Over the past 30 days, the global AI inference compute market has burned through enough electricity to power a small nation. Meanwhile, the price of a single GPT-4o API call has dropped by 38% since January. The chart didn't lie — the unit economics of intelligence are trending toward zero. And if you've been watching the blockchain, you've seen the same pattern play out in crypto mining, L2 gas wars, and the race to zero fees. Zhu Su, the former Three Arrows Capital co-founder known for his anti-fragile macro calls, recently dropped an analogy that cuts through the noise: "Oil may be the best analogy for AI, ultimately leading to commoditization." I've spent the last week chasing the ghost in the smart contract code of both industries — and the data tells a story that most AI bulls don't want to hear.
Zhu Su is no stranger to controversial takes. After 3AC's collapse in 2022, he re-emerged as a crypto commentator with a macro lens. His latest piece, published on a Web3 news outlet, draws a direct parallel between the oil industry's evolution — from wildcat drilling to OPEC-controlled commodity — and the trajectory of artificial intelligence. The core thesis: AI models, like crude oil, will eventually become a standardized, low-margin commodity requiring massive, often state-backed capital expenditure. The value will accrue not to the drillers but to the refiners, distributors, and infrastructure providers. This is not a new idea in traditional finance — banks have been warning about AI's capex burden for months. But Zhu Su frames it through the lens of a crypto-native who understands what happens when a nascent technology gets financialized.
Let's break down the analogy with on-chain data and market evidence.
Commodityization is already happening in AI inference. The API price decline is accelerating. According to data from Artificial Analysis, the cost per million tokens for top-tier models has fallen from $20 in early 2023 to under $3 today. That's an 85% drop in 18 months. The chart didn't lie — and it mirrors the trajectory of Bitcoin mining hashprice, which fell 75% from its 2021 peak to the 2023 trough. Both industries are experiencing the same dynamic: technological improvement plus competition equals margin compression. In crypto, we call this the "deflationary sink" of compute resources. In AI, it's called commoditization. Based on my experience manually executing flash loan arbitrage on Uniswap V2 in 2020, I learned that marginal cost advantages compound geometrically when liquidity is deep. The same principle applies here: as AI inference becomes cheaper, more use cases emerge, sucking in even more capital to build even cheaper compute. It's a cycle that ends only when the cost of intelligence approaches the cost of electricity.
Capital intensity is spiking at levels that rival oil majors. Microsoft, Google, and Amazon have committed over $150 billion in AI infrastructure capex through 2026. That's on par with the annual capex of ExxonMobil and Shell combined. The parallel with crypto mining is eerie. In 2021, publicly traded miners like Marathon and Riot announced $3 billion in mining rig purchases — only to see asset values crater in 2022. Based on my audit experience analyzing balance sheets for DeFi protocols, I can tell you that the same "capex death spiral" pattern emerges when firms overinvest in hardware without securing long-term offtake agreements. For AI, the offtake is API usage. For crypto mining, it was hashprice forward contracts. Both are vulnerable to demand shocks. If a recession hits AI adoption growth, those data centers become stranded assets — just like the Bitcoin ASICs rusting in warehouses after the 2022 crash.
State backing is real and will reshape the competitive landscape. The U.S. CHIPS Act, the EU's AI Act, and China's massive GPU stockpiling efforts mirror the national oil company model. Follow the scholar, not the token — if you want to understand where AI value will flow, track government subsidies and export controls, not model benchmarks. The Biden administration's export curbs on NVIDIA H100 chips to China are the AI equivalent of the 1973 Arab oil embargo. I saw this dynamic play out in my 2024 work analyzing spot Bitcoin ETF flows: when institutions entered, the price narrative shifted from retail hype to regulatory positioning. The same is happening now with AI. Nations are not just funding AI companies; they are securing the supply chain for compute, energy, and talent. This geopolitical overlay adds a layer of risk that pure commodity investors haven't fully priced.
