Raoul Pal: AI is activating a new business logic, and the computing power and energy bottleneck is triggering a supercycle

Raoul Pal and Jordi Visser analyze the AI-driven new business cycle, pointing out that computing power and energy will become key bottlenecks.
(Background: Legendary investor Raoul Pal—what exactly does the future world of AI + Crypto look like?)
(Additional context: Raoul Pal lists 12 reasons to go bullish against the tide: accelerating global liquidity expansion, and the crypto market’s “most oversold in history” reversal)

Table of Contents

Toggle

• Computing Power & Energy: The AI Supercycle Begins
• Bottleneck Effect: Capital Concentrates on Infrastructure
• Sector Rotation: From Chips to Biotechnology
• Personal Brain: The Knowledge Management Revolution
• Tokenization: Unlocking Dormant Asset Liquidity

Macro investor and co-founder of Real Vision, Raoul Pal, invites Wall Street strategist Jordi Visser to the podcast. In the interview, Raoul and Jordi break down how AI is building a supercycle powered by computation, energy, intelligent agents, data centers, and explosive growth in intelligence. They also examine how cryptocurrencies, tokenization, and the data economy can open up new markets.

Host (Raoul): Today, my regular guest and good friend Jordi Visser—this isn’t really an interview; it’s a joint deep dive into what’s happening, how to measure it, and how to seize opportunities. Jordi, how have you been lately? What are you thinking about?

Jordi: I want to start with your and Julian’s recent conversation. You talked about the shift from “labor and capital” to “computing power and energy,” and it really resonated with me. In the past, to push a business forward, you needed money, you needed to hire people, you needed office space. But in the “computing power and energy” world, the rules are completely different. I’ve been writing about the AI cycle; in this new world, if we can’t manufacture all the chips we need or don’t secure enough electricity, that creates bottlenecks from supply and demand imbalances and shortages. These bottlenecks may slow the pace of these companies’ profits, but it’s not because there isn’t demand—it’s because demand is too big.

Host (Raoul): I’ve built an indicator dashboard to monitor the index-like growth of intelligent output per unit of energy. In the past it was Moore’s Law; now, with GPUs and AI, it’s already shown exponential growth on a logarithmic chart (double exponential), which is what I call Riedel’s Law (exponential squared). When you factor in that data center construction is only about 30% of what’s expected to be announced, take into account competition between the US and China, and consider that no single leading frontier AI company can dominate on its own, you’ll see that it’s almost inevitable that this becomes a “supercycle.” These bottlenecks will only slow the pace. To break through them, we need to build infrastructure like power grids—this will be the largest capital expenditure (Capex) cycle humans can ever witness.

Jordi: I completely agree. In surveys like PMI (Purchasing Managers’ Index) when we look at business cycles, we often see that what’s behind a high PMI is actually supply chain bottlenecks and rising prices, while the new orders index may already have fallen to 50. But let me dig one level deeper: without enough data centers built, how are we increasing intelligence? It’s because, at the algorithm level, human feedback reinforcement learning and reasoning capabilities have improved. This isn’t the standard process of “drill more oil wells to get more oil,” nor is it “build more data centers to get more intelligence.” This intelligence unfolds through recursive self-learning and algorithmic improvements, which is what gives rise to double exponential growth. The speed is so fast that parabolic linear growth has become the norm—this confuses many people because it’s rarely seen in emerging markets in the past. From Jensen Huang’s January talk at CES about needing to build an AI agent economy, to the big players rushing in at the March Morgan Stanley TMT conference, people only then realized how enormous the numbers are. This phase is different from the past three years of purely boosting IQ. In the next year, the results of recursive self-improvement will shock you even more.

Computing Power & Energy: The AI Supercycle Begins

Host (Raoul): But the market currently can’t put all its attention and capital into one stage yet—it hasn’t fully worked out what the AI agent economy means. Historically, the total addressable market (TAM) for business has been humanity. Now, TAM is expanding infinitely. So I see market rotation into different segments; it won’t always be just NVIDIA going up. When bottlenecks like power show up, to solve the goal of intelligence per unit of energy, all “roadblocks” get cleared, and capital and attention concentrate at the bottlenecks. For example, the peptides and gene science you mentioned earlier—those will also see rotation.

