GPU正在变成石油。不是比喻，是字面意义上的——华尔街正在为它开发期货合约。
8月中旬，芝加哥商业交易所集团（CME Group）宣布计划于2026年10月5日推出两份算力期货合约，分别跟踪H100和B200 GPU的每小时租赁价格指数，每份合约约730 GPU小时（相当于一个月租金），现金结算，在纽约商业交易所（NYMEX）挂牌。这是全球首个标准化的算力期货产品，目前尚待CFTC（美国商品期货交易委员会）审批。
与此同时，CFTC已正式就算力衍生品的监管框架征求意见，ICE（洲际交易所）也在开发基于Ornn真实成交指数的竞品期货合约。算力的金融化时代，正式拉开序幕。
为什么GPU需要期货？因为价格波动太疯狂了
算力期货不是凭空发明的金融玩具，它有真实的商业需求。过去7个月，H100现货租赁价格翻倍，不同平台同日报价从$2.78到$7.18不等——价差超过2.5倍。对依赖GPU推理和训练的AI公司来说，这种波动性是致命的：你上个月做的预算这个月就不够用了，你签的年度合同在现货价格暴涨时变成了亏本买卖。
这跟航空公司对冲油价是一模一样的逻辑。航空公司不可能每次加油都按现货市场价付——油价涨一倍，机票价格没法立刻跟着涨。所以航空公司在原油期货市场锁定未来的燃油成本，把不确定性变成确定性。AI公司现在面临同样的问题：算力就是AI时代的石油，而石油需要期货市场来管理价格风险。
CME选择的标的指数来自Silicon Data——一家由全球交易公司DRW支持的GPU市场情报和基准定价公司。期货合约跟踪的是H100和B200的现货租赁价格，每个月结算一次。AI公司可以在期货市场"锁仓"未来6个月甚至12个月的算力成本，投机者可以押注GPU价格涨跌，做市商提供流动性——整个石油期货市场的剧本，在算力上原封不动地重演。
算力成为大宗商品意味着什么？
第一，算力定价权正在从云厂商手中向金融市场转移。以前AWS、Azure、Google Cloud说H100一小时多少钱就是多少钱，客户没有议价权。期货市场形成后，会有一个公开透明的基准价格，云厂商的定价权被削弱——就像石油公司不能单方面决定油价一样。
第二，算力的"囤货"和"卖空"成为可能。预测GPU供应紧张的机构可以买入期货合约锁定利润，预测价格下跌的可以做空。这会吸引传统商品交易员、对冲基金进入AI算力市场，带来更多流动性和价格发现——但也可能带来投机和波动。
第三，中国公司需要关注定价权问题。CME的期货合约基于美国市场的GPU租赁价格，主要反映的是Nvidia GPU的现货价格。中国国产芯片（华为昇腾、寒武纪等）的定价目前完全不在这个体系内。如果算力期货成为全球标准，中国AI公司在对冲算力成本时会面临定价权旁落的风险。InfoQ的报道也提到了这一点。
"当H100租金7个月翻倍时，AI公司才发现自己跟航空公司一样，需要在期货市场锁定成本。"—— 一位商品交易员的评论
当然，算力期货现在还处于早期。CME的合约尚待监管审批，流动性需要时间培养，B200的租赁市场还不够成熟。但方向是明确的：算力正在从技术资源变成金融资产，从云厂商的私有产品变成华尔街交易的大宗商品。
还记得038期我们报道H100半年涨价40%、现货全面售罄吗？仅仅两期之后，华尔街就已经在为这种价格波动开发金融工具了。AI的每一个痛点，都是金融市场的新机会。
明天见。
GPUs are becoming oil. Not metaphorically — literally, in the sense that Wall Street is building futures contracts for them.
In mid-August, CME Group announced plans to launch two compute futures contracts on October 5, 2026, tracking hourly rental price indexes for H100 and B200 GPUs respectively. Each contract covers roughly 730 GPU-hours (equivalent to one month of rental), cash-settled, listed on NYMEX. These are the world's first standardized compute futures products, pending CFTC review.
