Wall Street is starting to treat rented AI compute like oil, and Ornn has put one of the first tradable prices on the screen.
Renting Nvidia's Blackwell GPUs has become expensive enough, and volatile enough, that the New York Stock Exchange's parent now wants futures contracts tied to the price. You don't need to love the comparison to crude oil to see the point. A market this large needs a number everyone can argue over.
Ornn is the startup trying to supply that number. Founded in 2025 by Kush Bavaria and Wayne Nelms, who a16z crypto says met at MIT, the New York company publishes the Ornn Compute Price Index, a transaction-based benchmark for GPU rental prices. Its own April announcement said the index had been added to the Bloomberg Terminal and was already being used by more than 400 data center operators, investors and AI companies.
The Blackwell price spike is the hook. Several outlets citing Ornn's index reported that renting an advanced Nvidia Blackwell GPU rose to $4.08 an hour in April, up 48% from $2.75 two months earlier. That isn't gentle movement for infrastructure sold as on-demand capacity. If you're an AI company running agents that call tools, retry tasks and burn through tokens all day, that price is not an abstraction. It lands in your monthly bill.
The Price Wall Street Wants #
Intercontinental Exchange announced on May 19 that it planned to launch U.S. dollar, cash-settled GPU compute futures tied to Ornn's Compute Price Index, pending regulatory approval. The contracts are meant to settle against live-traded spot prices for GPU compute across major hardware types, not a cloud provider's posted rate card.
CoreWeave's Q2 Revenue More Than Doubled to $2.58 Billion on AI Demand CoreWeave's second quarter revenue more than doubled to $2.58 billion and its backlog swelled to $104 billion, beating Wall Street's estimates. But a widening $626 million net loss, heavy customer concentration around Microsoft and OpenAI, and a debt-fueled capex plan show why investors keep selling the stock even when it beats. - CoreWeave Q2 revenue doubled AI demand - AI cloud infrastructure cost analysis
CME Group has now moved too. On August 11, it said it would launch Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures on October 5, 2026, also pending regulatory review. Each contract will represent a month's worth of rent for the relevant Nvidia GPU, using hourly rental-rate indexes published by Silicon Data.
That is the real signal here. ICE, CME and Kalshi are not all circling compute because the phrase sounds fashionable. They are doing it because GPU hours have started to behave like a commodity with serious price risk. A buyer wants to cap future costs. An operator wants to lock in revenue. A trader wants to bet that everyone else has mispriced the squeeze.
Here's the thing: Ornn doesn't have to prove AI is a bubble for its product to matter. It only has to prove that compute prices move enough to hurt people.
The Debt Behind The Trade #
The financing side is where this gets sharper. Goldman Sachs estimates about $7.6 trillion of capital could be spent from 2026 through 2031 across compute, data centers and power. That figure is not a slogan. It is a map of who has to borrow, build, lease, hedge and hope the demand curve keeps climbing.
Neoclouds such as CoreWeave and Nebius sit near the middle of that map. They buy large blocks of Nvidia capacity and lock in long contracts - then rent that access back to AI labs and enterprise customers. Investors have rewarded that exposure, then punished it whenever the financing starts to look too heavy. MarketWatch reported Wednesday that Nebius fell after announcing a $4.5 billion convertible bond offering, while Investor's Business Daily noted that CoreWeave carries more than $46 billion in debt against about $5 billion in equity.
Debt changes how you read a GPU price index. If a rack of chips is collateral, the lender needs to know what those chips can earn next month, not just what they cost to buy last year. If rental rates fall, the spreadsheet breaks fast. If rates rise, the operator looks smart and the customer feels the pain.
Frankly, that is what Ornn is selling: not AI hype, but a cleaner way to price the risk everyone already took. A hedge fund can short compute instead of shorting an AI stock. A data center operator can protect future rental income - and a lender no longer has to pretend GPU collateral is easy to value just because Nvidia's demand looks strong today.
Nscale Races Toward a Fall IPO After Locking In Billions From Microsoft Nscale has hired Goldman Sachs and JPMorgan for a possible fall IPO near a $25 billion valuation, built largely on roughly $23 billion in Microsoft GPU deals. CoreWeave's post-IPO swings show public investors will test that backlog hard. - Nscale GPU startup IPO fall - Microsoft GPU rental company valuation
Ornn raised $5.7 million in October 2025, according to its seed announcement, then a16z crypto said on June 24 that it led a $33 million seed round for the company. Galaxy Ventures, Nordstar and SV Angel also joined, according to reports on the round. For a company this young, the timing is unusually useful: the market is already nervous, the exchanges are already building, and the Bloomberg Terminal already has a ticker for the benchmark.
Compute futures may fail to find deep liquidity at first. Many new contracts do. But the direction is clear enough. AI infrastructure is no longer just a technology buildout. It is a capital markets problem with power bills, GPU shortages, debt packages and now, price indexes. Ornn did not create that problem. It gave Wall Street something to trade while it tries to manage it.
Also read: TerraPower Breaks Ground in Wyoming to Power the AI Data Center Boom • Google Gets the Right to Buy $12.2 Billion of Marvell Stock • LivePerson Shareholders Vote Thursday on SoundHound's $3.33 Buyout Offer