Organized & compiled: Deep Tide TechFlow

Guests: EJ and Josh, hosts of the Limitless Podcast

Podcast source: Limitless Podcast

Original title: AI Found the Trades We Missed

Broadcast date: July 8, 2026

Summary of Key Points

The Limitless Podcast has reached its 200th episode. The two hosts did something a bit risky: they fed the transcripts of all 200 episodes to Claude and ChatGPT, asking AI to help them identify hidden investment themes they may have missed, review judgments made over the past year, and forecast the next steps for AI. Fourteen months ago, when they pressed the record button for the first episode, SpaceX was still a private company valued at $350 billion, Anthropic was valued at $61 billion, OpenAI at $300 billion, and GPT-4o was still the flagship model. Now, after going public, SpaceX has a market capitalization exceeding $2 trillion, Anthropic is approaching $1 trillion, and the number of ChatGPT users has more than doubled.

The show unfolds along two threads. One is retrospective: when they first declared "computing power is national security," SanDisk soared 35 times; their bet on memory chips saw Micron rise 180%, and SK Hynix became South Korea's largest company, with Samsung's profits surpassing NVIDIA's. They even mentioned on the show three months ago that they wanted to buy Cursor stocks, resulting in Cursor being acquired by SpaceX for $60 billion. If you created a portfolio of the companies whose founders they have interviewed, the annual return would be approximately four times, with the best being Valor Atomics at thirteen times. The other thread is forward-looking: mentions of terms like superintelligence and AGI in the show have plummeted by 90%, while mentions of Anthropic have quadrupled, surpassing OpenAI. Their three most promising directions are space training models, AI's implementation in vertical fields, and small models at the edge.

Highlights of Opinions

On the Validation of Computing Power and Memory

• "We first said that computing power is a national security issue. Whoever has watts, chips, energy, and cooling has the rest of the economy. Looking back after 200 episodes, this is even more true than it was then."
• "The average return for memory manufacturers is 153%. If you bought Micron at the end of last year, you're up 180%. SK Hynix has become South Korea's largest company, pushing Samsung aside. Samsung had just become the world's most profitable company, surpassing NVIDIA."
• "Three months ago we said we hoped to buy OpenAI stocks, but there was a company that could catch that tailwind, called Cursor. Three months later, Cursor was acquired by SpaceX for $60 billion."

On the Limitless Portfolio

• "If we formed a portfolio from the companies founded by those we've interviewed on the show, the total return from the date of the interview to today would be about four times, with Valor Atomics being the best at thirteen times."
• "Isaiah Taylor, CEO of Valor Atomics, was on the show just after securing a seed round; now it is valued at $2 billion. His logic is simple: data centers are running out of electricity, and he's building small modular nuclear reactors to solve this problem."
• "Boom Supersonic, which originally built supersonic planes, has now also started to produce gas turbines for AI data centers."

On Trend Migration

• "The term superintelligence was mentioned 60 times last year but only 6 times this year—a 90% drop. Cryptocurrency is nearly at zero. Robotics is down 60%, AGI has decreased by 54%."
• "Anthropic had only a quarter of the mentions of OpenAI last year, but this year that number has quadrupled, surpassing it: 806 to 758. A year ago we were saying Claude 3.7 was underwhelming and that ChatGPT was the king. This company has turned things around completely this year."
• "The AI company we mention the most is not OpenAI, nor Anthropic, it's Google. However, Google’s momentum has clearly slowed down."

On Upcoming Bets

• "Fable 5 charges $10 per million tokens for input and $50 for output. If you burn several million tokens each week, that cost hurts. Uber and Meta have already been cutting their AI token expenditures."
• "Most companies and ordinary people haven't really started using these AI tools. The models are already ridiculously powerful, but real-world deployment is still very small. The core question for the next 200 episodes is how the world extracts value from these tools."
• "Whoever solves the energy problem will be one of the most powerful companies on Earth. Whether AI succeeds or not, energy is the ceiling for everything."

