The message landed in my inbox on a Tuesday morning, forwarded by a partner at a Geneva-based fund. “Jordi Visser just dropped his AI manifesto. Want to tell me where he’s wrong?” I opened the PDF. The claims hit like a hammer: AI compute demand will explode 20-30x; half of the S&P 500 will be “worthless” within a decade; traditional macro analysis is “99.9% wrong.” The tone was urgent, apocalyptic, and intoxicating. Then I checked the data. Visser claimed Samsung’s 2024 profit would hit $217 billion. The real figure, according to consensus estimates, sits closer to $300-400 billion. A $217B profit would make Samsung the most profitable company on earth, exceeding Apple by a factor of three. That number felt like a hallucination.
Code speaks, but culture listens. Visser’s article, published via a blockchain/Web3 distribution channel, was never about technical precision. It was a narrative weapon—a story designed to reallocate capital. And as a Narrative Hunter, I don’t dismiss narratives; I dissect them. The question isn’t whether the story is true, but why it’s being told now, and what it ignores.
Jordi Visser is a macro strategist, not an AI engineer. He’s spent decades on Wall Street constructing big-picture theses. His current view: AI is an “exponential, unstoppable force” that will destroy all existing business moats, and the only safe harbor is a concentrated bet on compute infrastructure—Nvidia, Marvell, Caterpillar, Modine—plus digital assets. The framing is seductive because it offers certainty in a market that has none. But under the hood, his argument chain has more cracks than a dried-out riverbed.
Context: Visser’s analysis is a prime example of ‘narrative inflation’—a phenomenon I documented during the 2021 NFT mania as a Cultural Semiotics Ethnographer. Back then, the story was ‘NFTs are art’. Now it’s ‘AI compute is infinite’. Both narratives serve to justify extreme valuations and herd behavior. The structure is identical: take a legitimate trend (AI compute is growing), amplify it to absurd proportions (20-30x), ignore counterevidence (capacity bottlenecks, regulatory risks, algorithmic improvements), and sell a simple investment thesis. In the 2020 DeFi Summer, I identified the ‘yield trap’ when everyone chased insane APRs. Now, Visser is framing an ‘compute trap’—convincing retail and institutional investors that Nvidia and its peers have no upper bound.
Core: Let’s dig into the technical foundation. Visser claims that consumer AI agents (voice-activated, dynamic workflow tools) and full autonomous driving will require 20-30x current compute. No source, no scaling law projection, no breakdown between training and inference. As someone who reverse-engineered smart contracts for fun in 2017—back when gas optimization was an art—I know the difference between a plausible trend and a back-of-envelope extrapolation.
First, the training vs. inference confusion. AI training demand is indeed massive, but once a model reaches a certain capability, training may plateau (diminishing returns on more data and parameters). Inference demand will grow, but not at the same rate as Visser implies. Consumer AI agents haven’t even proven product-market fit yet. The number of concurrent users for a hypothetical 2026 agent is pure speculation. 20-30x is a poetic number, not a mathematical one.
Second, the bottleneck reality. Even if demand grew 20x, supply cannot follow. Advanced packaging (CoWoS) and HBM memory are already constrained. Nvidia’s lead times remain over a year. Data centers need 3-5 years for permitting and grid connection. Electricity supply is the next wall. Visser mentions Caterpillar and Modine—power and cooling—but he doesn’t calculate whether global power capacity can absorb a 20x compute surge. AI training can consume 50 MWh per session. A single Blackwell cluster pulls 60 MW. Multiply that by thousands of clusters, and you need new nuclear plants. The narrative ignores physics.
Third, the RPO misinterpretation. He cites “$2 trillion in remaining performance obligations” among cloud providers as proof of insatiable AI demand. In reality, RPO includes all future cloud services—storage, databases, networking—not just AI. And RPO is a contractual commitment, not a guarantee of delivery. It can be canceled or delayed. Visser’s conflation is like looking at an airline’s booked tickets and concluding every passenger will board a plane that hasn’t been built.
Now, the business moat argument. Visser says Salesforce and Adobe will lose their moats “overnight” because AI makes competition cheap. This is the same fallacy that dominated early internet hype: ‘Amazon will crush every retailer’. Some retailers died; most adapted. The true moat for enterprise SaaS isn’t just software—it’s data gravity, workflow lock-in, and a decade of customer trust. Salesforce has 150,000 customers with custom configurations. Replacing their CRM involves political risk inside the organization. AI agents help newcomers but they can also be integrated by incumbents. The death of classic SaaS is likely, but it’s a slow bleed, not a heart attack.
The Cassandra complex is real. I’ve been called a Cassandra before—in 2021, when I warned that NFT floor prices driven by social capital would collapse, no one listened. Now Visser is playing the same role for AI compute. He’s the prophet warning that everyone else is wrong. But prophets are dangerous when they sell maps to a land they’ve never surveyed.
Contrarian: The counter-intuitive truth is this: the best investment opportunities may not be in compute itself, but in the bottlenecks that Visser ignores—and in the underappreciated moats he dismisses.
First, the infrastructure bottleneck trade. While Nvidia soaks up hype, the real friction lies in power generation, cooling, and data center construction. Companies like Vertiv (thermal management), Quanta Services (grid infrastructure), and Brookfield Renewable (power supply) face less narrative volatility but more tangible growth. Energy constraints will throttle compute expansion long before chip supply does.
Second, the institutional defense. Visser recommends betting against all traditional companies. But many S&P 500 firms—especially in healthcare, energy, and defense—have regulatory or physical moats that AI can’t defeat. Eli Lilly (his own pick!) has a drug patent moat that no chatbot can violate. AI can accelerate drug discovery, but it can’t shortcut FDA approval.
Third, the narrative reversal risk. The digital assets he champions are now tightly correlated with tech stocks. A sudden regulatory clampdown (e.g., EU AI Act enforcement, or a US executive order on compute) could trigger a simultaneous selloff in both AI equities and crypto. The same narrative that lifts them can crush them.
Another rug pull? Or just another myth? Visser’s story is powerful because it sells a new paradigm. But paradigms require proof, not poetry. I’ve been in this industry long enough to see cycles repeat: 2017 ICO hype, 2020 DeFi yields, 2021 NFT identity. Every time, the narrative outruns reality. This time, the narrative is ‘infinite compute’. The reality is we’re still figuring out how to power one data center, let alone twenty.
Takeaway: The next 12-18 months will expose the gap between story and substance. Watch for signals that Visser’s narrative is fraying: a major AI safety event that triggers regulation, a decline in NVIDIA’s quarterly guidance, or a surprise breakthrough in algorithmic efficiency (like sparse MoE architectures reducing compute per token). If compute demand doesn’t hit the 20x mark on time, the assets riding on that assumption will reprice violently.
For now, the smartest move is to go long on bottlenecks and short on narrative. Buy the picks and shovels for energy and cooling. Avoid the digital assets tied to the infinite compute story. And maybe—just maybe—don’t abandon Salesforce quite yet. Code speaks, but culture listens. And right now, the culture is drunk on a story that tastes too good to be true.