CME compute futures delayed as GPU rental becomes a tradable contract
Economic Observer, republished by 36Kr, reports that two compute futures designed by CME Group and Silicon Data on Nvidia H100 and B200 rental prices were originally slated to list on NYMEX on Oct 5 but were delayed after CFTC flagged the product as novel or complex, extending the review to a Nov 9 completion date, with CME marking the initial listing date as to be announced. Meanwhile ICE and Ornn began publishing Ornn Compute Index reference data, Kalshi already offers GPU compute price prediction contracts and forward curves, and Architect and Nodal Exchange are moving on similar products. Interviews with Chinese compute rental firms, cloud vendors, model companies and AIDC park operators found no mature public compute price index domestically, so firms rely on long-term contracts, prepayments and reserved capacity, while Nanhua Futures unit Henghua International signed a memorandum with Silicon Data.
Financializing compute pricing would let AI firms hedge training and inference cost swings, but the report notes the lack of a domestic public index and the risk that derivatives import new trading risks before they hedge existing ones.
英伟达H100显卡一小时的租金，正在被做成可以交易的金融合约。
这听起来离普通人很远，但对大模型公司、云厂商和算力租赁商来说，GPU（图形处理器）租金每跳动一次，都可能吞掉一部分利润。
训练大模型需要长时间持续占用算力。算力涨价，AI（人工智能）企业的训练成本随之抬升；算力租金回落，重金购卡、自建算力集群的投资方，则要面临营收缩水与硬件资产贬值的双重压力。
CME Group（芝商所）和Silicon Data（算力价格指数公司）原计划推出挂钩英伟达H100、B200显卡租赁价格的期货合约，原定10月5日在NYMEX（纽约商品交易所）挂牌，随后因该产品存在“新颖或复杂”问题被监管延审推迟。
截至10月19日，距离CFTC（美国商品期货交易委员会）对CME Group算力期货完成延长审查，还有二十几天。目前，上市时间仍未确定。
经济观察报近日采访国内头部算力租赁企业、民营算力服务商、中小算力业主、央企算力采购方、大模型公司、云厂商和AIDC（人工智能数据中心）园区运营方发现，即便这款算力期货仍处于监管延审阶段，算力可被期货化定价这件事，也已经把一个现实问题推到中国企业面前——当算力租金像油价那样剧烈波动，谁来替AI企业兜住成本风险。
#CME Group#Silicon Data#CFTC#NYMEX#ICE#Ornn#Kalshi#Architect#Nodal Exchange#Nanhua Futures#Nvidia H100#Nvidia B200