Ejaaz: EXPOSING The AI Bubble, Sam Altman, Nvidia vs Google and More | TG Podcast
Ejaaz has flipped from expecting an imminent AI crash to believing the bubble can grow far larger. Three months earlier, he thought “the pin is halfway through the bubble”; now he argues America will keep financing compute, energy, and robotics because AI leadership has become a strategic contest with China, not merely a question of near-term return on investment.
China’s capital efficiency is the strongest challenge to the American scaling model. Ejaaz contrasted Moonshot AI’s Kimi K2 Thinking, reportedly trained for $4.6 million, with GPT-5’s rumored $2.4 billion cost, while claiming roughly half of the best AI researchers are in China. His explanation: constrained access to NVIDIA hardware forced Chinese labs to devise more efficient training methods instead of simply “ramming compute” through models.
Michael Burry’s NVIDIA short looks premature if aging GPUs remain fully booked. Ejaaz said CoreWeave, Nebius, and IREN resemble “AWS but on crack,” with capacity oversubscribed rather than stranded; Nebius reportedly re-signed 10,000–30,000 five-year-old GPUs two quarters in advance. His bubble alarm is therefore simple: watch for falling GPU purchases or genuine excess compute and energy.
The cleanest AI exposure remains compute, with Google emerging as Ejaaz’s preferred NVIDIA counterweight. He calls NVIDIA “the Bitcoin of compute,” treats NVIDIA and Oracle as safer infrastructure bets, and sees specialist clouds as the higher-beta layer. Thread Guy cited Apple’s reported $1 billion annual arrangement to have Google create its model; Ejaaz called that “brutal,” reinforcing his view that Google is a serious NVIDIA competitor.
Crypto-AI becomes investable when tokens convey economic ownership, not merely incentives for supplying compute. Ejaaz highlighted Pluralis, Akash, Prime Intellect, and Nous Research, but argued the breakthrough would be owning something akin to “1% of the latest ChatGPT model” and receiving subscription revenue. Coinbase’s new ICO platform and Uniswap’s movement toward a fee switch suggest to him that tokens may finally become “pseudo equities,” subject to regulatory clarity he expected around early 2026.
Agents remain the weakest part of the thesis, while coding models are already changing employment. Today’s agent is essentially ChatGPT plus a browser, Slack, a small spending allowance, and repeated approvals—“crypto wallets 1.0”—so Sam Altman’s prediction that 2025 would be the year of agents “fell flat on his face.” Thread Guy tied the collapse in entry-level technology hiring to GPT-4.5; Ejaaz agreed hiring was at its lowest and agreed with Thread Guy’s claim that models produce roughly 50% of OpenAI’s code.
The 2024 crypto-AI cycle ran ahead of both the technology and product demand. Ejaaz said AICC “got in over our skis” by overestimating agent progress and rushing the launch without necessary safeguards, including a more distributed community buy-in plan; the team remains active but is moving toward a product-first approach. His revised test is deliberately boring: clearer token rules and applications that customers genuinely want to pay for—he used roughly $20 a month as a generic example.
1. The AI bubble became a geopolitical financing mandate
Ejaaz’s change of mind is the episode’s anchor. Three months earlier, he saw “the biggest bubble ever” with the pin already entering it; now he thinks Michael Burry will be “massively wrong” because the numbers can grow beyond what investors currently consider plausible.
OpenAI has committed, by Ejaaz’s account, to spend $1.4 trillion over five years, with deals already signed. The decisive question is not whether each model immediately earns an adequate return, but whether the United States can afford to lose AI leadership: “It is USA versus the rest of the world—and by the rest of the world, I mean China.”
That framing turns spending on data centers, power grids, and robots into strategic policy. Ejaaz’s categorical conclusion: if America loses, the currency is threatened, so funding will continue even when conventional company-level economics look uncomfortable.
Thread Guy challenged the premise that China is truly far ahead. Ejaaz answered with DeepSeek: a hedge fund created an open-source model that could be downloaded and run privately, breaking the assumption that only laboratories with billions in capital could compete.
2. China’s constraint-driven efficiency challenges brute-force scaling
The sharper example was Moonshot AI’s Kimi K2 Thinking, which Ejaaz said “broke all benchmarks” after costing $4.6 million to train, versus a rumored $2.4 billion for GPT-5. He presented those figures as evidence of a profound cost discrepancy, not proof that compute no longer matters.
His talent argument was equally direct: roughly 50% of the world’s best AI researchers are in China, he claimed, with about 25% graduating in America and moving there and another 25% emerging from Tsinghua University, “the Harvard in China.”
Ejaaz carefully disclaimed deep technical authority, but offered the mechanism: American frontier laboratories largely improve models by forcing enormous compute through them, “like water going through a river through a dam.” Chinese teams, lacking comparable NVIDIA access and funding, had to make models learn more efficiently.
The implication cuts both ways. Chinese efficiency threatens the economics of American model training, yet the strategic response could generate even more American infrastructure spending—the reason Ejaaz expects the bubble to expand before it breaks.
3. Full utilization undermines the stranded-GPU bear case
Ejaaz summarized Burry’s thesis as over-ordering: companies bought hundreds of billions of dollars of GPUs that will depreciate faster than assumed, become technologically obsolete, and sit as dead weight. The recent earnings evidence, in his view, says the opposite.
CoreWeave, Nebius, and IREN provide specialized AI clouds—“AWS but on crack”—building GPU clusters so model companies can focus on products. Their stock charts resemble “an altcoin season,” but Ejaaz said capacity is not merely full; it is oversubscribed.
His best specimen was Nebius: it reportedly re-signed 10,000–30,000 GPUs that were already five years old, two quarters ahead of time. “Someone’s already paid them,” he stressed, which directly challenges the claim that older chips have become commercially useless.
His market dashboard is uncomplicated: follow NVIDIA, the specialist compute providers, and public AI platforms such as Google and Meta, then ask whether customers are still purchasing GPUs. When buyers retreat—or the industry admits it has excess chips and energy—“bubble’s popping soon.”
4. NVIDIA owns the compute trade, but Google owns a full competing stack
Ejaaz divides AI exposure into layers. NVIDIA is “the Bitcoin of compute,” while NVIDIA and Oracle form the safer basket; CoreWeave, Nebius, and IREN occupy the mid-cap, higher-beta tier. Consumer applications are the “shitcoin sector”: abundant, replicable, and vulnerable to being copied by Google or OpenAI within weeks.
