在2026年第一季度，全球企业在云服务上花费了1290亿美元，而美国五个最大的云巨头预计今年将投入7000亿美元的资本支出。人工智能的指数级增长已将处理能力（或称“算力”）的焦点从通用CPU转移到了专用的图形处理单元（GPU）。作为回应，金融中介机构正试图将这些物理芯片证券化。例如，在2024年，麦格理（Macquarie）为Lambda组织了一笔由GPU担保的5亿美元贷款，而拥有1吉瓦电力容量的CoreWeave则利用芯片和多年客户担保获得了黑石（Blackstone）等公司的贷款。
虽然大型算力销售商使用芯片作为抵押品，但买家需要金融工具来对冲这一昂贵投入的价格波动。初创公司Silicon Data和Ornn已分别与主要衍生品交易所——芝商所（CME Group）和洲际交易所（Intercontinental Exchange）合作，推出基于追踪英伟达（Nvidia）H100芯片每小时租用成本指数的算力期货。尽管Silicon Data的报价目前比Ornn仅追踪交易的指数高出10%（低于两个月前40%的差距），但由于云环境、硬件配置和地理位置的差异，标准化这些衍生品依然困难。
尽管存在这些标准化障碍，算力衍生品市场仍在扩大，这得益于即使在英伟达最新的Blackwell芯片进入市场后，Hopper代芯片依然坚挺的租赁价值。像Architect Financial Technologies这样的金融初创公司正在推出算力期货交易所，而像Roundhill Investments和ProShares这样的资产管理公司正申请建立算力期货交易型开放式指数基金（ETF）和杠杆产品。这些进展表明，在将处理能力商品化为人工智能时代基础货币的竞争中，速度正在急剧加快。
In the first quarter of 2026, global businesses spent $129 billion on cloud services, while the five largest American cloud giants are projected to invest $700 billion in capital expenditure this year. The exponential growth in artificial intelligence has shifted the focus of processing power, or "compute", from general-purpose CPUs to specialized graphics-processing units (GPUs). In response, financial intermediaries are attempting to securitize these physical chips. For instance, in 2024, Macquarie organized a GPU-backed loan of $500 million for Lambda, and CoreWeave, which controls one gigawatt of power capacity, has used chips and multi-year client guarantees to secure loans from firms like Blackstone.
While large compute sellers use chips as collateral, buyers require financial instruments to hedge against price volatility for this expensive input. Startups Silicon Data and Ornn have partnered with major derivatives exchanges—CME Group and Intercontinental Exchange, respectively—to launch compute futures based on indices tracking the hourly rental costs of Nvidia H100 chips. Though Silicon Data's quoted prices currently track 10% higher than Ornn's transaction-only index (down from a 40% difference two months ago), standardizing these derivatives remains difficult due to variations in cloud environments, hardware configurations, and geographic locations.
Despite these standardization hurdles, the market for compute-based derivatives is expanding, bolstered by the resilient value of Hopper-generation chips even after Nvidia's newer Blackwell chips entered the market. Financial startups like Architect Financial Technologies are launching compute futures exchanges, while asset managers like Roundhill Investments and ProShares are filing to establish compute-futures exchange-traded funds (ETFs) and leveraged products. These developments indicate a rapid acceleration in the race to commoditize processing power as the foundational currency of the AI era.
Source: How to turn compute into a financial asset
Subtitle: Entrepreneurs, exchange operators and AI firms are creating tradable instruments backed by processing power
Dateline: 6月 25, 2026 09:15 上午 | New York