The chart whispers, but the volume screams. Over the past 72 hours, a flood of institutional research has hit my terminal — all singing the same tune: AI will trigger a 20-30x explosion in compute demand, half the S&P 500 will die, and you better load up on Nvidia, Marvell, and digital assets. Jordi Visser’s note, circulated through Web3 channels, is the loudest yet. It’s seductive. It’s urgent. And if you take it at face value, you’ll miss the real signal buried in the noise.
Let me be blunt: I’ve seen this script before. In 2017, it was ICOs promising to disrupt everything. In 2021, it was NFTs as the future of ownership. Now it’s AI as the unstoppable force that will vaporize traditional businesses and pump crypto alongside. The narrative is compelling, but the technical and market reality is far messier. As a real-time trading signal strategist who’s spent years bridging retail speed with institutional scale, I’ve learned that speed alone isn’t enough — you need to filter the signal from the hype.
Context: Why This Narrative Is Flooding Crypto
Visser’s thesis, as parsed by industry analysts, rests on three beams: (1) consumer AI agents and autonomous driving will require 20–30x current compute, (2) traditional corporate moats will collapse in 5–10 years, making most equities irrelevant, and (3) the only hedge is to overweight “frontier AI stocks” and digital assets. The piece is being passed around Telegram groups and crypto newsletters because it validates a shared bias — that the old world is dying and crypto is the escape hatch.
But here’s the problem: Visser is a macro strategist, not an AI engineer. His 20–30x compute multiplier has no technical basis — no scaling law derivation, no inference throughput model, no consideration of algorithm efficiency gains. It’s an emotional extrapolation dressed in numbers. And when you dig into the data, the cracks widen.
Take the claim that Samsung’s 2024 profit will hit $217 billion. That’s not a typo — it’s a distortion of reality. Samsung’s actual expected profit is around $30–40 billion, depending on the semiconductor cycle. A 500% error in a core data point should make you question everything else.
Core: What the Compute Demand Thesis Misses
Speed is the only hedge in a real-time world. I built my career on velocity-first analysis, but velocity without accuracy is just noise. Let’s run the numbers on Visser’s compute explosion from the perspective of someone who models liquidity flows and hardware bottlenecks daily.
Training vs. Inference: The false equivalence. Visser conflates all compute demand into one lump. In reality, training demand grows when new frontier models are released — but each successive model faces diminishing returns. The era of scaling laws may be plateauing; even OpenAI has acknowledged that bigger models aren’t necessarily smarter. Inference demand, on the other hand, scales with user adoption. Yes, consumer AI agents could drive inference up, but that increase depends on real product-market fit, not just speculation. We don’t yet have a single widely-adopted consumer AI agent that can handle dynamic workflows. The technical hurdles — long-term memory, robust multi-step reasoning, low latency — are far from solved.
The 2 trillion dollar RPO mirage. Visser points to cloud providers’ $2 trillion in remaining performance obligations (RPO) as proof of insatiable AI demand. But RPO includes years of non-AI cloud services — storage, databases, legacy workloads. Attributing it all to AI is like looking at a restaurant’s total bookings and assuming every table is ordering the most expensive steak. Worse, RPO is a forward-looking contract, not guaranteed consumption. Customers can cancel or delay. During the 2022 cloud spending slowdown, we saw exactly that — RPO growth decelerated as firms optimized costs.
Hardware supply is the real governor. Even if AI workload demand does double every year, the physical capacity to produce advanced chips is limited. TSMC’s CoWoS packaging capacity is already maxed out through 2025. HBM memory is constrained. Building a new fab takes 3–5 years. Visser’s 20–30x implies we’ll somehow bypass these physics. We won’t.
Liquidity flows where fear turns into opportunity. And right now, fear is being manufactured to push capital into a narrow set of assets. The smart money isn’t chasing Nvidia at 50x forward earnings — it’s looking at the bottlenecks that actually exist: power infrastructure, cooling, networking. That’s where Caterpillar and Modine come in, but even there, you need to watch for overcapacity.
Contrarian: The Real Threat Isn’t Compute — It’s Narrative Inflation
Visser warns that half the S&P 500 will lose investment value in 5–10 years. That’s the kind of headline that gets clicks, but it ignores how moats actually work. Yes, AI erodes cost advantages. But regulatory moats (banks, healthcare), network effects (Microsoft Office, Visa), and high switching costs (enterprise ERPs) don’t disappear overnight. The internet didn’t kill General Electric in 5 years; it took 20, and even then, GE’s downfall came from financial engineering, not technology.
We didn’t see the last bear market coming until it was too late. In 2022, the same “AI will save us” narrative collapsed when interest rates rose. If Visser’s thesis is wrong — if AI compute demand grows at 5–10x instead of 20–30x, or if a regulatory shock hits (EU AI Act, export controls, safety incident) — the stocks he recommends could halve. And the crypto market, which he frames as a hedge, is highly correlated to tech sentiment. Bitcoin’s post-ETF approval performance shows it now trades like a risk-on tech proxy, not digital gold.
My opinion on Bitcoin: Satoshi’s vision is dead. The ETF made Bitcoin Wall Street’s toy. It’s no longer peer-to-peer cash; it’s a macro asset driven by institutional flows. That doesn’t mean it can’t go up, but it means the “hedge against AI disruption” argument is weak. If AI stocks crash, Bitcoin crashes with them.
Stablecoin yield products: another maturity mismatch. Visser doesn’t mention them, but the crypto sector is full of products like sUSDe that promise high yields from funding rates and basis trades. These work in bull markets, but as we saw with Terra, they blow up first when liquidity dries up. The AI narrative is pumping capital into these products now, creating a fragility that will snap in the next downturn.
Takeaway: What to Watch Next
The chart whispers, but the volume screams. Right now, the volume is screaming “buy the hype,” but the whisper says “watch the bottlenecks.” Over the next 6 months, I’m tracking three signals:
- NVDA’s next earnings — If data center revenue guidance disappoints, the 20–30x narrative dies instantly.
- Cloud provider CapEx commentary — Look for any mention of “efficiency” or “optimization” from AWS, Azure, GCP. That’s the first sign of a pullback.
- Crypto correlation — If Bitcoin fails to decouple from tech stocks during a selloff, the “hedge” thesis is dead.
Speed is the only hedge in a real-time world. But speed without skepticism is just a faster way to lose money. Don’t let the narrative run you over. Position for the reality, not the dream.