Episode 200: Feeding All Transcripts to AI

EJ: Welcome to episode 200. In this program, we've said 1.4 million words, 99.9% are Josh and I debating whose computing power is greater, whose model is better, and which AI company is the most worthy of investment. Before recording this episode, we did something a bit risky: we gave the transcripts of all 200 episodes to Claude and ChatGPT to help us find hidden investment themes we might have missed, to sort through what we've said and judged over the past year, and where AI might be headed next.

Looking back to when we recorded our first episode 14 months ago, the world was completely different. At that time, SpaceX was still a private company valued at $350 billion. Now it is public and has a market capitalization exceeding $2 trillion. Anthropic was valued at $61 billion, now it is nearing a trillion. OpenAI's valuation was $300 billion, and ChatGPT had about 500 million users, now more than doubled. The models are even more interesting: at that time, GPT-4o was flagship, Claude was at 3.7, Gemini was at 2.5. Most absurdly, when recording that first episode, almost all code in the world was still handwritten. Now, it has turned around.

So what we are going to do today is review what happened over the past year and look forward based on these trends. Let's start with the judgments we've gotten right.

Computing Power is King: A Validated Early Bet

EJ: Josh, do you remember episode four? On June 5, 2025, we said: computing power is now a national security issue. Whoever has watts, chips, energy, and cooling has the rest of the economy. Our meaning was straightforward—computing power is king; whoever owns GPUs and can power them can produce the best models.

Looking back after 200 episodes, this is even more true than it was then. In the last few months, Anthropic has signed four new computing power contracts. OpenAI has been ramping up production wildly; their aggressive bets on acquiring computing power have proven very correct. They haven’t placed any restrictions on users, which supports the core demand for reasoning, allowing AI agents to operate 24/7. Computing power is king; this is one of our earliest judgments.

Josh: I just checked, when we said that, SanDisk rose 3500%. That's a 35-fold return. If you had told someone back in June 2025 that the federal government would invest in companies like Intel and that their stocks would rise this much, we might have invested too, but we would also be pretty surprised at how important this computing power race has become. The most scarce resource in the world right now is the energy and memory that powers GPUs. Noticing this early is truly remarkable. I genuinely wish we had invested at that time.

Memory Chips: 153% Average Return

EJ: Another asymmetric bet is AI chips, especially memory. When predicting last year, our core logic was: memory is expensive; it constitutes a large portion of the entire AI system. To run GPUs, you need to remember the entire conversational context with ChatGPT and Claude. Memory prices are likely to rise, and the stock prices of memory manufacturers will follow.

Guess what the average return is for these top three memory manufacturers?

Josh: Infinite. This has probably been the best single investment you could make over the past year.

EJ: It's not that exaggerated, but it's close. 153%. If you bought Micron at the end of last year, you're up 180%. If you managed to buy SK Hynix, it has become South Korea's largest company, pushing Samsung, which had monopolized for decades, aside. Samsung had just become the world's most profitable company, surpassing NVIDIA.

This is a comprehensive arms race where these companies keep surpassing each other.

Cursor: Acquired by SpaceX for $60 Billion

EJ: In March 2026, not long after, we said this was an asymmetric bet: I really wished I could buy OpenAI stocks, but there was a company that could ride this wave, called Cursor.

Three months later, Cursor was acquired by SpaceX for $60 billion.

Josh, should we start managing a fund?

Josh: We need a fund. Does anyone want to invest in us?

EJ: If we really turned these judgments into investments, the returns should be quite substantial.

Limitless Portfolio: Four Times Return in a Year

EJ: In the early days of the show, we interviewed many founders in cutting-edge technology. These companies were carefully selected because we saw great potential in them. We secured exclusive interviews with the CEOs and founders. If we had invested money on the day of the interview, the total return to today would be roughly four times, in just over a year.

The best performer is Valor Atomics, which is at thirteen times. In episode 10, we invited CEO Isaiah Taylor, who builds small modular nuclear reactors. The logic is straightforward: data centers are lacking electricity, and he is here to solve that problem. When he was on the show, he had just completed a seed round, and now it is valued at $2 billion.