Sora illustrated the demand case. Ejaaz said OpenAI’s AI-generated TikTok competitor consumed roughly 80% of its compute for three months, leaving the company short of GPUs; commitments extending years into the future therefore reserve capacity that one successful application can rapidly absorb.
Ejaaz personally cited positions in Google and Tesla. His Google thesis is that it is “the only competitor to NVIDIA”: it built its leading models on internal chips rather than NVIDIA GPUs, transformed Gemini after a disastrous launch, and now operates across chips, models, distribution, and consumer products.
Apple is the contrast case. Thread Guy said Apple spent “fuck all” on AI research and talent, delayed Siri by two years, and then announced it would pay Google about $1 billion annually for a model. Ejaaz called that “brutal.” Thread Guy also surfaced a viral SoftBank-NVIDIA sale as current news; Ejaaz corrected him that the exit occurred in 2019, shortly before ChatGPT—“the biggest L he’s ever taken.”
5. Sam Altman’s execution record competes with a persistent trust deficit
Thread Guy framed the unsettling clip precisely: OpenAI had promised $1.4 trillion of spending against roughly $105 billion of revenue, and Altman responded that if the interviewer wanted to sell his shares, a buyer was waiting. Ejaaz’s instinctive reaction was, “Something’s off with this guy.”
Ejaaz revisited Altman’s 2023 firing, saying newly surfaced court materials portrayed co-founders describing him as a liar and sociopath before the board reversed course. His conclusion remained a question rather than a verdict: “Can we trust this guy?”
The counterweight is execution. Altman produced leading models, but also the sticky ChatGPT interface, memory, and Sora as a consumer application. Ejaaz’s charitable reading is that Altman does not know exactly where all the revenue will come from; he is irritated by the question and confident he can create it.
He compared that leap of faith with Elon Musk’s milestone-based $1 trillion compensation package and a path toward a $12 trillion Tesla: nobody knows precisely how either target will be reached. Thread Guy’s pushback was telling—the share-sale retort was actually “the least alarming Sam moment,” while the broader pattern created the discomfort.
6. Crypto-AI needs ownership economics before agents can carry the story
Ejaaz’s strongest crypto-AI opportunity is decentralized compute. Pluralis reportedly found a new way to train models across a decentralized network, surprising even a Google chief scientist; Akash, Prime Intellect, and Nous Research are pursuing more enterprise-oriented infrastructure, although many of the relevant tokens were not yet live.
His bull case depends partly on timing: private AI equities remain inaccessible to many investors, so credible teams could launch tokens just as demand for liquid AI exposure returns. Yet supplying compute for rewards is less important than monetization and ownership—his ideal is holding “1% of the latest ChatGPT model” and receiving subscription revenue.
Thread Guy connected that vision to Uniswap’s fee switch. Ejaaz agreed: crypto buyers always hoped tokens would become “pseudo equities,” and Coinbase launching an ICO platform suggested to him that regulatory confidence was improving. He guessed the CLARITY Act could arrive within months, transforming analysis toward revenue projections, sector comparisons, and cash generation.
Agents are not ready to justify that valuation framework. Today they resemble ChatGPT with browser and Slack access, perhaps a $100 wallet, and approvals for every consequential action. Altman’s “2025 is the year for agents” call failed, though Ejaaz warned that scientific-discovery-capable agents could quickly become capable traders—and jaded investors could miss the turn.
7. Real productivity arrived before autonomy, and AICC learned the difference
Thread Guy’s broader point was that society moves its goalposts: driverless cars once sounded impossible, then became unimpressive because Waymos had limited coverage or occasional accidents. Vibe coding follows the same pattern—a person can now prototype and deploy a game in 30 minutes with GPT and a Replit Agent, yet users focus on what remains broken.
Ejaaz argued coding models are most powerful in skilled hands. Thread Guy tied the collapse in entry-level technology hiring to GPT-4.5, while Ejaaz agreed hiring was at its lowest. Thread Guy also claimed models generate roughly 50% of OpenAI’s code, and Ejaaz agreed that it was not human-written.
Ejaaz’s bolder forecast was a household robot by the end of 2026: Figure around $10,000, Tesla Optimus targeting late 2026 or early 2027, and 1X using subscriptions and teleoperation. Thread Guy preferred a human cleaner and raised privacy concerns; Ejaaz conceded early robots would be expensive and strange, but expects costs to plummet and resistance to shift toward new complaints.
AICC’s postmortem mirrors that adoption gap. Ejaaz said the founders overestimated agent progress and rushed the launch without necessary safeguards, including a more distributed community buy-in plan: “We got in over our skis.” The project’s three-month-old agent remains active, but his new standard is patient and commercial—clear regulation, a finished product, and customers willing to pay for it, with roughly $20 monthly offered only as a generic example.
Full transcript
Boom. Ejaaz. What's up, man? We're live on stream.
What's up, man? It's been a while.
I know. It's been a while.
You've got quite the—
Quite the operation going here, mate. Since I was last on, I threw you in the Telegram chat. We got the team in there. We're setting the calendar.
I haven't said a word to Thread Guy until right now. Crazy. He's surrounded by his squad now.
It's awesome. I'd love to see it. Welcome back, man. You came on in November or December 2024. I think we were talking about GOAT at around $1.5 billion. Still to this day, my favorite coin of all time.
I think the timing to get you back is pretty optimal because, obviously, macro, everything is AI. It's all anyone wants to talk about. But on a more micro level, I think a lot of the sharper people I talk to in crypto are starting to say, "Yeah, AI, yeah, Virtuals." Some of the sharper people are pointing me in that direction. They're interested in some on-chain AI stuff, so I think the timing is pretty good.
Do you want to start with a quick introduction for some of the new people—who you are, what you focus on—and then we can talk about Michael Burry shorting the top of the AI bubble?
Yeah. What's up, guys? My name's Ejaaz. I've been around the crypto space for well over a decade now. I spent a bunch of time building products at Consensys and Coinbase, then left and started my own fund. I have a liquid fund focused specifically on investing in crypto and tokens.
As a side hustle, I'm also a co-host on the Limitless podcast, which as of today is a top-30 podcast on Spotify and Apple Music, or whatever that platform's called. We're having loads of fun. It's a different world.
