how powerful are these individual megapacks? And I was looking it up. And one megapack alone is enough to power 170 average American homes for about a day, which is insane, right? And then I was like, okay, well, what is that compared to the amount of power that's

in a single Tesla car, which, you know, you own one of these, Josh? And it's 67 of them. So I'm trying to think about, like, how this kind of like stacks upon each other as, like, Elon keeps kind of like rolling all these, like, energy tradeoffs into one super efficient thing. And I kind of like envision a world where you can kind of create a city in the middle of a desert or somewhere where it's very resource scarce and have a fully functioning society of people out there

that can just kind of like survive on their own with these megapacks. It also gets me thinking about eventually when we end up with proficient space travel and we end up landing on different planets and stuff, we could probably be able to power all of this from some of these megapacks, which is super, super cool in my opinion.

Yeah, and it's not to discount the fact that we still need to power these megablocks. So the megablocks are just batteries. They don't actually generate any net, energy, but in the case that you can supply energy to these blocks, well, they will take care of all the rest. And like you said, EJAS, where there's a place where there's not a lot of infrastructure available or a lot of infrastructure gets wiped out, you could just plug these into a electricity generating source, and it will take care of the rest. And it's a really cool,

compact thing. I mean, it's four megapacks. They turn them into one block, and then that one block provides about 20 megawatt hours per block. And it's a huge amount. And the speed at which they're able to actually deploy these, I think, is one of the most interesting things, where in the presentation, they said, they can get one of these deployed in 20 days. And that means with enough of them, you can get up to a gigawatt of energy in 20 days. And if you'll remember from our past episodes, a few of the data centers like Open AIs Giants Data Center or XAI's Colossus Data Center, they're actually targeting for one gigawatt or two gigawatt of energy, which is a huge amount that cannot be deployed very, very quickly with these megablocks. So it's a really powerful way of

not only subsidizing the grid, but unlocking energy that we previously would not have actually access to. I think a lot of the times we say, and I've been guilty of this, like, we don't have enough energy. And the reality is we don't. But we could have a lot more if we just distribute it properly because we're losing half of it, almost half of it, to just inefficiencies in the grid. And this will hopefully smooth that out over time. Being a show where we discuss AI a lot, particularly AI models, and we've mentioned how much energy and compute it requires to pre-train some of these models, right? I'm wondering whether they are more efficient ways that we can train models using these megablocks or these megapacks, whichever one

you want to consider. I saw someone kind of like have a good take on X the other day where he basically said at nighttime typically it's harder to train some of these or pre-train some of these AI models because it's not as efficient at night time for many different reasons. For example, it might be cooler. And if you could supply some of these megapacks or these megablocks and have that continually trained certain models, I think that we, we'll start to see some of these things in Colossus 2 was the point I was going to make. And I'm curious whether there is any kind of overlap with Elon's new data center that he's building out. Where is it now?

Is it in Texas in Austin? I believe it's Tennessee. Tennessee. Okay, cool. Yeah, I'm curious whether he might start implementing a bunch of these different components that he's building with Tesla into some of his AI-driven efforts. Yeah, I'm sure they will be. Because if you remember, I think, I forget what percentage of it was, but almost all the electricity that went into the Colossus one, structure in the first factory was from megapacks.

And a megablock is just a juiced up steroid version of that. And that's mostly because, I mean, in their case, the issue was that grid power isn't always completely smooth. And when you're powering up and down 100,000 GPUs at once, the energy swings are very large, very quick, and they needed a way to smooth that out. So the megapacks solve that. Now, with a megablock, it's just like that on steroids. It is a tremendous amount of energy that you can smooth out into these data

centers that doesn't mess up a single training run. And there's this crazy stat. I mean, I mentioned they can be rolled out in 20 days, but 200 of these sites would enable a 20% increase in total U.S. electricity generation without building a single new thermal generator. It's just by better utilizing what we already have. So if these things start to get deployed at scale, it can seriously make a huge impact in the amount of energy that we have and the cost per kill a lot more efficient, more importantly, that we always talk about, without actually needing to spin up more solar solar farms or gas turbines or whatever type of energy source we need because we have it. It's just a matter of just deploying it more efficiently.