But here's where the analogy gets interesting for crypto. The crypto-AI intersection — decentralized compute networks, tokenized GPU resources, zero-knowledge proofs for model verification — exists precisely because of this commoditization pressure. If AI is becoming oil, then projects like io.net, Render Network, and Akash are building the AI equivalent of oil pipelines and refineries. They're aiming to be the midstream, not the upstream.
Volatility is just liquidity with a pulse. Crypto AI tokens have been some of the most volatile assets in the 2024-2025 market, with RNDR swinging 20% in a single week. But beneath the surface, the nest was empty for many of these projects — total value locked on decentralized compute networks remains below $500 million, a rounding error compared to centralized cloud spend. Yet the narrative is compelling: if AI is becoming a commodity, then the infrastructure that intermediates its consumption should capture value. That's the thesis behind io.net's token model, which rewards GPU providers in a liquidity pool-like structure. I've scanned the block for the missing brick in this thesis, and it comes down to demand side commitment. Without guaranteed offtake from enterprise AI clients, these networks risk becoming ghost towns — like the hundreds of dead L2 chains we saw in 2023.
Contrarian: The oil analogy has three critical blind spots that every crypto investor should exploit.
First, oil is a physical commodity with storage costs, transportation constraints, and geopolitical concentration. AI is digital — it can be replicated at zero marginal cost, open-sourced, and distributed globally in seconds. The commoditization of AI may happen much faster than oil's century-long journey, and it may not lead to the same OPEC-like cartelization. Open-source models like Llama 3 and Mistral are already eating the proprietary API market. The chart didn't lie: open-source model performance is closing the gap faster than anyone expected. In crypto, we saw this with the rise of L2 solutions commoditizing L1 execution. The lesson: don't bet on the drill when the pipeline can carry any barrel.
Second, the analogy ignores the potential for AI to become a public good rather than a commodity. Crypto's ethos — permissionless access, user-owned networks — directly opposes the oil industry's extractive, centralized model. If decentralized AI networks achieve scale, they could flip the script: instead of a few state-backed giants controlling AI, we could have a global, community-owned compute commons. That's the thesis behind projects like Bittensor and Gensyn. But it's also a huge stretch. My 2021 deep dive into Axie Infinity's "scholar" exploitation taught me that when economic incentives are misaligned, even beautiful decentralized systems become extractive feudal structures. Bittensor's token model creates a similar dynamic: subnet owners can extract rents from miners, and without proper governance, the network centralizes around whales.
Third, and most dangerously for investors, the oil analogy implies that AI companies will eventually trade like Exxon — low P/E, high dividend, slow growth. But current valuations for OpenAI, Anthropic, and even crypto-AI tokens like RNDR and TAO are pricing in software-like growth multiples. If the commodity thesis is correct, these valuations are due for a massive correction. Based on my experience covering the 2022 Terra/Luna collapse — where I was the first to publish on-chain data confirming the depeg — I can tell you that narratives can flip faster than a flash loan attack when the data contradicts the hype. I've already seen warning signs: OpenAI's revenue growth is decelerating, Anthropic is burning through cash, and crypto AI token volumes are heavily concentrated in a few addresses. Speed eats stability for breakfast, but in this case, the speed of commodityization may eat the valuation whole.
Takeaway: So where does this leave us? As a crypto news editor who has watched the industry cycle through hype and crash, I see the oil analogy as a useful stress test for every AI-related token. Ask yourself: is this project building a better drill (model) or a better pipeline (infrastructure)? The former will get commoditized; the latter may survive. But even infrastructure isn't safe — it needs moats like network effects, regulatory capture, or energy-cost advantages. The ultimate question for 2025 is not whether AI will become a commodity — but whether crypto can build the infrastructure that prevents that commodity from being controlled by the same old nation-states. I'll be scanning the block for the missing brick in that answer. Follow the scholar, not the token.