Over the past 30 days, the top AI-token protocols have shed 45% of their on-chain value locked. The narrative of 'decentralized compute for the AI future' is bleeding. Investors are waking up to a harsh reality: AI is not a novel resource to be tokenized—it’s a commodity waiting to happen. This isn't my opinion; it's the structural signal buried in Zhu Su's recent AI-oil analogy.
Zhu Su, co-founder of Three Arrows Capital, drew a direct line between the evolution of oil and that of artificial intelligence. His core thesis: AI will follow oil’s path from a high-margin, differentiated resource to a low-margin, capital-intensive commodity. The analogy is crude—pun intended—but it carries a devastating logic for the crypto industry’s AI ambitions. If AI becomes oil, then the race is about scale, not innovation. And crypto protocols, by design, resist centralized scale.
Let me break down the context. Zhu Su argues that AI requires massive, state-backed capital—think NVIDIA’s GPU clusters, not your home miner. He points to the externalities: energy consumption, job displacement, and systemic risk. The endgame? AI services will be priced by cost, not by model quality. The winners will be those who control the cheapest compute and largest distribution, not the best model. This is a direct threat to the current crypto-AI playbook, which sells tokenized access to 'exclusive' AI power. If the power becomes a commodity, where is the scarcity premium?
This is where my own forensic work kicks in. Over the past three years, I have audited over a dozen decentralized compute protocols—Akash, Render, io.net, and others. I’ve stress-tested their token economics, their settlement layers, and their incentive alignment. The common flaw: they treat AI compute as a finite, tradeable asset. But Zhu Su’s analogy reveals a deeper truth. Compute is not oil; it is a service whose marginal cost trends to zero. In an audit I conducted for a leading GPU-sharing network, I discovered that their token price was entirely decoupled from the actual utilization rate of the hardware. The market was pricing in a hope of AI scarcity that the protocol’s own architecture could never enforce. Logic holds until the ledger bleeds.
The core innovation of my analysis is this: the oil analogy actually masks a more dangerous pattern for crypto. Oil commoditization happened slowly, over decades, because extraction and refining are physically constrained. AI commoditization could happen in years, driven by open-weight models, hardware standardization, and algorithmic efficiency. We are already seeing the harbingers. Meta released Llama 3.1 for free. Mistral offers open models rivaling GPT-4. The price per million tokens on OpenAI’s API has dropped over 80% in two years. If AI service costs continue to halve every twelve months, the entire value proposition of ‘buying AI compute now to earn later’ collapses. Trust is a variable, not a constant. In crypto, we trusted that compute scarcity would endure. The market data suggests otherwise.
Here is the contrarian twist: the oil analogy fails precisely where it tries to be predictive. AI is not oil. Oil is a molecule with fixed energy content. AI is a function whose capabilities can expand without bound. The commodity view assumes AI capabilities plateau—that we stop making breakthroughs. Yet the industry continues to push toward AGI, multimodal reasoning, and autonomous agents. If capabilities keep improving, models are not interchangeable commodities. A GPT-5-level model will still command a premium over a fine-tuned Llama 3.1, just as a Ferrari commands a premium over a Toyota. Crypto’s blind spot is not in overestimating commoditization; it is in underestimating the possibility of a permanent high-value tier of AI. Protocols that focus solely on commoditized inference will miss the real value: trustworthy, verifiable AI execution for high-stakes decisions. Decentralization is a promise, not a guarantee.
To be clear, I am not dismissing the commodity thesis. I am mapping its limits. The most likely outcome is a two-tier market: a low-cost, generic AI layer (the oil) and a premium, secure, and verifiable AI layer (the specialty chemical). Crypto can own the second. By leveraging zero-knowledge proofs, on-chain attestations, and decentralized governance, protocols can offer AI services that are auditable and resistant to model tampering—something a centralized oil company cannot provide. This is where my own research on AI-agent smart contract orchestration comes in. I have architected systems where AI decisions must be validated by a decentralized set of verifiers. The cost is higher, but the trust is provable.
Takeaway: The AI-oil analogy is a useful warning, not a roadmap. If you treat AI as a commodity, you will compete on price—and lose to centralized giants who can scale cheaper. If you treat AI as a trust problem, you unlock a market that no commodity player can address. The future of crypto and AI is not in tokenized compute. It is in tokenized truth. The question is: can the crypto industry pivot fast enough, or will it bleed out clinging to a flawed analogy?
