Zhu Su, co-founder of the infamous Three Arrows Capital, dropped a rhetorical bomb last week that ricocheted through both AI and crypto Twitter: oil is the best analogy for AI, and its inevitable endgame is commoditization. The post was short, the implication deep — and for those of us who live in the intersection of narrative mechanics and capital flows, it wasn't just a macro speculation. It was a signal. A whisper about where the next cycle's liquidity will hunt.
I’ve spent the last sixteen years watching narrative carcasses rot in the bull market sun. I’ve seen ICOs sell dreams without code, DeFi protocols collapse under their own governance fat, and NFT communities turn into ghost towns overnight. Tokens are receipts; memes are the religion. But every cycle, the underlying asset class — the narrative vessel — changes. In 2017 it was “utility tokens.” In 2020 it was “DeFi Lego.” In 2021 it was “NFT art.” Now, in 2024, the hottest narrative is “AI + Crypto.” And Zhu Su just handed us a framing that could either legitimize the hype or mark its peak.
Context: Three Arrows, Zhu Su, and the Commodity Playbook
Zhu Su is not an AI researcher. He’s a creature of financialized narratives — a man who rode the crypto supercycle on leverage and shadowy debt. His firm, Three Arrows Capital, famously blew up in 2022, leaving a crater of contagion. But survivors often see the battlefield with sharper eyes. His oil analogy isn’t technical; it’s structural. He argues that AI, like oil, will require massive state-backed capital, create wide societal benefits, generate negative externalities (energy consumption, job displacement), and ultimately become a low-margin commodity.
Why should a crypto native care? Because the same framing is currently being applied to AI tokens — Render (RNDR), Akash (AKT), Bittensor (TAO), and a dozen others. These tokens are being marketed as “the oil of the AI revolution.” The narrative: own the compute, own the future. Chaos is the alpha, but coherence is the asset. If Zhu Su is right about AI commoditization, then the current premium on AI tokens might be a mirage. But if he’s wrong — if AI remains a high-differentiation, winner-takes-most market — then the best is yet to come.
Core: The Narrative Mechanism of Commoditization
Let’s dissect the analogy’s mechanics. Oil commoditization happened because crude oil from different sources became interchangeable after refining. The transport network (pipelines, tankers) and downstream products (gasoline, plastics) captured most of the value. Upstream extraction became a race to the bottom based on cost efficiency.
Apply this to AI. If base models (GPT-4, Claude, Llama) become as interchangeable as crude — if fine-tuning and application layers become the real value — then owning GPU compute is like owning a well. The profit margin collapses. We didn’t find a coin; we found a consensus. The consensus here is that AI tokens are pricing in a scarcity that might not exist in five years.
But here’s the catch: crypto markets don’t trade on five-year fundamentals. They trade on narrative velocity. And right now, AI tokens have the highest narrative velocity since the 2021 NFT mania. Look at TAO’s market cap: it surged past $3 billion in Q2 2024 on the promise of a decentralized network for AI model training. But token holders aren’t buying compute; they’re buying a story — that decentralized AI will outcompete centralized giants.
From my experience managing a $50 million token fund, I’ve seen this pattern before. During the ICO boom, I was on the other side: I launched a fraudulent token project that raised $40k purely on narrative. I learned that tokens are receipts; memes are the religion. The religion of AI tokens is powerful because it taps into both the tech utopian dream and the fear of missing the next internet. The fact that Zhu Su, a notorious macro gambler, is now applying oil logic to AI is a bearish signal for those holding AI tokens purely on hype. It suggests the smartest money is already thinking about exit liquidity.
Contrarian: Why AI Is Not Oil — And Why That Makes AI Tokens Even More Dangerous
The most obvious counter-argument: AI is software, not a physical resource. Software has near-zero marginal cost to reproduce. Once a model is trained, running inference costs pennies per query. Oil, on the other hand, must be extracted, transported, and refined — each step adds cost. AI’s commoditization is therefore an even more extreme race to the bottom. The winner might be the company with the cheapest inference, like Google or Meta, not a decentralized compute network.
But wait — there’s a deeper contrarian layer. What if AI commoditization doesn’t happen because the bottleneck shifts from compute to data? Proprietary data (medical records, legal documents, user behavior) is not interchangeable. And data is what starts moats. In that world, AI tokens that own data marketplaces (e.g., Filecoin, Arweave, Ocean Protocol) could become the real oil — not as a commodity, but as a differentiated resource. This is where my contrarian instinct kicks in. Chaos is the alpha, but coherence is the asset. The coherence requires us to separate the narrative from the fundamentals.
I recall advising a hedge fund in 2024 on integrating crypto. One of the hardest sells was explaining that Bitcoin’s value doesn’t come from its “digital gold” narrative alone — it’s reinforced by institutional infrastructure (ETFs, custody, derivatives). Similarly, AI tokens need real-world adoption beyond speculation. Render is being used for actual rendering work by VFX studios. Akash hosts machine learning workloads. TAO has a functioning subnet ecosystem. But the revenue of these networks is a fraction of their market caps. The narrative premium is enormous.
Takeaway: The Next Commodity Might Be Attention, Not AI
Zhu Su’s analogy is a useful mental model, but its real value is in forcing us to ask: what becomes the scarcest resource in a world of commoditized intelligence? My bet is attention — curation, trust, and community. The same dynamics that made crypto communities (memecoins, NFT clans) outlast DeFi protocols are the ones that will persist. We didn’t find a coin; we found a consensus. The consensus is that AI tokens are overpriced in the short term but structurally important in the long term. The smart move: track narrative fatigue. When the AI hype cycle peaks — as all hype cycles do — the liquidity will rotate back to foundational assets (BTC, ETH) or to the next new story (DePIN? RWA?).
In my newsletter, I often sign off with: “Tokens are receipts; memes are the religion. Chaos is the alpha, but coherence is the asset. We didn’t find a coin; we found a consensus.” Zhu Su just gave us a receipt. Now we decide if the religion spreads or burns.