Meanwhile, the CFTC has formally opened a comment period on compute derivatives regulation, and ICE (Intercontinental Exchange) is developing competing futures based on Ornn's real transaction index. The financialization of compute has officially begun.
Why Do GPUs Need Futures? Because Price Volatility Is Insane
Compute futures aren't a fabricated financial toy — they address real commercial pain. Over the past seven months, H100 spot rental prices doubled, with same-day quotes across platforms ranging from $2.78 to $7.18 — a 2.5x spread. For AI companies dependent on GPU inference and training, this volatility is brutal: the budget you signed last month is insufficient this month; an annual contract signed at lower rates becomes a money-loser when spot prices spike.
This is exactly why airlines hedge jet fuel. Airlines can't pay spot market prices every time they refuel — if oil doubles, ticket prices can't adjust overnight. So airlines lock in future fuel costs on crude futures markets, converting uncertainty into certainty. AI companies now face the same problem: compute is the oil of the AI era, and oil needs futures markets to manage price risk.
CME's chosen index provider is Silicon Data, a GPU market intelligence and benchmark pricing firm backed by global trading firm DRW. The futures track H100 and B200 spot rental prices with monthly settlement. AI companies can "lock in" compute costs 6-12 months out; speculators can bet on GPU price moves; market makers provide liquidity — the entire oil futures playbook is being replicated for compute, line by line.
What Does Compute as a Commodity Mean?
First, compute pricing power is shifting from cloud providers to financial markets. Previously, AWS, Azure, and Google Cloud set H100 hourly rates unilaterally with no customer bargaining power. A futures market creates a transparent benchmark price, eroding cloud vendors' pricing power — just as oil companies can't unilaterally set crude prices.
Second, compute "hoarding" and "short selling" become possible. Institutions expecting GPU shortages can buy futures to lock in profits; those expecting price drops can short. This will attract traditional commodity traders and hedge funds into AI compute markets, bringing liquidity and price discovery — but also speculation and volatility.
Third, Chinese companies should watch the pricing power question. CME's futures are based on US-market GPU rental prices, primarily reflecting Nvidia GPU spot rates. Chinese domestic chips (Huawei Ascend, Cambricon, etc.) are entirely outside this system. If compute futures become a global standard, Chinese AI companies hedging compute costs face pricing power risk — a point also noted by InfoQ's Chinese coverage.
"When H100 rental prices doubled in seven months, AI companies discovered they're just like airlines — they need futures markets to lock in costs."— A commodity trader's observation
Of course, compute futures are early. CME's contracts await regulatory approval, liquidity needs time to develop, and the B200 rental market isn't yet deep enough. But the direction is clear: compute is transitioning from a technical resource to a financial asset, from cloud vendors' proprietary product to a Wall Street commodity.
We reported H100 prices rising 40% in six months with spot exhaustion in Issue 038. Just two issues later, Wall Street is already building financial instruments for that volatility. Every pain point in AI creates a new opportunity for financial markets.
See you tomorrow.
算力就是AI时代的石油，而石油必须有期货市场——H100租金7个月翻倍，逼出了全球首个算力期货。
—— Dawn Vision编辑部
Compute is the oil of the AI era, and oil needs futures — H100 rental prices doubling in seven months forced the world's first compute futures into existence.
— The Dawn Vision Editorial Desk
CME · 算力期货 · H100 · B200 · 10月5日 · NYMEX · Silicon Data · DRW · CFTC · ICE · 算力金融化 · 大宗商品 · 价格对冲
CME · compute futures · H100 · B200 · October 5 · NYMEX · Silicon Data · DRW · CFTC · ICE · compute financialization · commodities · price hedging
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理的 9 个源信号生成，经编辑部人工审核。素材来源：上海证券报、新浪财经、ai2.work、InfoQ。
This article was generated by the Dawn Vision cognitive engine processing 9 source signals, with human editorial review. Sources: Shanghai Securities News, Sina Finance, ai2.work, InfoQ.