Jordi: Exactly—energy is a very obvious bottleneck. That also explains why NVIDIA and Siemens recently announced a partnership on solid-state batteries in China, which use a lot of silver rather than lithium. Everyone should pay attention to how much silver solid-state batteries actually require. If we have energy storage innovations to handle peak power demand, the US power grid is actually sufficient until 2030. Jensen Huang once divided AI into five layers: the bottom is energy, chips, and infrastructure; then come the models; and at the top is the application layer. But in the application layer, people always keep focusing on the old-fashioned SaaS model. Today’s application layer, which truly absorbs the most capital, is “human software”—the capital absorbed by Eli Lilly’s (a multinational pharmaceutical company headquartered in the US) GLP-1 weight-loss drugs. They have data centers with thousands of GPUs on-site. I believe the cash flow generated by GLP-1 is, in practice, financing the development of the next phase of human biotech software.

Host (Raoul): It’s like the cash flow generated by Musk’s used-car business feeding back into new ventures. When it comes to solving bottlenecks and efficiency, Musk, to address a global copper shortage, upgraded the Cybertruck’s voltage architecture from traditional 12V to 48V, directly reducing copper usage by 70%. When capital and attention are given, with human and AI intelligence we can always find ways to bypass roadblocks—for example, when we hit oil bottlenecks, we invented shale oil extraction; when we hit data center bottlenecks, we switched to fiber-optic transmission.

Jordi: People still don’t realize the significance of an AI agent world. Think about it: if, back in January, someone said the world’s population would suddenly increase by 7.5 billion, we’d feel resources would run out immediately. But the AI agent world creates dozens of billions, even hundreds of billions of new “thinkers” entering the world—and they consume only one thing: computing power. Digital workers don’t need to buy houses or send their kids to college. People have to reshape their old business-cycle thinking. As Musk said, if billions of agents tirelessly solve problems for us, we’ll enter an era of abundance—where humans may choose whether they even need to work. These massive AI agents are executing countless “Manhattan projects,” ultimately solving all problems.

Host (Raoul): And it’s not just one single big model thinking. OpenAI has 1 billion users; each user is using a different instance of this vast intelligence, and intelligence growth is exponential as a result.

Jordi: To support myself, I started building my personal “knowledge brain.” I transcribed Jensen Huang’s talks, and even uploaded and integrated hours of transcripts of David Ricks, the CEO of Eli Lilly. Using the professional knowledge content of a group of people—or even an individual—as AI work material is far more focused and profound than directly searching across the entire human internet.

Bottleneck Effect: Capital Concentrates on Infrastructure

Host (Raoul): I’m also using a vector database to build my “GMI brain”—containing all the long-form content I’ve written over the past 21 years, video transcripts, and my tweets. I’ve also built a tool called “Lens,” based on my exponential era framework and first principles, which can analyze everything from the US election to market issues. But the problem I face now is that I’m doing everything by myself, and my time is completely not enough.

Jordi: I worked at a hedge fund for 20 years. After it closed, I decided to no longer work for anyone or manage anyone. Asking investors for money and staring at charts in the middle of the night to explain trading strategies made me lose interest. I realized that 8 billion people worldwide don’t know how to respond to the AI era, so I focus on content creation—helping people understand trends with easy-to-understand language. I have no employees—just one assistant—but my business grows quickly and steadily. That’s thanks to AI agents handling everything. Building an AI business comes with astonishing profit margins and very low costs, so you can achieve rapid linear growth and even explosive growth.

Host (Raoul): Then how do you squeeze out time to do all of this? I feel like my time gets swallowed by learning new tools.

Jordi: I have more time than you because I don’t have to handle as many interviews, business trips, or company management tasks. Also, I frequently have ChatGPT help me do the “subtraction”—I ask it, “Should I spend time on this new tool?” and usually it tells me, “Don’t worry about it; it will get easier next month.” That saves me a lot of trial-and-error time.

Host (Raoul): That makes sense. I’m now focusing my energy on building the underlying database engine—what I call my “personal vault.” It stores all your personal files, photos, and phone call recordings, becoming your personal operating system, which can be used for your personal brain or for monetization. Once information on every computer and phone can be retrieved by AI in one second, many people haven’t realized how disruptive that will be.

