Hook
A sell-side strategist screams that AI will make half the S&P 500 worthless within five years. Cloud providers have $2 trillion in remaining performance obligations. Compute demand is about to explode 20 to 30 times. The narrative is seductive. The data is a house of cards.
I’ve seen this pattern before. In 2021, I ran a bot team during the Bored Ape minting race. We spent $2,000 on RPC nodes to secure speed. Twelve NFTs later, $40,000 in profit. The lesson? Speed beats narrative every time. But speed without verification is just noise. The same applies here.
When the code bleeds, the ledger keeps the truth. Let’s audit this thesis.
Context
Jordi Visser, a macro strategist at 22V Research, published a piece circulating through blockchain/Web3 channels. His core argument: AI agents, full autonomy, and humanoid robots will drive compute demand to infinity. Traditional moats collapse. Buy Nvidia, Marvell, Eli Lilly, Caterpillar, and Modine. Allocate 10–20% to digital assets and frontier AI. Dump everything else.
On the surface, it sounds bold. Under the hood, it’s a fragile extrapolation dressed in urgency. Visser is a macro guy, not an AI engineer. His claims about compute multiples and corporate destruction rely on selective data and omitted risk dimensions. As a quantitative strategist who audits protocols for a living, I dissect this sort of narrative with surgical precision.
Core
Let’s break down the critical flaws. Each one is a leverage point that could snap the thesis.
1. The 20-30x Compute Leap Has No Technical Foundation
Visser claims consumer AI agents and full autonomy will require 20 to 30 times current compute. He provides zero models, zero inference throughput data, zero context window assumptions. This isn’t analysis. It’s a guess drawn from a curve that assumes linear scaling of both technology adoption and hardware efficiency. Reality is messier.
During my 2019 audit of BZRX, I learned that technical precision is the only honest currency. Visser’s prediction confuses training and inference. Consumer agents will drive inference demand, but training demand hits diminishing returns after model maturity. The 20-30x figure lumps them together. If you separate them, inference might grow 10x over five years, while training plateaus. That’s still huge, but it breaks the exponential narrative.
Also, the underlying assumption that consumer AI agents (voice-interactive, dynamic workflow agents) are ready for mass deployment ignores engineering bottlenecks. Long-term memory, multi-step reasoning, and robustness are not solved. I’ve built trading bots that rely on these capabilities. Trust me, they fail often.
2. The $2 Trillion RPO Is Not an AI Demand Signal
Visser cites cloud providers’ $2 trillion in remaining performance obligations as proof that compute demand is infinite. This is a classic data misinterpretation. RPO includes all cloud services – storage, databases, legacy IaaS – not just AI compute. It also represents signed contracts, not immediate spending. Customers can cancel or defer. In a downturn, cloud budgets get slashed first.
I’ve structured options strategies around cloud capex cycles. The correlation between RPO and realized compute spend is loose. Using it as a proxy for "no idle capacity" is a rookie mistake.
3. The Nvidia Monopoly Assumption Ignores Competition
Visser bets big on Nvidia as a "ten-year low valuation" buy. He ignores AMD MI300X, Intel Gaudi, and startups like Cerebras and Groq. These are not theoretical. AMD’s MI300X is already deployed in Azure clusters. If Nvidia’s pricing power erodes, the margin story collapses. The forward PE of 40-50x already prices in perfection. Any market share loss triggers a re-rating.
During the 2020 DeFi Summer, I watched lending protocols fight for liquidity. The winner wasn’t the one with the best tech – it was the one with the lowest fees. Nvidia faces the same dynamic. Infrastructure superiority matters, but not when the competitor is cheaper and good enough.
4. The Samsung Profit Data Is Simply Wrong
Visser claims Samsung’s profit is $217 billion. That’s absurd. Samsung’s 2024 forecast is around $30-40 billion. He might have confused revenue with profit, or used a peak year without context. Either way, this isn’t a rounding error. It’s a data fabrication that undermines every other number in the piece. If one key data point is off by a factor of five, why trust the rest?
5. The "Half the S&P 500" Collapse Timeline Is Extreme
Visser says 5-10 years, 50% of companies lose moats. History says no. The internet took 20+ years to disrupt brick-and-mortar retail. Even then, Walmart survived. AI erodes cost moats, but regulatory moats (banks), brand moats (Coca-Cola), and network effects (Visa) are stickier. The 50% figure is a shock value statement, not a forecast. I’ve been through the Terra collapse. Panic selling wiped 80% of my portfolio. I shorted the rubble and profited $15,000. That crisis taught me that survival comes from hedging, not from predicting doomsday on a fixed schedule.
6. Ethical and Regulatory Risks Are Completely Ignored
Visser’s thesis assumes no major AI safety incident, no regulatory clampdown, no social backlash. This is the biggest blind spot. The EU AI Act is already in force. China requires model approval. A single deepfake election scandal could freeze consumer AI adoption for years. Compute demand would crater.
In 2024, I built a Python script to scan Deribit options data for volatility arbitrage. The regulator hadn’t written the rules yet. Now they have. The same will happen to AI. Code is law until the oracle fails.
Contrarian
The mainstream takeaway from Visser’s piece is "buy Nvidia and ride the wave." The contrarian view is that the wave is already priced in, and the real opportunity lies in the bottlenecks and the hedges.
The Bottlenecks Are the True Alpha
Compute demand is real, but supply constraints – advanced packaging (CoWoS), HBM memory, power, data center construction – will cap growth. The 20-30x figure assumes unlimited supply. Physics disagrees. Companies that solve these bottlenecks (Caterpillar, Modine, Vertiv) have more predictable revenue than GPU makers. I’d rather own the picks and shovels than the miners.
Short the Bubble, Not the Innovation
If Visser is right about disruption, traditional stocks fall. But AI stocks also fall if a recession hits or capex is cut. The correlation is higher than he admits. A smarter play is to short overvalued incumbents (e.g., legacy SaaS) while buying deep out-of-the-money puts on the AI hype basket. That’s a true hedge.
The Robotaxi Reality Check
Visser invokes full autonomy as a demand driver. Waymo and Tesla are still bleeding money. Regulatory approval for scale is years away. The computing power required for L5 is immense, but the timeline keeps slipping. I’ve analyzed the Deribit volatility surface for mobility stocks. The option market implies a low probability of mass adoption before 2028.
Takeaway
Visser’s thesis is a generous fantasy dressed as hard truth. The direction is right. The magnitude is wrong. The data is fudged. The risks are buried.
Institutional money is already rotating into AI infrastructure. Retail is late. The battle trader’s edge is not in buying the narrative – it’s in shorting the narrative’s excess when the margin of safety evaporates.
black box
Arbitrage is just violence disguised as math.
When the code bleeds, the ledger keeps the truth.