OpenRouter is similar; when founder Alex Atallah appeared on the show, it was valued at $500 million, now $1.3 billion. The logic behind OpenRouter is what we have repeatedly discussed: companies won’t only use one model; they will employ different models based on varying scenarios. This is also why Cursor was favored, and that’s why SpaceX spent $60 billion acquiring it. OpenRouter can access early versions of models before they are officially released, including Claude, ChatGPT, and models from China. Developers can use a variety of models without restriction. The user intention data they aggregate is incredibly rich, allowing them to determine what models to create next.

Zipline is also interesting, specializing in drone delivery. When they were on the show, they were displaying early prototypes, and now they operate in several major cities. Perplexity has risen from $18 billion to $21 billion, primarily because Samsung set it as the default AI agent on all phones.

Boom Supersonic is a particularly interesting case. Originally focused on supersonic planes, they have now begun producing gas turbines for AI data centers.

Josh: A 400% return is already insane compared to the market. If we could actually invest that would be fantastic.

EJ: I would agree. But this simulated portfolio tells us that the past year has formed several clear trends: energy is the bottleneck, nuclear energy is one solution; the model routing layer addresses a genuine need; and AI’s deployment on mobile has begun.

Trend Migration: Superintelligence Exits, Anthropic Rises

EJ: Over the past 14 months, our discussion topics have changed dramatically. After analyzing the transcripts from 200 episodes with AI, the trend migration is quite clear.

The term superintelligence was mentioned 60 times last year but only 6 times this year—a drop of 90%. Cryptocurrency is nearly at zero; it is a cold winter. Robotics is down 60%, and AGI down 54%. We have mentioned these big terms less; perhaps it's because they seem to be reaching a point where the boundaries are becoming blurred. When you see these mythos-level models, they might feel like AGI already. Robotics is interesting; it seems we are in an intermediate phase: many companies are building them but haven't officially released them. A new version of Optimus is coming but hasn’t been unveiled, and Figure is also developing new robots but hasn't made them public yet. I guess there will be intensive releases later this year, leading to a resurgence of discussions about robots.

The most astonishing development is Anthropic. Last year, we mentioned it about one-quarter as much as OpenAI and ChatGPT. This year, their mentions have quadrupled, surpassing OpenAI: 806 to 758. A year ago, we were critiquing Claude 3.7, thinking it wasn't that good, and that ChatGPT was the king. This company has completely turned things around this year. Claude Code has only been out a little over a year; last year almost no one used it, but now everyone is.

Another interesting finding: the AI company we mention the most is not OpenAI, not Anthropic, but Google. However, Google’s momentum has clearly slowed down. In recent months, their product iteration speed was absurdly fast, almost new things every week, and they were all quite good. But then it slowed down. Now the real two major players are OpenAI and Anthropic, with no changes in sight in the short term. Perhaps Grok could become a comeback candidate during the year, as SpaceX’s AI team is pushing full steam ahead.

Next 100 Episodes: Three Key Tracks to Watch

EJ: If I have to pick the three most important trends to focus on next, the first is space training models. We first mentioned StarCloud, a startup from Y Combinator that started sending H100 GPUs into space to train models. Now, the entire strategy of SpaceX AI is to launch a large number of satellites to train models in space, likely Grok and other models. This trend will only strengthen, and SpaceX will be the leader.

The second is the implementation of AI models in vertical fields. General large models are excellent for chatbots, but not so effective in niche areas requiring specialized knowledge. Anthropic and OpenAI have both been forming joint ventures over the past few months, raising billions of dollars, sending engineers into these fields to find optimal solutions. I believe the focus will shift to finance and life sciences next. Anthropic has already released Claude Science, OpenAI is releasing genetic benchmarking tests, and has acquired Retro Science Labs for new drug development. Google has AlphaFold working on molecular sequencing and new drug discovery. I am particularly optimistic about this direction, as it could be very close to AGI for science.