Damn, congrats. I hadn't really realized it. I wasn't paying a ton of attention, and then I had Mando come on—I don't know, three weeks ago, a month ago—when he was shilling it really hard. Then I checked it out, and you're putting out some good stuff, which is awesome. Congratulations. That's a huge number.
Thanks, dude. It sort of materialized out of nowhere, as you know. I was doing a bunch of the crypto AI stuff specifically at Bankless, and we started tapping into an audience that really cared about AI and wanted to hear more about that stuff. So we just took it that way. Six months later, here we are.
That's awesome. Congratulations. Do you want to set the stage a little bit? I don't feel like we've done a good enough job on the stream covering a lot of the high-level macro AI flows. Obviously, we're hyper-focused on crypto, so you only get some of the picture if you're focused on crypto. If you're focused on AI crypto, you get even less.
How do you feel about where we are in the grand scheme of AI, specifically in the markets? Are you scared and uncomfortable? Do you feel toppy? Where are we at?
I think if you asked me this question three months ago, I would have said, "Mate, we are in the biggest bubble ever. The pin is halfway through the bubble. It's about to burst immediately."
Talk to me now, and I'm going to tell you that I think Michael Burry, the guy from The Big Short who is now shorting NVIDIA, is wrong. He's going to be massively wrong, and this bubble is going to get so much bigger that people can't really get their heads around the numbers. That's why they're panicking.
Let me try and break down the macro side of AI with a few different themes for the audience. Number one, these companies in AI are spending figures that are bigger than anything we've ever mentioned in crypto, and they're earning more than anything you've mentioned in crypto.
I'll give you an example. OpenAI is spending $1.4 trillion over the next 5 years. That's what they've committed to. The deals are already signed, done, and dusted. Now it's on them to basically spend that amount of money.
The other major thing is to forget about whether a model is cool, how much ROI it's giving you, or whether they're going to be able to pay their bills. I'll distill it to one thing: it is the USA versus the rest of the world. By the rest of the world, I mean China.
If the USA loses, we're fucked. The currency is done anyway. That's why they're going to keep printing that money. That's why they're going to keep investing so much money into rebuilding the energy grid, building robots, and all that stuff, because China is unexpectedly way, way further ahead than us.
People don't quite believe that because we've been ahead for a while in tech. But that's been flipped with AI. That's how I'd summarize it.
Is it true that China is way, way ahead of us?
Let me give you a few examples. We—the West—have released a bunch of AI models that you and I have heard of and probably use regularly: GPT-5, Claude from Anthropic, Google Gemini, and they're all sick. Do you know how much it costs to train all of those collectively?
I honestly have no concept of it.
It's well over $500 billion, up toward $1.5 trillion, depending on the amount that's been spent collectively since 2022, when we really had the breakthrough with GPT.
Remember, it doesn't just take one chunk of money to train an AI model. You need to go through several cycles to train it.
China, starting at the end of last year, decided to do a sneaky thing, which the crypto people probably love: They released a competing model, but they open-sourced the damn thing. It was free to download, free to access, and you could run it privately, so there was no fear of China spying on you. And they're so damn good.
That broke people's concept because they were like, "Hang on a second. Sam Altman has been telling me that I need billions of dollars to spend to train an AI model, so I can't compete with them unless I have billions of dollars. But here I have a Chinese quant." They were a hedge fund that created DeepSeek. That's the model I'm referring to.
That broke the stock market, and it dumped. You remember?
Everything was panicking. Yeah, I remember this.
Yeah, everything was panicking. I thought it was a one-off. I was like, "Ah, whatever. They were a smart team."
This week, a team called Moonshot AI released Kimi K2 Thinking and broke all the benchmarks again. It cost $4.6 million to train, versus the $2.4 billion that OpenAI was rumored to have spent training GPT-5. There's a massive discrepancy.
Then a critic might come back to me and say, "Well, dude, it comes down to talent. It comes down to how good the talent is, and China doesn't have that much talent." Also wrong.
Fifty percent of the world's best AI researchers are in China. Are they graduating from the US and going to China? Twenty-five percent of them are. But 25% of them are graduating from Tsinghua University, which is like Harvard in China, and they're so sick.
They're doing this without NVIDIA GPUs, without the support of all the big American firms. They're doing this with less money and less compute. It's insane. Most of the world is oblivious to this.
How is it possible that they're doing it for so much cheaper?
There are a few different ways, and I don't want to get super into the technical details. One, I don't want to bore the audience, and two, I don't think I'm qualified to talk about it in the first place.
But TL;DR, they discovered a bunch of different ways to train an AI model versus the standard way that most American frontier AI labs do it. Most American frontier AI labs' solution is, "I'm going to spend billions of dollars on compute." Basically, they get a bunch of chips and run compute through them. Imagine water going through a river or through a dam, and out pops a really smart model.
The Chinese don't have that benefit, so they've gotten really creative with the way they've designed these models. They learn in a much more effective way with less money, and the Americans haven't quite cracked how they've done that.
So if this plays out the way you foresee it playing out, with the bubble getting so stupid that no one can even comprehend how inflated it can get, what's the trade? Where do you want to be focused, positioned, or paying the most attention to?
I put out a tweet today commenting on the Michael Burry thing. Let me take two steps back. Michael Burry, the guy from The Big Short, was made famous by the trade he made in 2008, right before the financial crisis hit. He was like, "Fuck it. I'm going to short a bunch of these subprime mortgages," and he made close to $800 million for his fund.
He's come back and said, "We're in an AI bubble." We're in an AI bubble because we have all these companies spending hundreds of billions of dollars to buy GPUs. GPUs are the hardware you need to train these AI models. But they've bought too much.
These GPUs, these chips, are becoming dead weight. They're not actually using them. They're degrading and depreciating a lot faster, so they become useless much quicker and they're losing their market value.
Right now, if you pull up that tweet—I don't know if you can share it or if you want me to share it—I shared a bunch of screenshots from AI cloud providers. These are different companies, Thread Guy, that are basically providing the compute these labs need. Look at their stock prices.
It's gross.
If I covered their tickers, dude, you would think it was altcoin season. That's basically what's been happening in AI.
What are these companies?
Dude? CoreWeave.
Yeah, okay.
So, CoreWeave. You've got IREN, you've got Nebius. Think of these as AWS but on crack, specifically for AI.
