算力期货的金融化
人工智能算力正在经历金融化的过程，GPU小时已经在现货市场交易，并被转化为数十亿美元的债务工具，同时预售容量也已成为常态。
多家初创公司（如Ornn、Architect Financial、OneChronos、Compute Exchange）正在构建期货交易所和定价指数，并有项目（如GAIB、USD.AI）致力于将GPU能力上链。
文章指出，由于基础设施紧张、成本飙升以及对风险对冲的需求，算力衍生品市场正在形成。
尽管面临GPU贬值、供应集中和标准化缺失等挑战，但现已出现GPU资产支持债务、期货交易所和衍生品市场等现象，预示着算力金融化正在萌芽阶段。
查看原文开头（英文 · 仅前 3 段）
Executive Summary
The financialization of AI compute is no longer a theoretical prospect. GPU‑hours are already being traded in spot markets, securitized into multi‑billion‑dollar debt instruments, and pre‑sold through capacity reservations. Simultaneously, dedicated startups (Ornn, Architect Financial, OneChronos, Compute Exchange) are building futures exchanges and pricing indices, while tokenization projects (GAIB, USD.AI) aim to bring GPU capacity on‑chain. Government policy, specifically the Trump administration's AI Action Plan, has explicitly endorsed a "healthy financial market for compute" [14]. At the same time, the infrastructure that underlies compute is under strain: hyperscaler CapEx has reached $213 billion in 2024 and is forecast to rise to as much as $1.3 trillion per year by 2032 [32], energy bottlenecks are driving price spikes, and inference spending has overtaken training as the dominant operational cost [5], [6]. This environment creates the raw material for derivatives: scarcity, price volatility, and a large pool of participants who need to hedge.
Yet, the path toward a mature compute‑derivatives ecosystem is far from certain. GPUs depreciate over an economic life of only 3–5 years, far shorter than traditional financeable assets [8]. The supply chain is highly concentrated—NVIDIA holds over 80% of cloud accelerator instances [44]—and vertical integration by large AI firms threatens to shrink the addressable spot market [11]. Standardization of a contract unit remains unsolved, and legal recognition for tokenized GPU title is absent in major jurisdictions [28]. The evidence assembled in this report shows a market in its earliest, fragile stages, with both enormous potential and profound structural risks.
※ 出于版权考虑，仅引用前 3 段。完整内容请阅读原文。