The code never lies, but the analysts do.
Last week, Jordi Visser, a macro strategist with a penchant for provocative forecasts, published an investment thesis that has been circulating through blockchain and Web3 channels like a virus. His core claim: AI will trigger a 20-30x surge in compute demand, destroy half of the S&P 500 within a decade, and investors should allocate 10-20% of their portfolios to digital assets and frontier AI stocks.
I read it three times. Not because it was insightful, but because it was a masterclass in narrative construction—carefully curated data points woven into a compelling but structurally unsound argument. The kind of narrative that crypto natives should recognize instantly: the same pattern that pumped Terra, that inflated NFT floor prices, that turned DAOs into exit liquidity vehicles.
Let me be precise. The thesis is built on a single faulty assumption: that compute demand grows linearly with hype. It doesn't. Compute follows technical constraints, not marketing slides.
The Context: Visser's Historical Blind Spot
Visser positions himself as a maverick—the strategist who saw the 2000 dot-com crash coming, who warned about 2008, who now claims that 99.9% of strategists are wrong about AI. This is a classic appeal to authority, but authority without evidence is just a longer con.
His article, originally published on 22V Research, was republished by a blockchain-focused newsletter. That's the first red flag. Why would a mainstream macro piece be amplified by a crypto outlet? Because the narrative is perfectly aligned: both crypto and AI thrive on speculative demand, on promises of exponential growth, on the idea that the old rules no longer apply.
The article is not a technical analysis. It is a macro investment thesis with no code, no on-chain data, no audit trail. For someone like me, who has spent years dissecting smart contract failures and tokenomic collapses, the lack of verifiable evidence is a deal-breaker.
Core Insight: The Data Breach
Every crypto analyst knows the cardinal rule: verify every number. Tap the block explorer. Check the transaction hash. Don't trust the dashboard.
Visser violated this rule in the most egregious way. He cites Samsung's "profit" at $217 billion. The actual figure for 2024 is somewhere between $30 billion and $40 billion, depending on the quarter and the semiconductor cycle. That's a factor-of-7 error.
This is not a rounding error. This is a fundamental misunderstanding of the company that supplies the memory for Nvidia's H100s. If he cannot get Samsung's profit right, how can he estimate 20-30x compute demand?
More importantly, this data point is not peripheral to his thesis. He uses Samsung as a proxy for the entire semiconductor supply chain. The error propagates through his entire argument.
Math doesn't lie, but the inputs do.
The Rest of the Structural Flaws (Condensed for Precision)
1. Confuses Training and Inference Demand Visser treats all AI compute as a monolith. In reality, training demand is lumpy and concentrated among a few hyperscalers. Inference demand is distributed, latency-sensitive, and harder to monetize. The 20-30x claim treats both as one smooth curve, which ignores the reality that scaling laws for training are hitting diminishing returns, and inference efficiency improves with model distillation.
Based on my audit experience tracing GPU utilization across mining pools and AI compute racks, the actual inference-to-training ratio is shifting. By 2026, inference will dominate, but at lower margins and with more competition. Visser's model doesn't account for this.
2. The $2 Trillion Remaining Performance Obligations (RPO) Misread He claims the cloud providers' $2 trillion in RPO proves there is no idle capacity. This is a misreading of accounting data. RPO is unearned revenue from multi-year contracts, including plain old cloud storage, database services, and enterprise IT migration. AI compute is a fraction of that. More importantly, RPO can be canceled or renegotiated—it's not a guarantee of future compute demand. It's a liability, not an asset.
3. Ignores the Entire AI Ethics and Security Dimension This is not an oversight; it's a deliberate omission. Any serious analysis of AI's impact must include the risk of a catastrophic safety failure, regulatory lockdown, or public backlash. Visser's "AI is IQ 140 polymath" framing is dangerously anthropomorphic. AI is a stochastic parrot with no understanding of truth. A single widely-publicized hallucination-induced accident in autonomous driving could freeze deployment for years. That would crater the compute demand he predicts.
Crypto natives should understand this better than anyone. We saw how one Tornado Cash sanction reshaped DeFi. We watched Terra's collapse erase $40 billion in hours. The same tail risk exists in AI.
4. The Competitive Landscape is a Fantasy Visser names Nvidia, Marvell, Caterpillar, Modine, Eli Lilly. That's a portfolio of "AI winners" that ignores the most important battles: the race between open-source and closed-source models, the rise of custom ASICs from Google and Amazon, the potential for a new chip architecture to disrupt Nvidia's CUDA moat. He treats Nvidia's position as permanent, which is exactly what people said about Intel in 1999.
Contrarian Angle: Where the Bulls Got It Right
I do not dismiss the entire thesis. That would be intellectually lazy. Let me state clearly what Visser gets right:
- AI infrastructure spending is real. Microsoft, Amazon, Google are ramping CapEx. Data center construction is accelerating. Companies like Caterpillar and Modine will benefit, though the timing and margins are uncertain.
- Traditional software moats are eroding. The idea that Salesforce's CRM lock-in is permanent is questionable when an AI agent can replicate core workflows. Visser is early, but not wrong.
- The scale of compute required for autonomous driving and humanoid robots, if they materialize, dwarfs current AI training. That's a 5-10 year scenario, not 1-2.
But the magnitude of his predictions—50% of S&P 500 losing value, 20-30x compute demand—are based on extrapolation, not evidence. He mistakes a strong trend for a guaranteed asymptotic curve.
Takeaway: The Accountability Call
I don't care if Visser is right or wrong about AI in 2035. I care that he is presenting a speculative narrative as a fact-based investment thesis, and that this narrative is being laundered through blockchain channels to recommend digital assets.
We, as on-chain detectives and critical thinkers, have a responsibility to apply the same scrutiny to AI narratives as we do to smart contracts. Audit the data. Check the assumptions. Identify the single point of failure.
In Visser's thesis, the single point of failure is the $217 billion Samsung profit error. Once that crack appears, the entire structure fractures.
Floor prices are just consensus hallucinations. So are 20-30x compute demand projections without verifiable code.
Trust is a vulnerability with a capital T. Don't trust the strategy. Verify the data.