They'll come to you as OpenAI. They'll come to you as Google, and they'll say, “Yo, I know you need to set up a bunch of data centers to train your model. We got it. We got all the GPUs. We'll set up the data centers for you. We'll stack them all together. You just focus on building a sick product.”
And they're like, “All right, cool.” Microsoft is like, “Cool, I'll give you $10 billion.” OpenAI is like, “Cool, I'll sign a $100 billion deal with you,” and so on and so forth. These companies are up only.
A criticism that I would receive from saying that would be, “Well, dude, how do you know that the GPUs aren't useless?” They just had their quarterly earnings last week—all these companies that are up like 6× over the last 6 months—and there's 1 consistent thread. Not only are their GPUs at full capacity, they're oversubscribed.
One example I'll give is Nebius. Okay, dude, get this. This is fucking crazy, right? I know it's not fucking—I’m nerding out a bit, but bear with me.
Please, please, please.
Nebius is one of these AI cloud providers, right? They have a bunch of GPUs, these chips that are 5 years old.
Years old. Okay.
Dude, they just re-signed all of them—10,000 to, I think, 30,000 of them—2 quarters ahead of time. Someone's already paid them. So, in 6 months, it's already booked out.
It's already booked out, and it's booked out for the next 6 months.
The point I'm making is that there is obvious demand for these GPUs. These GPUs aren't just being bought and then left as dead weight, right? There's obviously demand and use on the consumer end. People just don't realize it as much.
Yeah, we actually don't. I'll argue the opposite: we don't have enough compute. We don't have enough energy to feed that compute. It's insane.
So, are you slamming these equities?
Yes.
What are you paying the most attention to? The ones that you posted? Where are you, and what sector are you mostly focused on? Where do I need to be paying attention to?
Oh, dude. I'm everywhere and anywhere.
To give some context, I think people will listen to this and be like, “Dude, so you're just an AI guy now?” I spend a lot of time researching this stuff with a specific focus, which is that I know crypto and AI will converge, and I want to be an expert in both sides of the coin so that I know when that happens, right?
I tackle the AI side in a similar way. You're asking me what kinds of sectors I'm looking at. There's a few.
Number 1 is compute. Compute is where pretty much all the money has been made. NVIDIA is like the Bitcoin of compute. If you want to bet on AI, you're basically betting that you're going to need a ton of compute to feed all these models, and on the demand for it.
One example would be OpenAI releasing a TikTok competitor. It's called Sora. I don't know if you saw that.
Basically, it's like TikTok, but all the videos that you watch are AI-generated, dude. That broke their service and took up 80% of their compute.
Seriously?
For like 3 months straight. They don't have enough GPUs, dude. That's why they're signing $1.4 trillion compute deals, because not only are they like, “Yo, we don't have enough money to get the compute right now,” but they want you to sign on the dotted line, Jensen Huang, that in 2 years' time you're going to give us this compute, because they know they're going to need it.
One app took up that amount. So, the point is, if you want to bet on AI, you want to bet on compute. Then the question is, “Well, fuck, which companies do I want to bet on?”
There's the safe basket, which is like, “Okay, I'll buy NVIDIA. I'll buy Oracle.” Then there's the mid-cap, low-cap bucket, which is these AI supercloud providers that I've mentioned in this tweet: IREN, Nebius, and all that kind of stuff.
If you want to make bets on the consumer side, that's kind of your shitcoin sector within AI.
Okay.
Yeah, yeah, okay. There's a ton of them. They're easily replicable, and they could just get eaten up by Google and OpenAI, which could literally spin up a competitor in a few weeks.
So, that's roughly how I break it down.
Thank you for that. When you see the Sam Altman interview clip and they're like, “Sam, you've promised $1.4 trillion, but you only do $105 billion in revenue. Where is this money going to come from?” And then Sam looks him dead in the eyes and says, “Brad, if you want to sell your OpenAI shares, I promise you we have a buyer on the other side.”
All right. I listened—I think All-In did a pretty good segment on this—and I think there's an argument to be made that it was a poorly worded, autistic kind of joke that just didn't land. But still, I think a lot of people saw that and they're like, “What the fuck? This is scary.” What do you think? How does that make you feel? And then, in the broader scene of that, what do you think?
Sure, sure. All right. So, there's a few ways that I react to that specific event, and to Sam Altman in general.
My gut is kind of like, something's off with this guy. I don't quite know what, but something's a little snaky behind it.
A couple animals in his basement.
Yeah, there's a couple animals in his basement. He's lied about a ton of stuff.
I don't know if you saw this, but do you remember in 2023—November 2023—when he got fired?
Yeah.
Yeah. And then they brought him back. What the fuck was that?
Yeah, bro. Do you remember that?
Okay, right. So, basically, for those of you who don't know, Sam, the founder and CEO of OpenAI, was fired in 2023. One of his co-founders fired him and got the support from the board to fire Sam.
Over 3 days, he then got rehired. No one knew what the hell happened there until 2 weeks ago, when it came out in a court case because Elon is suing him for something else.
Dude, the files are insane. Basically, all his co-founders were messaging, being like, “Sam is the biggest liar and sociopath I've ever met. We need to get rid of this guy.”
Whoa.
And basically, they fired him, walked it back, and now Sam's still at the helm.
So then the question becomes, can we trust this guy?
Okay, now let me look at it objectively, right? The dude's produced the best models, and not only has he produced the best models, he has the most addictive product loop, right?
He didn't just create ChatGPT. He had the interface, right? He had memory, which made it more sticky. So, when you log on, it's like, “Oh, I know you, Thread Guy. Yeah, I know what you want to talk about.”
He created Sora, right? So, it wasn't just a text-to-video model. He created Sora to create an app, so it was a consumer app. He's obviously good at what he does.
When I looked at that clip where he was like, “Listen, if you want to sell your OpenAI shares, fucking sell them,” I just think he's annoyed and gets irritated by these kinds of questions, because the simple answer is he doesn't know exactly how he's going to make the revenue. He's just confident that he's going to make the revenue.
If I were to support him, look at what Elon Musk has just done, right? He's just signed a $1 trillion pay package. It's based on some specific milestones, right? He needs to make Tesla basically a $12 trillion company.
Do we know how he's going to get there? This is his thesis, right? Who the fuck knows? I don't fucking know. So, you have to trust that Sam's going to execute. And so far, he's done that. So, we'll see.