And that's exactly what Megapak does. And that's why it's really, it's so important because if this gets rolled out at scale, we could see the cost per kilowatt drop very, very much. And also, it can be rolled as data centers very, very quickly. So in the case that someone needs a lot of power, well, they don't have to go through hell and back to get some energy there. They could just roll one of these bad boys out and juice it all up. Okay, so EJJES, we kind of talked about why this is so important.

but I understand that they did some pretty amazing things in order to achieve this that you can hopefully walk us through. Well, okay, listen, I'm a massive fan of Pokemon, and the main reason why I was is because these Pokemon cards have some of the coolest stats ever. And it seems like we have it for the Megapak3. So introducing the Pokemon card version of the Megapagepak 3, 5 megawatts per hour of usable AC energy. its weight is a casual 86,000 pounds. This I found super interesting. It operates between minus 40 degrees Celsius to 60 degrees Celsius. So basically it's impervious to any temperature fluctuation.

You can have a pretty much manufactured anyway. That's a humongous range. And 75% of the mass of a megapact 3 is battery cells. Now, having come hot off the Apple episode last week, Josh, where that stat basically rings true for the new iPhone Air. That's going to be delivered to my building in about a couple of days' time. I kind of see a trend developing here, which is pretty cool. A single module weighs as much as a cyber truck.

These things are much larger than I thought. I kind of thought this might be like a kind of like a jetpack type thing that astronauts might wear, but maybe that's in V3. And it will partly use battery cells from Tesla's new 7 gigawatt an hour, LFP battery facility in Nevada coming late online in 2025. That's cool because we've spoken about this new battery facility quite a bunch of times on this show, but we've never really kind of like understood maybe the strategy of where he's feeding a bunch of these cells. This is probably one of the major outlets that will get there as well.

Super cool. And I kind of understand now why Yilan is buying a billion dollars worth of additional shares this morning. That news broke. Pretty hilarious. stock prices doing pretty good as well. I think people are excited. This is one of those cases where if any company in the world were to achieve any of these things, it would be massive news. And there's one company that's at the forefront doing all of these things. I mean, the AI5 chip is an incredible feat of engineering. It's not done yet. They're finishing the design process. But assuming they do

deliver like they always do, maybe it'll probably be late, but assuming they do, it's going to be a remarkable piece of tech. And it's going to run a ton of the very first humanoid robots that we see walking around. And now they're also rolling out this energy plan to subsidize the grid that's been faltering and hasn't been like upkept correctly. And it's just, it's amazing to see the progress of some of these innovations. And not only how quickly they're rolling it out, but how effective they are. I mean, every single one of these new product rollouts is the best in class. And I don't see that changing anytime soon. So I, it's a good time to be bullish.

It's a good time to be bullish on Tesla on the entire ecosystem around them on the XAI team who also benefits from this. I mean, there's a lot of all you. start to see all of the vertical integration, how it all feeds into each other and they all help each other in this, this really nice ecosystem. And it gets me fired up. I mean, this is good stuff, dude. This is a, these things when they work at scale will change the world, particularly the megablock. I mean, this is, this is huge. For a long time, I think the idea of electricity or rather like solar powered electricity, powering everything kind of seemed like a gimmick. And the major reason for that wasn't because solar wasn't powerful enough. It's because the distribution of it sucked.

And now we have a vehicle that can clearly distribute this energy way more efficiently, 50% of energy to be precise. So I'm super excited about this. And hopefully, if you will listen to this, this was educational for you. And you learned something new about Elon Musk, who seems to be releasing paradigm shifting technologies every couple of weeks now, Josh. I'm very bullish, Tesla. It is probably the stock that I own in the largest quantity outside of some of my crypto stuff. Oh, nice.

Okay. Oh, yeah. Dude, how can I not be bullied? We talk about it about 50% of episodes. I'm glad. Well, you never know who's actually on board, right? I'm glad to hear this. No, no, I am, I've got my skin, teeth, and bones in the game.

So I'm super pumped about this and I'm excited to see where it goes. But again, if you think that there is a better topic that we could be discussing, if there's something that Elon has released or is talking about that we've missed and that you think we're not covering because our Tesla bags are consuming. us and there's a competitor that needs to be described. We chat to us. Our DMs are open. Our comments are open. We also like hearing feedback, particularly the feedback that is spicy, that disagrees with

us, that wants to hear different takes. Josh and I, we like to fight, but mainly we agree with a lot of the stuff that we discuss. We kind of discuss topics that we enjoy. And if there's anything else that you would like to hear about, please reach out. Like, subscribe, share with your friends, and we will see you on the next one. Which will probably be made to connect. Big hardware episode coming this week, so don't miss it. It's going to be a big one.

Hell yeah. Thank you guys for watching. We'll see you guys in the next one.