The whisper came from a familiar corner of the crypto graveyard. Zhu Su, co-founder of Three Arrows Capital, recently posted that oil might be the best analogy for AI — predicting an eventual commoditization of large language models, capital-intensive infrastructure, and state-backed consolidation. For those of us who spent 2017 auditing ICO whitepapers in windowless rooms, the word "commoditization" carries a specific weight. It’s the echo of a promise that never materialized — until now. But as an open source evangelist who has watched both crypto and AI evolve from the fringe to the mainstream, I sense something deeper: the oil analogy is dangerously incomplete. It captures the surface currents of capital and computation, but it misses the undercurrent of protocol design that could change the entire flow.
Context: The Analogy and Its Blind Spots
Zhu Su’s argument is straightforward. AI, like oil, requires massive upfront capital (compute clusters, data pipelines, talent). Over time, the technology becomes standardized, margins compress, and the value migrates upstream to infrastructure providers (GPUs, energy) and downstream to application layers. The externalities — energy consumption, job displacement, systemic risk — become societal burdens, much like carbon emissions from oil. He points to state-backed investments in AI as evidence of its strategic importance, paralleling how nations once fought for oil fields.
On the surface, it’s a coherent narrative. But I’ve lived through the ICO boom and the DeFi summer, and I’ve seen how the same "commoditization" logic led to the collapse of projects that thought liquidity mining was a durable moat. The oil analogy treats AI as a homogeneous resource, but code is not crude. A model can be forked, fine-tuned, and distributed at near-zero marginal cost. An oil well cannot. This difference is not semantic — it’s structural.
Core: Protocols Over Pipelines
The real insight from the oil analogy is not the commoditization outcome; it’s the infrastructure layer. In the oil industry, control over pipelines, refineries, and distribution networks determines profitability far more than ownership of crude reserves. In AI, the equivalent infrastructure is not just GPUs — it’s the coordination layer that governs who can access compute, how models are verified, and whether the value flows to a central entity or to a network of participants.
This is where blockchain enters. During my time contributing to the Veritas framework — an open source protocol for verifying AI-generated content on-chain — I saw how the union of cryptography and AI could create a new class of infrastructure: a trust-minimized pipeline for model inference, training contributions, and data provenance. The oil analogy assumes a centralized refinery, but smart contracts can act as a decentralized refinery, allowing anyone to stake compute, serve inference, and earn tokens proportional to their contribution. The result is not commoditization in the traditional sense, but a market where the resource (model capability) remains diverse because the incentive structure rewards differentiation, not homogenization.
Consider the current reality: every week, a new open source model emerges — Llama, Mistral, Gemma — each with architectural trade-offs. If AI becomes a commodity, these models would converge to a single optimized design. Instead, we see fragmentation, because the protocol layer (Hugging Face, ONNX, TEE-based verification) enables them to coexist. This is exactly what happened in cryptocurrency: Bitcoin became digital gold, Ethereum became a settlement layer, and thousands of chains found niches. The same pattern will repeat in AI, provided we build the right protocols.
Based on my experience auditing the Ethera DAO in 2017, I learned that centralization is often hidden in the governance token distribution. The same lesson applies to AI: the true bottleneck is not model performance but the control of data and compute allocation. If a single entity like OpenAI owns both the model and the API gateway, it can extract rent indefinitely, just as a pipeline owner charges tolls. But if the gateway is a smart contract — auditable, open, and governed by a community — the rent dissipates. The oil analogy fails to account for this architectural shift.
Contrarian: The Real Risk Is Not Commoditization — It’s Capture
Counter-intuitively, Zhu Su’s commoditization thesis may be too optimistic. Commoditization implies that competition will drive prices to cost, benefiting consumers. But AI infrastructure exhibits strong network effects and economies of scale that mirror the oil industry’s worst tendencies: natural monopoly. The cost to train a frontier model has already surpassed $100 million, and inference costs are dropping much slower than expected due to hardware shortages. In such an environment, the "commodity" might not be the model but the right to access the compute that runs it.
We’ve seen this movie before. In Ethereum’s early days, gas prices were a pain point, but they were visible and market-driven. Today, rollup fragmentation has created a hidden monopoly of sequencers — single entities that order transactions and extract MEV. The same dynamic could emerge in AI inference: a few centralized providers (Azure, AWS, GCP) control the majority of GPU inventory, and they can set prices arbitrarily. The oil analogy would call this "commoditization of the resource," but it’s really capture of the distribution layer.
The antidote is not less capital but better protocol design. Projects like Bittensor are attempting to create a peer-to-peer network for machine intelligence, where miners train models and validators judge their outputs. Others, like Akash Network, are building decentralized compute marketplaces. The challenge is not technological — it’s alignment. How do we design incentives so that the network rewards quality rather than grinding for tokens? In my work with the Soulbound Narratives community, I saw how small, trusted groups could produce better content than open, anonymous pools. Scaling that to AI training requires a careful balance of reputation, commitment, and cryptographic verification.
Takeaway: The Void Between Tokens Holds the True Value
Zhu Su’s oil analogy is a useful warning, not a prescription. It reminds us that every technology, no matter how transformative, can become a rent-extraction vehicle if its protocol layer is neglected. The value in AI — as in crypto — will not be in the models themselves but in the protocols that govern their creation, access, and verification. Open source is not a license; it is a covenant. And that covenant must be written in code that anyone can audit, fork, and improve.
Silence in the ledger speaks louder than code. If we listen, we will hear the faint hum of a thousand small models, each serving a niche community, verified on-chain, and owned by no one. That is the forest that will follow. The oil rigs of AI will eventually rust, but the protocols — if we nurture them — will last.
Growth without belonging is just noise. Let us belong to the network, not the refinery.