Let me be direct: Jordi Visser's recent macro note—claiming half the S&P 500 will be bankrupt within a decade while AI compute demand explodes 20-30x—is the kind of narrative candy that passes for deep insight in a bull market. As someone who has spent years mapping liquidity flows across crypto and traditional markets, I have learned to distrust clean, linear extrapolations. The real world is messy, bounded by constraints that no spreadsheet can capture. Visser's argument is seductive. It plays into every fear and greed reflex. But when you stress-test it against the code of actual market mechanics, the faults are glaring. Code is law, but incentives are the reality. And Visser's incentive is clear: produce a shocking thesis that generates attention for his research firm, 22V Research. That does not make him wrong, but it makes his analysis selectively blind. Let me walk through the critical flaws—from data fabrication to missing risk dimensions—and offer a more defensible framework for positioning in both AI and crypto assets.
Hook: The 2170 Billion Dollar Error
The most glaring red flag in Visser's note is his claim that Samsung's operating profit is $217 billion. For context, Samsung's actual 2024 operating profit is expected around $30-40 billion, depending on memory cycle recovery. Even at peak 2022, it was $34 billion. The $217 billion figure is roughly six times that—an error of such magnitude that it signals either deliberate sensationalism or a complete disregard for basic financial data. If we cannot trust a simple profit figure, how can we trust a prediction about 20-30x compute growth or the collapse of half the S&P 500? This is not a typo; it is a warning that the entire edifice of Visser's argument may rest on similarly shaky foundations.
Context: The Macro Watcher's Lens
I operate as a macro liquidity auditor. My training in applied mathematics and my years tracking stablecoin flows, Bitcoin whale movements, and DeFi yield mechanics have taught me one thing: narratives break faster than chains. When a respected macro strategist like Visser publishes an extreme call, I do not dismiss it outright. I examine the assumptions, stress-test the data, and identify where the logic chain fails. In this case, the failure is systemic. Visser conflates training and inference compute, misinterprets cloud providers' remaining performance obligations (RPOs), ignores hardware and energy bottlenecks, and completely omits the most critical variable—regulatory and ethical risk. As an analyst who has lived through the Terra collapse and the NFT liquidity crisis, I recognize the pattern: a brilliant narrative that ignores tail risk often precedes a violent reversion.
Core: Dissecting the Compute Demand Argument
Visser claims that consumer AI agents will require 20-30 times current compute capacity. He offers no model, no scaling law analysis, no breakdown of inference vs. training. The number appears plucked from intuition. Let me apply my own framework—the same one I used to model stablecoin velocity in 2017. If we assume each consumer agent processes 50 queries per day with 10,000 tokens per query (generous), that is 500,000 tokens per user per day. Multiply by 1 billion users: 500 trillion tokens per day. State-of-the-art inference at GPT-4o level costs roughly $0.01 per 1,000 tokens (API price). That would be $5 trillion per day in compute costs—absurd. Clearly, costs must drop 100x-1000x for this to happen. Visser's 20-30x may be plausible in the long run, but he presents it as near-term certainty. Code is law, but incentives are the reality. His incentive is to sell a scarcity narrative for NVIDIA and other infrastructure plays.
Now, examine the RPO claim. Visser points to $2 trillion in cloud providers' remaining performance obligations as proof there is no idle capacity. In reality, RPO is a multi-year contractual backlog including storage, database, and basic IaaS—not just AI compute. Much of that $2 trillion was signed before the generative AI boom. Moreover, RPO can be cancelled or delayed. I have seen similar over-optimistic interpretation in crypto: people mistaking notional open interest in perpetual swaps for real liquidity. The same cognitive bias is at play here.
Another missing dimension: the hardware bottleneck. Even if demand materializes, the supply of advanced packaging (CoWoS) and HBM memory is constrained. NVIDIA's lead times extend to 12 months. Building a new fab for 3nm chips requires $10-20 billion and 3-5 years. Energy constraints are even more binding. A single AI training run can consume 50 MWh. Scaling to 20-30x without a parallel revolution in energy production is physically impossible in a decade. Visser recommends Caterpillar and Modine for data center infrastructure, yet fails to assess how permitting delays, grid capacity, and renewable energy targets will throttle growth.
Contrarian Angle: The Decoupling That Isn't
The most dangerous part of Visser's thesis is his suggestion that investors should allocate 10-20% to 'digital assets and frontier AI names' as a hedge against the collapse of traditional equities. This implies a decoupling between AI stocks and the broader market that is not supported by history. In 2022, when the macro environment soured, NVIDIA lost 50% of its value. Bitcoin dropped 75%. Correlation was near 0.8. The idea that AI and crypto will thrive while the rest of the economy crumbles fails the stress test. More likely, if Visser's prediction that half the S&P 500 becomes worthless comes true, the resulting credit crisis and recession would hit all risk assets, including his recommended positions. This is not decoupling; it is a concentration of risk.
Furthermore, Visser ignores the disruptive potential of open-source models. Meta's Llama 3 is closing the gap with GPT-4. If open models commoditize intelligence, the economic value accrues to the infrastructure layer (hardware) but not to model providers. Even for hardware, AMD, Intel, and custom ASICs (Cerebras, Groq) are eroding NVIDIA's monopoly. The assumption that NVIDIA will maintain its pricing power and growth trajectory for five more years is optimistic, not deterministic.
Takeaway: Position for Reality, Not Narratives
As a crypto investment analyst, I see parallels to the 2021 NFT mania: a 'this time is different' narrative that ignores basic economics. The correct approach is to map real liquidity flows, identify where value is being created, and hedge against systemic risks. For AI, that means favoring companies with diversified product lines and strong balance sheets over pure-play hype names. For crypto, it means watching on-chain metrics like stablecoin supply on exchanges and transaction fees—not listening to macro pundits who cherry-pick data.
Code is law, but incentives are the reality. Visser's incentive is to generate clicks and client interest. Yours, as an investor, is to preserve capital while capturing asymmetric upside. Do not buy into the 20-30x compute explosion narrative without verifying the assumptions. Do not allocate 10-20% to a thesis that could evaporate in a single regulatory action or energy crisis. The market will eventually audit this thesis, and the ones who relied on narratives rather than fundamentals will pay the price.
Follow the liquidity, not the headlines. Volatility reveals structure. And when a strategist uses a $217 billion profit figure that is six times reality, the only proper response is to question everything else they said.
