Zhu Su just dropped a thread comparing AI to oil. The market barely blinked. In Toronto, I traced the silence that broke the ICO boom, and this feels eerily similar. Over the past 48 hours, Su’s take landed on my feed with the signature boldness of a man who once bet billions on Terra—and lost everything. But beneath the surface, the analogy is far more than a hot take. It is a roadmap for how we value compute, data, and the very infrastructure of the next decade. And for those of us who live in the crypto trenches, it carries a warning that the herd is not ready to hear.
Su argues that AI, like oil, will eventually commoditize—become a low-margin, high-volume, capital-intensive resource controlled by states. He points to massive energy consumption, job displacement, and the need for coordinated regulation. His thread echoes the narrative that transformed oil from a speculative frontier into the backbone of the 20th century. But Su is no mere analyst. He is the co-founder of Three Arrows Capital, the hedge fund that collapsed in 2022 after leveraging into LUNA and GBTC. Since his reappearance, he has positioned himself as a provocateur, trading on his past clout. Yet his AI-oil analogy deserves a forensic audit—not because it is wrong, but because it misses the subtle contract that binds our digital tribes.
The core of Su’s thesis is undeniable: AI compute is becoming the new crude. Training a single frontier model costs hundreds of millions of dollars in GPU time and electricity. Nvidia’s H100s are the new oil rigs. Microsoft, Google, and Amazon are the new Standard Oil—vertically integrating from chips to cloud to applications. The parallels are striking. But where Su sees commoditization, I see a deeper tension that he ignores entirely: the role of decentralized consensus in preventing the very monopoly he predicts.
Based on my audit experience during the 2017 ICO boom, I learned to spot misaligned incentives. When I analyzed 21.co's whitepaper, I found a vesting schedule that effectively gave insiders control over supply—a classic rug-pull waiting to happen. Today, the same forensic lens reveals a blind spot in Su’s analogy. AI models, unlike oil, are not purely physical. They are defined by code, data, and governance. That governance layer is where crypto’s programmable money and smart contracts offer a radical alternative. Su assumes that AI infrastructure will centralize because that’s what oil did. But he underestimates the power of tokenized incentives to align distributed compute, storage, and data curation.
How we taught the streets to read the blockchain during DeFi Summer applies here: yield farming taught us that liquidity can be bootstrapped through smart contracts. Similarly, decentralized compute networks like Filecoin and Render are already commoditizing GPU time, but they do it through permissionless markets, not state-backed monopolies. The key difference is that oil is fungible and non-programmable; AI compute is programmable through smart contracts. This opens the door to competition that Su dismisses.
The contrarian angle that Su—and most of the market—misses is that the AI-oil analogy breaks down at the point of differentiation. Oil has no network effect beyond logistics. Once you have a barrel of Brent, it is identical to a barrel of WTI. But AI models are never identical. The value of a model lies in its fine-tuning on proprietary data, its alignment with user preferences, and its integration into ecosystems. This is exactly where decentralized protocols can shine. Imagine a DAO that governs a model’s training data, rewarding users for contributing high-quality examples. That model cannot be commoditized because its data and governance are unique. The invisible contract binding our digital tribes is trust—not compute. And trust cannot be drilled from the ground.
From an investment perspective, Su’s analogy suggests that AI companies will eventually trade like oil majors—low margins, high capex, dependent on regulation. But current valuations of OpenAI and Anthropic imply they are high-growth software companies. This mismatch is either a bubble or a bet that commoditization will not happen for decades. In crypto, we have seen this before: projects like Chainlink and Uniswap were labeled as commodities by critics, yet their tokenized ownership and network effects created moats that defied commoditization. The lesson is that programmability transforms commodities into ecosystems.
Leading the herd through the volatility fog requires looking beyond Su’s headline. The real signal is not that AI will become oil—it is that the current AI landscape lacks the decentralized consensus layer that made crypto resilient. Without that layer, Su’s prediction becomes a self-fulfilling prophecy: compute will centralize, and the few will control the many. But the crypto community has the tools to build an alternative—if we act before the silence spreads.
The takeaway is not to sell your AI tokens or to buy oil futures. It is to watch the infrastructure layer—not the models, but the markets for compute, data, and training. Catching the signal before the market blinks means recognizing that Su’s analogy is a double-edged sword: it warns of centralization, but it also highlights the massive opportunity for decentralized alternatives. The question is whether we have the courage to build them before the state-backed oilmen arrive.