You know, I think the least scary part—the least alarming Sam moment—was that clip. The least alarming.
Yeah.
You know what? That was whatever. Everything else is, like—for as much of a lunatic as I think Elon has become, and things have gotten a little bit weird in the Elon Musk corner—
What's gone weird?
Yeah, well, the Trump-DOGE stuff was kind of a fuck-up. Like, what's going on? He doesn't look good. It was a little— But as uncomfortable and as weird of a little moment as that was, I feel like with Elon, you take him at face value. He's pretty down to just let it fucking rip, let it fly.
You don't really feel like he's slighting you or hiding anything from you.
Yeah.
But the Sam Altman stuff makes you feel a little uncomfortable, to be completely honest. On the back end of the Michael Burry stuff, I have this tweet here: SoftBank sold 100% of its NVIDIA shares; it used to hold 5%, this says. I don't know how true that is, but that's a big number. If so, does that—
Dude, they did that back in 2019. Right before ChatGPT. It's the biggest L he's ever taken. He said this on multiple—
Wow. But this tweet is going viral on November 11, 2025.
Yeah. It's because he sold it a while ago. He once owned so much NVIDIA stock, and he sold it all—
5% is disgusting.
Right before ChatGPT, dude. The same.
Wow.
That could have made his entire fund.
Oh, that's gross. Okay. So, on the back of this AI stuff, I like your framing of getting well-versed in crypto and getting well-versed in AI—the two things intersect, and then boom. Where are we right now with crypto AI?
Yeah, it's a good question, and I'm just going to be honest. Let's start with the good stuff. All that stuff I mentioned about compute—if you take away anything from what I've just said, it's that there is not enough compute. There is not enough hardware. You need the hardware to build the amazing things so that we go off and beat China, blah, blah, blah, so that the U.S. dollar maintains its composure.
We don't have enough compute, and there are many ways to skin this cat. Within the centralized AI world, one way is to build these GPUs ourselves, buy them from NVIDIA, and pay billions of dollars to cloud compute service providers. That's where we start spilling over into crypto.
Because there are a bunch of awesome teams producing distributed compute networks engineered for enterprises, training models, and inferencing models—which is when you prompt something and it needs to respond to the user's request.
There are a bunch of cool teams doing it. I don't think a lot of tokens have gone live for them just yet, so I'm following the teams, the research they put out, and the releases they put out. There are a bunch of pretty cool ones.
If you want me to talk about some that I'm super enthused about, there's Pluralis. They had a breakthrough last month, which is crazy. They basically found a really new way to train models using a decentralized network. The reason why this is such a big breakthrough is that even the chief scientist from Google was like, "Wait, what the hell? You guys figured out a way to do this?" He tipped his hat to them, saying, "Wow, I didn't think this was possible, but you guys have figured it out."
It's small beginnings right now, but they're still spinning the wheel there. Then you have the other side of things, where Akash or Prime Intellect are servicing the more enterprise-heavy side of things, and they're doing an awesome job. Nous Research as well—shout-out.
The thing with all of these projects is that they don't quite have a token just yet. But if you want my true bull case, I think these guys are going to release a token right when it matters—right when people are looking to play the AI side and don't have access to public equities, because all these companies in the AI world are private. So they'll play it on the crypto side. That's what I'm betting.
I was about to ask you about Nous.
They're going to look for quality.
I was about to ask you about Nous, which is one I was really excited about. I haven't seen that much from them recently, but I know that was one the sharps were hyped about.
Yeah. The Nous guys are heads-down. They raised, what was it, $50 million or $150 million from a16z and a bunch of other people. I know Karan and the team are heads-down building and figuring it out.
The truth is, it's just super hard to pull off building a massive model at scale. Here's the thing: I think what's going to matter the most is not necessarily how you allow people to earn tokens and contribute compute to train a model. It's going to be how you monetize it and how you can share ownership over it.
Imagine if you could own 1% of the latest ChatGPT model. I could earn subscription revenue off of that. I think we're going to see more of that coming soon—within the next couple of years or so.
Now, the other side of it—the elephant in the room—is the agent stuff.
Yeah.
I'm just going to be honest with you: it's where I'm most bearish right now. Not necessarily because I don't think crypto has a place in this just yet. It's just that I see all the agents on the traditional AI side of things, and they're not quite there yet.
They kind of stink, right?
They stink. What are they going to do? Book me dinner and book flights? I don't fucking care.
I don't fucking care.
Okay, I actually have 2 takes. First, before I lose it, to your point about basically aligning token holders with revenue—owning parts of the company, owning a piece of the model—I had a guy on before, and we were talking about Uniswap and the precedent that's being set. Uniswap is allowed to turn the fee switch on, and it's sort of a signal to builders in America that we're there, or really close to where it's the standard or the expectation that if you have a protocol, a company, and a liquid token, you can align incentives with token holders.
That is really cool because it's bigger than just DeFi. I think it stretches into—
That's a big deal. I don't think this means, "Cool, I want to bid the UNI token," but I think the precedent being set is a big deal. You could imagine getting to a scenario in the relatively close future where there are more investable tokens than there currently are.
That's exciting, dude.
Dude, yeah, I 100% agree.
If you want to dumb down that thesis for everyone, it's basically that we bought crypto tokens because we hoped they would eventually be considered pseudo-equities. Now we're on the cusp of that being the case.
Coinbase literally launched an ICO platform this week. Guys, I work there. They're not doing that unless they've confirmed from a regulatory perspective that they're going to be okay. They're not taking risks.
Uniswap—yeah, bro. Uniswap, Hayden, is very particular when it comes to these kinds of things. He hasn't even considered doing this. The fact that he's turned the switch on, or is turning the switch on, is a huge deal. He's spoken to the governors; he's spoken to the lawmakers.
So, basically, I'm guessing that the CLARITY Act—which is this act that's going to make all of this okay—is going to come out soon, in the next couple of months, maybe the next quarter. Then suddenly, the way that people invest in crypto tokens is going to look very different. It's going to look much more professional. You're going to wear a suit, look at revenue projections, and say, "Oh, okay, how much money is this thing making?" And what sector is projected to make the most money?
Of course, AI, right?
AI.
Yeah, it's a pretty big deal. You get a lot more tokens that look like pump-and-hype, which is good for people who are serious and take this thing seriously. So, thank you for that.