The third is edge computing and local models. Uber and Meta have already been cutting their AI token expenditures because cutting-edge models are too expensive. Fable 5 charges $10 for every million input tokens and $50 for output. If you are burning millions of tokens weekly, that cost is painful. More companies will begin to pivot to locally run small models—possibly Chinese open-source models, American open-source models, or models like those being trained by Apple and run on local hardware. OpenAI may release new devices by the end of the year, and Apple is also working on them, with many companies developing devices like glasses or chest-mounted units. These devices are centered around personal AI agents.

Energy and Deployment: The True Bottleneck of AI

Josh: If I were to pick two areas of greatest concern, the first is how the world extracts value from these AI tools. There exists a large asymmetry: a few companies have built tools that are unimaginably powerful, but the world hasn't figured out how to use them yet. Most people and most companies haven’t truly started utilizing them. The models are already unbelievably powerful, but deployment dispersion is still very small. The core question for the next 200 episodes is this: how to distribute this intelligence into daily devices and how to redefine ordinary positions within companies. That's what truly excites me.

The second is energy. Whoever solves the energy problem will be one of the most powerful companies on Earth. Whether AI succeeds or fails, energy is the ceiling for everything. There are several companies doing interesting things right now. Travis Kalanick's new company Adams is building precious metal-driven narrow scenario robots to distribute materials. Tesla’s Optimus is developing humanoid robots, and if they can manufacture at the scale of Model 3 and Model Y—currently the world's best-selling cars—that would be significant.

EJ: Let me add one more. China may soon be closing off a batch of models. This morning, news came that the Chinese government has requested AI labs to sever external access, effectively their version of export controls. They don't want Westerners, especially Americans, to gain access to their models. This may shift to a paid model or even direct privatization. I think this is not good news for the world, but for the next 6 to 12 months, it is a genuine geopolitical risk.

Another longer-term but noteworthy trend is recursive self-improvement (RSI). This refers to AI models being smart enough to create the next generation of themselves. This is a challenging issue, but many smart people predict it will be achieved around early 2027. If it comes to that, AI models could work for you 24/7. However, this also means we will rely on these AI agents more and share more personal data, which concerns me a bit.

Bullish Index: 192 to 68

EJ: Finally, I’d like to share a data point. In the past 200 episodes, we have said "bullish" 192 times and "bearish" 68 times, a ratio of 2.8 to 1. Both of us are extremely optimistic about the future. Regardless of how frightening and intelligent AI tools become, I believe they will bring abundance, GDP explosions, scientific breakthroughs, and new drugs. Furthermore, this transcends age—you could be 24, drop out of Harvard to create chips competing with NVIDIA, or you could be a student in an AP science class discovering something worthy of NASA sponsorship. The latter is the real story of a NASA intern today.

Josh: You can't just jump to this segment without mentioning the rest. The bullish versus bearish ratio of 2.8 to 1 is quite amusing. Also, on the term "exponential," I have claimed it 48 times, while you've claimed it 28 times. The database says "this physicist has a favorite curve." Then there is the so-called "waffle index"—38.9 hours compared to 27.9 hours, one person spoke 40% more than the other. The database notes "the difference in speaking time is significant, with one host speaking 40% more."

EJ: I don't know who you're talking about. I don't tend to hog the microphone.

Josh: Right, you just contribute the value you can.

EJ: 200 episodes. Thanks to those who have been listening from day one. This podcast itself is an accident—David from Bankless asked me if I would like to do my own podcast, and I said sure, then I went on vacation and came back to find that the podcast was already there. Then we just did it, with everyone’s support, creating some decent content. Looking back, it’s been a really cool experience in life. But it still feels early, and there’s so much more to discuss; AI remains the hottest topic, but there are still plenty worth talking about downstream. That was the original idea behind Limitless—to explore the frontier.

So here we are at episode 200; to all of you still listening, you are legendary. Thank you for your ongoing support. If you haven't subscribed or liked yet, try it today; there is no better day than episode 200.

Josh: We are still in the top 30 of global technology podcasts. Our ranking is increasing.

EJ: Here’s to 200 episodes. Cheers, see you in the next 200 episodes.