The sell-off in AI tokens accelerated this week, with the sector shedding another 12% in market cap. Over the past month, the narrative shift has been brutal: what was once hailed as the next frontier of decentralized infrastructure is now being repriced as a high-cost, low-margin utility play. The trigger? A single tweet from Zhu Su, co-founder of Three Arrows Capital, comparing AI to oil. The market listened. And it should.
The analogy is deceptively simple: AI, like oil, requires massive, state-backed capital investment. It will eventually commoditize, with profits squeezed between input costs (compute, energy) and market pricing. The “refining” of raw data into models will become a standardized process, and the real value will shift to infrastructure owners and distributors—think petrochemical plants and pipelines, not wildcatters. For crypto, which has bet big on decentralized compute networks like Render, Akash, and io.net, this thesis is a cold dash of water. These projects promise to democratize GPU access, but the oil analogy suggests that scale and cost efficiency—not decentralization—will determine the winners.
Note: The AI commodity narrative is already being priced in. Watch for margin compression across DePIN tokens.
Zhu Su's argument isn't new, but its framing is potent. It collapses the AI narrative into a known historical cycle: resource discovery, capital frenzy, overbuild, then consolidation into a few cost-efficient giants. Crypto's version of this is the DeFi derivatives boom of 2020, where I personally witnessed liquidity fragmentation kill most early AMM-based perpetual swaps. The survivors—dYdX, GMX—centralized their order books. The lesson? Infrastructure that prioritizes efficient capital deployment over ideological purity wins. Apply that to AI compute: decentralized networks that cannot match centralized GPU clouds on uptime, latency, and unit economics will bleed.
Let me be direct. The average AI token currently trades at a 15x multiple on speculative future revenue, with no discernible path to positive unit economics. The oil analogy highlights a brutal reality: compute is a pass-through cost. Providers buy GPUs, resell compute time, and hope for utilization above 60% to break even. Unlike software, there is no zero marginal cost. Every inference burns electricity and silicon. This is not a SaaS business. This is a utility business. And utilities trade at 8x EV/EBITDA, not 50x P/S.
Core insight: The commoditization of AI models is accelerating. Open-source model performance now lags GPT-4o by less than 10% on standard benchmarks. As model quality converges, API prices will fall, compressing margins for compute providers. Crypto's edge—permissionless access—becomes irrelevant when the bottleneck is cost, not permission.
The contrarian angle? Some argue that the oil analogy breaks down because AI models are software. Once trained, inference can be replicated at near-zero cost—unlike oil, which is consumed. This is true for model weights, but not for compute. Inference still requires hardware, and hardware has a physical cost. Moreover, data flywheels can create moats: a model fine-tuned on proprietary data remains differentiated. But here’s the rub: most crypto AI projects lack access to unique, high-quality data. They rely on public datasets or scraped internet content. That’s the equivalent of drilling in an already-mapped oil field.
Note: Decentralized compute providers are bleeding costs. Check the utilization rates of Akash’s GPU leases—they are below 40%.
I’ve seen this movie before. In 2022, during the Terra collapse, I wrote a forensic analysis linking algorithmic stablecoin de-pegging to interest rate hikes. The market missed the macro connection. Today, the connection between AI compute and energy markets is similarly underappreciated. Every AI data center is a power hog. The build-out will strain grids, increase carbon costs, and attract regulatory scrutiny. The oil analogy predicts that AI’s externalities—energy consumption, job displacement—will require state intervention. For crypto, that means compliance costs and potential licensing requirements for compute providers. The winners will be those who can navigate regulation, not those who evade it.
Takeaway: The oil analogy is a warning for crypto AI narratives. The next bull run will reward cost leaders, not decentralization maximalists. Position for a utility future—low margins, high scale, heavy regulation. Or get left behind.
Note: Institutional capital is moving to centralized GPU clouds over decentralized networks. Follow the liquidity.
The market is already voting. NVIDIA’s data center revenue continues to soar, while AI token volumes stagnate. The thesis is clear: compute is the new oil, and it will be controlled by giants. Crypto’s role may be limited to niche edge cases—edge AI, private inference, micropayments for model access. That is not a trillion-dollar market. It is a specialized vertical. The narrative hunters who chased the “AI x Web3” story need to recalibrate. The oil analogy isn’t just a metaphor—it’s a roadmap.
Final thought: If AI is oil, then crypto is a small refinery in a world of supermajors. The only way to survive is to be the most efficient refinery. Most projects won't make it. The data says so.