Jordi: I completely agree. The most typical example is when a loved one passes away—handling estate files is a massive nightmare. Over the past 18 months, I’ve been going through this myself. I gathered all files into one folder and connected them to Claude Opus 4.5. When someone calls to ask about assets or files, Claude can immediately find the exact answer and send it out—showing the power of future personal assistants and databases.

Sector Rotation: From Chips to Biotechnology

Host (Raoul): I’m now using a tool called Granola. It can not only do real-time meeting transcriptions, but also connect to large models and become my long-term knowledge base. All conversations get fed into my knowledge brain, which will never forget what we discussed last time. Today’s biggest bottleneck for AI companies is “insufficiently persistent memory,” and this long-term database layer breaks through that limit.

Jordi: Right now, I run OpenClaw—the Kim K2.5 system for Chinese models—on my Mac Mini, and I run GPT-5.5 on a top-tier M5 chip Apple laptop. It serves as my assistant and runs my investment portfolio technical algorithms. It can always help me recall the inspiration memo from two weeks ago that I wrote on a plane.

Host (Raoul): To avoid the friction of exporting data back and forth between different devices, I just bought the top M5 Apple laptop too, and now I carry it around everywhere.

Jordi: By the way, if you don’t hire people, the cost savings and extremely high profit margins of AI businesses will push you to go crazy investing in hardware. To run the Hermes agent, I’m planning to buy a Nvidia DGX. After listening to Dennis Casabus’s interview on Y Combinator, I’m sure open-source models will keep getting smaller and better, and eventually we’ll move toward “edge AI.” In the future, running local models on your own devices and computers will be an extremely important part.

But learning to use AI—especially on edge devices—is like learning to ski or play golf: you have to abandon old habits and put in a lot of practice. When I’m walking and encounter inspiration while listening to podcasts, I immediately pause, record with Whisper, and then generate a draft with ChatGPT. So people have to buy the best phones and computers, treat it like a paid education investment, and use it constantly—walking, driving, taking flights—to explore and develop their own workflow.

Host (Raoul): This kind of around-the-clock thinking really does generate lots of inspiration. I’ve been watching the issues of government debt and healthcare longevity. US household net worth is $180 trillion, while debt is only around $40 trillion, so it’s not a big problem. But what I care more about is this: as lifespans extend and medical spending takes a larger share of welfare, how will AI profoundly impact this welfare system?

Personal Brain: The Knowledge Management Revolution

Jordi: The debt-to-GDP ratio will collapse. As for welfare and population aging, AI can certainly help reduce welfare costs, but the harder part is how to restore a sense of value in the elderly within society or the economy. The boundary of “work” today is already blurred. Many YouTubers’ podcasts are themselves designed to capture human attention—this is a classic kind of job in the post-AI era. In the face of future-work anxiety, I recommend reading The Daily Stoic. For thousands of years, humans have worried about “What if we don’t need to work?” yet humans always find ways to adapt and evolve.

Host (Raoul): I completely agree. On resolving debt, another key factor comes to mind: tokenization. Two-thirds of the world’s massive assets (like real estate, private equity, venture capital, art, etc.) are illiquid. Once these assets are tokenized, transparency and liquidity get created for dormant assets. And once liquidity exists, GDP will inevitably rise. That’s why tokenization is central to how I think about longevity, ownership, and welfare issues.

Jordi: I’ve written a series on the “Invisible Economy”—the AI agent economy. Crypto tokens are essentially machine-readable bundles of information. Google processed trillions of tokens last year, and now it has reached the quadrillions. To train AGI into ASI (superintelligence), they need to absorb all digitized information—everything from universities to scientific data. The world’s biggest market will no longer be the human asset market, but the data market that AI craves. In the future, countless AI agent programs will conduct millisecond API-call trading. It’s a staggering but invisible trading market.

Host (Raoul): That’s an excellent perspective. And because of this large trend, we need to re-examine the concept of “bubbles.” A bubble is the combination of price and time. If the Mag 7 rise to $20 trillion within a year, that’s a bubble. But if they grow 20 to 30 times over 15 years, that’s structural tailwind. People always feel that if it rises too fast, it must be a bubble—but actually, large companies’ earnings growth and the surge in their stock prices are moving in sync, and even the P/E ratio is declining. Indicators across industries show hockey-stick straight-line spikes.