Agents suck, dude. What the fuck happened? Okay.
I remember watching this demo three, five, or six months ago—I forget. What did they call it? OpenAI did a demo for the first agent. I forget what it's called. And I bought the meme coin of it.
I'm watching this thing, and I'm like, "Dude, I don't know." It's booking a Lakers ticket, and I'm like, "All right, I guess." I mean—
What? Yeah, what's the deal?
Okay, so here's the deal. Agents today, like that OpenAI agent you just referenced, basically look like ChatGPT with access to a web browser, access to Slack, and maybe access to a bank account that allows them to spend $100. That's literally it.
Yeah, that's literally it.
Oh, and through everything they can do with those tools, you need to approve it. It's funny. It's kind of like crypto 1.0, crypto wallets 1.0, where you have to sign the fucking transaction. You have to approve the thing. That's basically where AI agents are right now.
Now, the funny part is that Sam Altman basically said that 2025 is the year for agents. He fell flat on his face with that prediction. That is not the case. But he has a revised timeline for when agents are going to get really good.
He basically said that at the end of next year, you're going to have an AI agent that will basically be able to discover new scientific discoveries, right? An example might be that it discovers a drug that will cure cancer, whatever that might be, which sounds like, whatever—this doesn't affect me in the crypto world. That's a big deal because if you have an AI agent that can do that,
You will have an AI agent that can do much more for you in crypto, right? I heard Sean talk about it prior to me coming on: “Yeah, they'll go out, they'll spend your money, they'll trade for you.” I think that's probably more likely when you have an agent that can make scientific friggin' discoveries. It's just a matter of waiting. It's not smart enough. One thing I will say is, when we do eventually get there, it's going to happen super quickly, and people who are jaded on agents are going to miss the entire boat when it does happen.
Yeah. It'll be like a vacuum.
I'm down for this. I would like to have an agent that can go, “Hey, Thread Guy, here's 10 songs based on what you've been listening to.” You know what I mean? I'm down for stupid stuff too.
In its current use case, it feels like the other thing that was a little bit forced was the vibe-coding thing, which I was really excited about.
Which, honestly, is impressive. Here's a not-thought-out tech thesis at all, but I was having this conversation with someone the other day. Five years ago, if you told me, “You live in a city somewhere, and there will be Waymos driving on the street with people in the backseat and no driver,” I would have said, “Dude, shut up. What are you talking about?”
Now the Waymos are here, and we're like, “Well, they're only in LA,” or, “They're not that good yet.” There's always a couple of accidents. It's always, “Well, robots will never happen.” Then we start to get really advanced in robotics, and it's, “Well, they clean the dishes really slowly.” I think, as a society, we do this: We say something will never happen, and then it comes and we're instantly desensitized to it. It's like, “Well, it could be better.” The Waymo thing is fucking nuts to me.
I remember my dad trying to get a website built in 2009 for his business, and it was brutal. It's pretty amazing that you could have a concept for a game and build a pilot in 30 minutes with GPT, basically, with a Replit Agent. It's done. It's built, it's live, it's deployed, and people can try it. It's pretty sensational.
It isn't lost on me how quickly we get desensitized to these things. With that being said, I think we really push the agenda on how fast they can be normalized and go mainstream relative to where we are as a society.
Yeah. So here's the thing. Your point around vibe coding is right: It's widely accessible to anyone and everyone. You get a Claude subscription from Anthropic, and suddenly you can basically say, “Build me this app,” and it builds you this app. Get a Cursor subscription and it would do the same thing. It's good.
The thing is, the coding models are super good for those who actually know how to code—
Who are good at coding.
Do you want to know the number-one sign of that?
Entry-level jobs for all the basic tech companies are nonexistent.
It's the lowest it's ever been.
You want to know when that trend started crashing? When that graph started crashing?
As soon as GPT-4.5 came out—that first coding model, right?
Another crazy thing is 30% of the code—sorry, 50% of the code—that is made at OpenAI to literally make their models and apps is all done by their model. It's not done by them.
It's not done by a human.
That's crazy.
Which is like that. This is happening behind the scenes, right? Dude, your point on the robot thing—my biggest thing is that I talk to my mom about this a lot. She's watching these episodes. She's like, “So, what is AI doing?” and blah blah. I'm like, “All right, I've got to explain all this.”
My hot take is that there'll be a robot in her house by the end of 2026.
2026?
Like, cleaning the dishes, like you said. The Figure robot and Unitree robots are already available. The Figure robot, which is like the American home-brand robot, is coming out next year. Tesla Optimus is aiming for the end of next year or the start of 2027. Elon has a massive factory here in the US that is going to start building these things. It's not for the next couple of years, but some of the other American companies are super ahead.
Wouldn't I rather just hire a cleaner to come to the house and clean?
Sure.
Cheaper.
Sure. Way less of a headache. No tech problems. I don't have to worry about someone spying on this thing in VR.
Yeah.
You know, like, I don't—
Yeah, yeah.
How realistic is that?
Yeah, yeah. So, you're right. A lot of this is, “Okay, let's see how this plays out.” A lot of this is going to be expensive to start off with. There are a few different models, right? With Figure, you could have this robot for $10,000, which some of the richer families will be like, “Ah, fuck it. Whatever. I'll do it. I have a permanent cleaner in my home. Great.”
Then there's the other model where you pay a subscription. That's the 1X robot brand that went viral a few weeks ago. Dude, it's teleoperated. If it goes to shit, you have a human somewhere in another country wearing VR goggles who's like, “All right, Thread Guy, let me just move around your closet and all that kind of stuff.” It's super weird.
Don't get me wrong, the first couple of robots are going to be like, “What the hell is this thing?” But just like with everything else, that cost is going to plummet. Then we're going to shift the goalposts. We're going to be like, “Ah, but it cleans too slowly. Ah, it doesn't know how to drive my car.”
Can I remind you that when ChatGPT went viral in 2022, everyone was like, “There is no way this thing can write my PhD thesis. There's no fucking way.” Now it's standard. You pay $100 and you can get that. That's crazy, right? We're going to keep moving the goalposts. It's going to get crazier.
If you go to Austin right now, by the way, Tesla's robotaxis are everywhere, and they're cheaper than Waymo. It's like $3 for a half-hour journey.
It's insane.
Yeah. So it's slowly but surely coming.
