Samsung’s profit was $217 billion. That’s the number Jordi Visser dropped in his widely circulated AI thesis—a number that doesn’t exist. Samsung’s 2024 expected profit hovers around $30-40 billion. A 7x exaggeration. But the market didn’t fact-check. It reposted, retweeted, and rehypothecated the narrative.
I’ve been here before. In 2017, I manually audited smart contracts during the ICO boom. The math didn’t add up then either. Promises of reentrancy-proof code that turned out to be Swiss cheese. Today, the same pattern repeats with AI narratives—slick storytelling with no arithmetic.
Visser’s piece, published on a Web3 news feed, claims consumer AI agents will drive 20-30x compute demand, that autonomous driving and humanoid robots will consume infinite chips, and that half the S&P 500 will lose investment value within a decade. Bold. Memorable. And built on a foundation of sand.
Let’s walk through the mechanics. The core hook is simple: AI destroys moats → traditional stocks suffer → buy Nvidia and digital assets instead. It’s a clean narrative flow that appeals to crypto-native readers who already distrust legacy finance. But when you stress-test the assumptions, the cracks appear.
First, the compute multiplier has no technical basis. Visser claims 20-30x demand relative to current levels. No model. No reference to scaling laws, inference throughput, or context length growth. In my DeFi strategy work, I never propose a yield projection without running Monte Carlo simulations on impermanent loss and gas costs. Here, we’re expected to accept a 30x number because it sounds huge. That’s not analysis—it’s marketing.
Second, the profit data fail a basic smell test. Samsung’s $217B “profit” is actually closer to its 2022 peak revenue. Profit was a fraction of that. This error alone should disqualify the piece as rigorous research. In crypto, we demand audit reports before trusting a pool with $100k. Visser asks for 10-20% portfolio allocation based on flipped decimal points.
Third, the risk architecture is orthogonal to reality. The article completely omits AI safety, regulation, energy bottlenecks, and chip supply constraints. The EU AI Act is law. China requires model approvals. A single major safety incident—like an autonomous vehicle fatality or an election-disrupting deepfake—could trigger a regulatory backlash that slashes compute demand overnight. Visser’s world has no black swans. Mine does. I learned that lesson in May 2022 when Terra’s peg broke in seconds.
Now the contrarian angle: The AI hype cycle isn’t just wrong—it’s dangerous for crypto investors precisely because it feels right. We are primed to believe in exponential disruption. DeFi summer told us liquidity mining was a free lunch. The NFT boom told us JPEGs were the new art market. Each time, the narrative outran the underlying economics. Each time, the code didn’t lie.
Visser’s thesis is a narrative, not a strategy. It borrows authority from real trends—GPU demand is strong, data centers are expanding—but then extrapolates them into infinity without modeling deceleration. He confuses early-stage adoption with terminal velocity. The same mistake that led LPs to assume Uniswap V2 yields would stay above 50% APY forever.
What’s the takeaway for a DeFi yield strategist reading this? Treat AI token plays, GPU-backed RWA protocols, and compute-derived yield products the way you’d treat a new algorithmic stablecoin. Demand audited data. Check the actual P&L of the underlying assets. Ask what happens if NVIDIA’s forward PE drops from 40x to 20x. Can the yield source survive?
In my own portfolio, I allocate capital based on stress-tested scenarios. I’ve seen what happens when a protocol’s TVL drops 40% in a week because one key assumption breaks. The Terra crash taught me that any asset lacking multiple orthogonal risk factors is a ticking bomb. Visser’s recommended 10-20% allocation to digital assets and frontier AI? That’s not diversification. That’s correlation stacking.
The real signal here is the process, not the conclusion. Visser is right that AI will disrupt industries. He’s wrong about the speed, the magnitude, and the risk-free nature of that disruption. The market will eventually correct the narrative when reality fails to match the 30x compute curve. Just as DeFi corrected when yields normalized and impermanent loss hit.
Audits don’t capture narrative risk. But they do capture code risk. In a market driven by stories, the only defense is rigorous, mechanism-level analysis. That’s what I bring to every article I write. And that’s why I’m short the hype, long the stress test.
The question isn’t whether AI will change everything. It’s whether you’re betting on the technology or on the story. One is built to last. The other will end in a liquidation event.