Jordi: Over-knowledgeable veteran traders can actually miss opportunities. They carry too much baggage from similar “internet bubbles” or the memories of the 1929 Great Depression, leading them to force-fit wrong assumptions and comparisons onto the current market. It’s like using Claude’s large model to write code—sometimes admitting you don’t know everything and dropping unnecessary context can make the system work better.

Host (Raoul): I’m curious about your take. Right now, the profit expectations for AI hardware and infrastructure stocks are very good. That’s unfavorable for narrative-driven assets (like Bitcoin and crypto), because capital gets pulled away. Many people haven’t realized the breakout power of the AI agent economy yet. Now the big funds managing trillions have already gotten on board, and profits are very stable. This won’t just lead to capital rotating into longevity-theme drug stocks and other sectors in a normal pattern—it will also put the crypto market to the test. When traditional stocks produce such huge returns, what force will shift capital back into crypto?

Tokenization: Unlocking Dormant Asset Liquidity

Jordi: First, even giant stocks will stop surging if liquidity and attention are insufficient. We also saw this kind of rotation between 1995 and 2000. As you said, “infrastructure bottlenecks” will cause these AI giants to enter a digestion plateau because their capacity can’t keep up with demand. And when capital overflows, blockchain technology—decentralized identities (IDs) and blockchain tech that satisfies AI agent trading attributes—will show its absolute advantage again. I’m especially focused on Layer 1 protocols. Although some people think AI will disrupt the existing SaaS software ecosystem, I see it differently: people still embed AI into existing billing and accounting software every day via APIs, and software still has value.

Host (Raoul): The AI hardware world indeed faces constraints on commodity output. For example, semiconductors need specific lithography gases and petrochemical inputs. Once production bottlenecks occur, infrastructure expansion may not reach the scale people expect, which will squeeze out part of the valuation bubble in Capex. I believe there are two prerequisites for crypto to experience a “third wave.” First, AI’s physical infrastructure bottlenecks cause capital to flee, and it then seeks safer targets that are based on the digital world and don’t require physical expansion (like AI agent application layers). Second, tokenization brings liquidity to the traditional massive dormant assets, and institutional capital starts to question the chips it holds and seeks entry into the crypto world again. We’re currently in a bottleneck digestion phase that requires extreme patience. What do you think about recent large-scale IPOs from some tech companies?

Jordi: Companies like Google are raising money now because this is a competition for limited capital. OpenAI, Anthropic, or SpaceX can’t borrow through credit markets—they can only pull cash from the stock market. I believe these three mega IPOs may signal a peak in infrastructure Capex trades in the short term, but not a peak for the entire market. Capital will flow into software and other sectors.

Host (Raoul): Yes. In my report, I mentioned that we just experienced an “AI Capex buffet.” Everyone ate so much they’re full. Now we need 3 to 6 months of consolidation and digestion to reset capital. As these underlying large trends keep fermenting, if stocks like NVIDIA stop rising, the probability and speed of capital flowing back into crypto assets will increase dramatically.

Jordi: Crypto definitely needs this digestion period too. It’s like when gold ETFs launched: the adoption of Bitcoin ETFs and US presidential support pre-empted a huge wave of demand. With global liquidity expanding only about 10% per year, the capital that was front-loaded for a year would inevitably need a year to digest. After the brutal 2022 bear market, and the emotional release in 2024, a painful consolidation period is unavoidable.

Host (Raoul): But when you go to massive crypto conferences like Miami Consensus, you’ll find almost no retail investors—although some retail investors can afford expensive tickets. Mostly you’ll see major banks, financial institutions, and established players focused on stablecoins and asset tokenization. This proves that the underlying technology trend is extremely strong. Once the infrastructure tailwind in the blockchain/blocks space starts to cash in and gets monopolized by a few core projects, token prices will inevitably move upward.

Jordi: Many people unfortunately get lost in the “trees” of short-term trading, ignoring the “forest.” If you’ve read Elliott Wave Theory, you know we’re waiting for that “third wave” that makes you a lot of money. Just like I caught the big rally in Micron, I’m patiently waiting for the “third wave” in the crypto market—the so-called Banana zone.