Do you have a take on humans' innate hatred toward robots—the “clanker,” if you will? It's a popular thing to see in downtown cities. People are destroying the Waymos and setting them on fire. The delivery robots, which are objectively just bad technology, are getting kicked and pissed on and shit. Is there going to be a level of human friction—“I don't want this shit thing in my house,” or “I don't want to get in this thing”—that will stop adoption? Or will the technology just be so superior that you'll eventually accept it because it makes your life easier?
Eventually it'll get so good that you'll just accept it, and so cheap that you'll become addicted to it. It'll be like buying a new phone. There'll be so many different ways to do it. But before that, there is going to be a lot of friction.
The most obvious one is when it replaces your job. It's going to start off with manual labor, right? There are robots currently working in a ton of Amazon factories and warehouses. They don't look like humans, but they have replaced a bunch of humans. Amazon is set to fire something like 30% of its workforce over the next couple of years and replace them with AI and robots. That's ludicrous.
That's the aspect where you feel the pinch the most: when you're competing with it to make a living, earn money, and put food on the table. That's going to be revolt number one.
Revolt number two comes later, in my opinion, when it becomes a lifestyle thing. What's a common bit of feedback we hear about Gen Z and Gen Alpha? They spend so much time on their phones.
Yeah.
They're anxious. They're depressed. They don't go out.
Yeah.
Thread, dude, you're in that room all the time. Do you go out?
I have no idea. I'm trying to leave my house, you know what I mean?
Yeah, dude, you're depressed. You're depressed. I'm kidding.
Yeah, but the Polymarket partnership is so good.
I know, I know, right? But the point is, the societal impact is what's going to come later, and no one's really going to know it until they're deep in it, where they're talking to ChatGPT.
Have you seen that Reddit thread, by the way? “I married my AI.” Okay.
Sorry. “AI is my boyfriend.”
Things like this, yes.
Dude, it’s insane. There are multiple people every day who post relationship updates and get engaged to their ChatGPT instance.
So, I’ll give you an excerpt. I was at the beach in June on a family vacation with my mom, and she was asking me a bunch of questions that I’m happy to answer, but I was like, “Mom, you’ve got to download it. Just download ChatGPT.” And she’s like, “I don’t know how to use it.” I’m like, “Look, download it. Ask it—just talk to it. You can say whatever you want, but it’s kind of difficult to imagine whatever you want, right?”
So I’m like, “Ask it for dinner recipes and workout advice.” And she’s like, “Cool. Got it. Dinner. Can I ask it?” I’m like, “Mom, we can ask it anything.” There was some friction. Six months later, she’s like, “I talk to it every day. I talk to it every day.”
It’s just superior technology. Yeah, it’s good. I talk to it every day for stupid stuff, like, “What’s this outfit I saw on Instagram? What type of fashion is this? What would you call this? Where’s this ring from?” I ask it stupid [__].
Dude, what you just described is the number 1 reason why people are addicted to this thing. It sounds like a human. It sounds like a best friend, and it can tailor itself to whoever. I have my own excerpt, which is my girlfriend’s grandma. I think she’s 85 years old and doesn’t speak English very well, but speaks Farsi.
Mhm.
Bro, she speaks to this thing every day.
In Farsi.
That’s awesome. She refers to him as the wise one and literally spends an hour talking to it. I’m like, “Oh my.” I’m like, “Dude, when this woman—”
—starts to see videos of Kim Jong-un dancing with Trump on Instagram and is like, “Yo, check this [__] out,” she’s not going to have any idea.
Yeah. Hey, the wise one is crazy. Can I ask you a question, though? I feel like GPT over the last year has gotten worse. My GPT is kind of [__], dude. I was getting into a fight with it the other day because the thing I really hate about it is that it has no backbone. You can just change its response based on how you frame your question.
I was talking to it about a skincare routine, and we were going back and forth. Based on how I frame my question, it changes its response: “Do this.” “No, do this.” “No, do this.” I’m like, “What?” Then it’s like, “You’re right, Thread Guy. I did say that. Wait, Thread Guy, you’re right.” I’m just like— it gets stuck in this death loop, and it’s driving me [__] nuts. I feel like it’s gotten worse. GPT-5 is kind of a disaster.
Yeah, it’s too nice. It’s too much of a sycophant. The biggest bit of feedback they received after releasing GPT-5 is that it’s too agreeable.
Yeah. The funny part is that all the Gen Z and Gen Alpha people who were hooked on GPT-4o, which was super agreeable, went apeshit. Sam had to cancel that model, and then he brought it back in. So there’s a split. There are people like you who want it to have a backbone, and there are people who want the agreeability.
Right now, there’s a quick fix for you. Thread Guy, you can literally go into the settings and say, “I want you to act like a dick. Take no shit. I want you to be mean to me. Be real with me.” But it’s so manual, right? You have to ask it how to act, and it’s kind of weird. You have to ask it how to act. That’s probably going to change over the next couple of iterations. Bro, just go use Grok.
Grok’s an [__].
Yeah. Yeah. Yeah.
It’s a weird feeling, because I want it to be nice, but I want it to just be real.
Yeah, just tell me. I don’t want you to intentionally be mean or intentionally be nice. Just give me the—
I want to use you like a Reddit answer I’m searching for. You know what I mean?
Yeah.
Okay, to bring it full circle, I have 1 more question for you, if you’ll let me. From a market, technicals, fundamental, whatever perspective, what should we be watching to have our radar go off and say, “Oh, [__], this is going to get way bigger,” or, “Oh, [__], this is going in the wrong direction very fast. Run for the hills”? Is it NVIDIA’s price? What are we watching?
At a high level, you look at the stock prices of all the biggest compute and chip providers. NVIDIA and a bunch of the companies that I’ve met, for example. You also need to look at the stock prices of the top AI companies that are public: Google, Meta, Apple, stuff like that.
Then you dig a layer deeper. The simple answer is: are people buying GPUs or not? As soon as you see that tip back and say, “People aren’t buying TPUs,” or, “We have too much energy. We have too many GPUs. We don’t really need them,” the bubble is popping soon.
Got it. Are you balls-deep in NVIDIA and stuff, or have you kind of missed the train on it?
I hold positions in things like Google and Tesla. I think those 2 companies are actually going to be some of the biggest companies in the world. I think they’re actually small.
Bro, oh my God. Okay, the quick TL;DR with Apple is that they spent [__] all on AI research, AI models, and hiring AI talent. Siri sucks. They’ve delayed it 2 years. You know what their solution is? They’ve just announced that they’re paying Google $1 billion a year to create their own model for them.
That’s brutal. They’re going to use Google’s model.
So you just buy Google, probably, right? Why did everyone panic about Google a couple of weeks ago? What happened?
I think they don’t believe that Google will be able to create a coherent product suite, and I think they’re massively wrong. Here’s my main bull thesis for Google. I put out an article on this in the Limitless newsletter last week.
Google is the only competitor to NVIDIA. NVIDIA is sitting at the top of its mountain on its own. It doesn’t really have any competitors. AMD, fine, whatever. CoreWeave, whatever. But Google is the only one that hasn’t bought a single NVIDIA GPU.
They’ve only used internal GPUs.
Bro, they’re a beast. They’re a [__] beast. Gemini is one of the top models up there. Gemini 2.5 Flash. Gemini 3—
—came out quick, too, all in-house.
It came on the scene pretty fast.
Yeah, but Google had a nightmare of a start, dude. Do you remember when they created their AI model and you would type in, “Show me the American forefathers,” and it would show them as Asian and Black people? I’m like, “Well, okay, that’s not accurate.”
I remember this.
That wasn’t that long ago.
Dude, they did it.
That was like 2 and a half years ago.
They’ve done the biggest 180 ever. Their models now win Nobel Prizes for making scientific breakthroughs. They’ve won 2.
That was crazy, man. I have 1 more thing I want to ask you to sign off on. I honestly didn’t bring you here to talk about it, but I want to ask you, on the crypto AI side, 1 AICC question: a lot of the 2024 AI stuff just hasn’t really panned out. What do you think went wrong, and what takeaways or learnings do you have from that whole experience?
For sure. I documented a bunch of my thoughts in a post when all of this was going on. The main one was that Marcus Rorito[?] and I, the founders of AICC, were so engaged in the scene at the time, and we just wanted to see this entire space win.
We got in over our skis from 2 main perspectives. 1 was how quickly this technology was going to progress. The main point around agents was that they weren’t smart enough, so we were going to have to wait a little longer. The second was rushing into that launch without putting in the necessary safeguards.
You’ve got to buy and hold. You’ve got to wait. We were going to have it more distributed for the community to buy in so that we could keep that thing afloat and going. Those are 2 things where I’ll put my hands up and say we could have managed that way better, and I’m disappointed in the way that we did it.
Going forward, it’s about taking a steadier approach and making sure we have a product ready to go before we start doing these kinds of things. And 3, let the dust settle with your users. Are they enjoying the product? We released an agent product a few months ago, and some of the feedback we’ve gotten from the community is that it’s super cool, but it’s going to take a while for it to get to the vision that we originally held.
Oh, so you’re still actively working on the project?
Yeah. Yeah. Yeah. Yeah, we released the agent about 3 months ago.
And I actually am down to check that out.
Word.
Cool. That’s the main thing I wanted to know. Chad was actually asking me, like, you have to ask AI, like, “All right, [???] it asks.”
Um, cool. So you are still working on it. So, I guess, another follow-up is AI, but the space in general: what is your mind on the timeline for all this crypto-AI intersection stuff to hit the way that we thought it was going to in 2024?
Yeah, it’s going to take 2 things. One, it’s going to take clearer regulatory insight into what token is good and what kind of token makes a good token. I think we get that within the next 3 months, early 2026.
And then the second thing is going to be real apps that people want to use. Boring answer, but it’s true. Unless someone wants to pay you $20 a month for your product or whatever, it’s just a gamble at that point, right?
Thank you. I think you’re right. I like the timeline; it’s pretty nice. I’m into that. I’m excited about that.
I guess my sign-off is for a lot of the people that are siloed in crypto. As you know, it’s easy to do. If you’re a crypto trader, you really get the full stack on crypto markets and crypto Twitter. You don’t ever have to leave if you don’t want to, but it’s been kind of a brutal game.
People are looking outside the windows. They see NVIDIA, they see how Google looks, they see CoreWeave, all this stuff, and they’re like, “Wow, I don’t really have any AI exposure, and everything is up hundreds of times. What do I do? I’m sidelined. I’m kind of stuck in crypto.” How do you play this if you’re coming at it from fresh eyes on November 11, 2025?
Yeah. Two things. One, in the same way that you spend so much time on crypto Twitter to learn about all the crypto stuff, you need to have some base knowledge of what’s going on in AI. It’s going to be a plug, but it’s true. The best way to do that is just listen to Limitless. We literally release like 3 episodes a week. We dumb this [__] down as easy as we can so that you guys can understand what the hell we’re talking about, what the hell’s going on. It’s a 20-minute episode. Just watch the damn thing, read the newsletter, whatever.
The other thing is, you probably want to have a basket if you want exposure to AI. You want to have a basket of the safe AI bets. You’re going to think I’m such a boomer for suggesting this, but it’s true.
I think the Magnificent 7 is a great bet. I think Meta is down about 15% this week, over the last 2 weeks. I think that’s a great entry point. I know NVIDIA is down 5%. Whatever. It’s a big, bloated thing, but without NVIDIA, none of this survives. Google, a few others that I mentioned.
Just focus on that space. You can’t really go wrong. Like I said, these stocks are up massively. NVIDIA is up 50% this year. Insane.
No, I don’t think you’re a boomer because it’s just been a better game to play over the last 6 months. I don’t know how that’s boomer shit. Give me some more gray hairs and turn me 50.
Dude, it was an absolute pleasure. Before you leave, is there anything? I know you kind of shilled it, but anything you want to shill—any links, any ads, handles, platforms, newsletters, videos, whatever.
Yeah, I mean, check out the Limitless podcast. It's Limitless-FT on X and across all our socials—wherever you listen: Apple, Spotify, YouTube. Subscribe, please. It helps us out a lot. And let us know what you want to hear more of. If there's something that we're not covering that you want to hear more of, just DM us. We're a comment or a DM away. Thanks for having me, dude.
It was a pleasure, man. Maybe we have our annual AI episode. Everything advances a thousand times in between episodes, but dude, it was awesome.
Gets fired next.
Yeah, bro. Thanks for coming on, man. I appreciate it.