a16z Backed a $33 Million Bet on Trading GPUs Like Oil. Right Now It Sells One Number.

Andreessen Horowitz's crypto fund led a $33 million seed round into Ornn, a New York startup that wants to make raw GPU time tradable the way oil, power, and real estate are tradable. The round closed this week, with Galaxy Ventures, Nordstar, SV Angel, and a handful of earlier backers along for it. The pitch is big: build the price benchmarks, the hedging tools, and the capital-markets plumbing that the compute market never grew on its own. The reality is narrower. Most of what Ornn is selling today is a single number.

What is actually live

That number is the Ornn Compute Price Index, or OCPI. It is a benchmark for the price of GPU compute, and unlike the list prices clouds advertise, Ornn says it is built from actual cleared trades across hardware types, regions, and contract lengths. The index is distributed on the Bloomberg Terminal, which is the part of this story that should make you pay attention. Getting a price onto the Terminal is how a number becomes a reference rate that other people quote against.

Ornn has extended the same idea to model inference. Its Token Price Indices track the realized cost of output tokens from the big model providers, so a company can see what it is actually paying per million tokens instead of what the rate card says. That is a useful thing to measure, and measuring it is most of what Ornn does right now.

The market it says is missing

The argument for Ornn is straightforward, and it is mostly correct. Compute has become one of the most capital-intensive commodities on earth, with trillions of dollars headed into GPUs, data centers, and power over the next decade. Every other commodity at that scale has a price curve, a way to hedge, and instruments that let investors take exposure without owning the physical asset. Compute has close to none of that. a16z's Ali Yahya and Elizabeth Harkavy, who led the deal, called it "a multi-trillion dollar asset class financed largely through private spreadsheets and handshakes."

A data center operator borrowing against future rental income has no clean way to lock in what that compute will sell for. An investor who wants exposure to AI infrastructure growth has to buy a whole company or a whole building. Ornn wants to sit in that gap and sell the missing financial layer: futures so operators can hedge forward prices, residual-value models so a GPU's worth in three years is a market estimate instead of a straight-line guess, and capital-markets products that turn data-center revenue into something investors can buy.

The futures don't exist yet

Here is the gap between the pitch and the product. In May, Intercontinental Exchange, the company that owns the New York Stock Exchange, announced it would list cash-settled, dollar-denominated GPU compute futures referencing Ornn's index, covering H100, H200, B200, and RTX 5090 hardware. That is the headline instrument, the thing that would make compute genuinely tradable. It is not trading. The ICE contracts are "pending regulatory approval," which means they are an announcement, not a market. Residual-value protection and the capital-markets products are roadmap too.

So the honest version is this: Ornn sells a benchmark and some derived indices today, and it has a credible exchange partner lined up to build the rest, subject to regulators it does not control.

The question to ask about any index

An index is only as good as the trades feeding it, and that is the thing worth watching with OCPI. A benchmark built from "cleared trades" is more honest than one scraped from advertised prices, but it inherits whatever its inputs are. If most of the transactions printing into the index route through Ornn's own venue, the index measures Ornn's order flow, not the whole market. The spot market for GPU time is thin, lumpy, and concentrated among a few large buyers, which is exactly the condition where a young benchmark can look more representative than it is. Ornn has not published transaction volumes, so there is no way yet to judge how deep the data underneath the number goes.

None of that makes the index fake. It makes it early. Oil benchmarks took decades and a lot of disputed prints to become trusted. The reason ICE matters here is that exchanges have spent a century building the settlement and surveillance machinery that turns a price into a contract people will post margin against. If those futures clear regulatory review and start trading with real open interest, OCPI graduates from an interesting data product into market infrastructure. If they stall, Ornn is a well-funded analytics company with a good chart on the Terminal.

Who is building it

Ornn was founded in 2025 by Kush Bavaria and Wayne Nelms, who met at MIT and have backgrounds spanning quantitative trading and systems engineering. That mix matters more than usual here, because the hard part is not the trading idea. It is convincing data centers, clouds, and institutional desks to print their trades into someone else's index and then trade contracts settled against it. That is a trust problem and a liquidity problem before it is a technology problem.

The money says the bet is on the financial rails, not the model layer. a16z's crypto fund led it, which tells you they see this as market-structure infrastructure, the kind of thing that either becomes a standard or becomes nothing. Worth watching whether the futures actually list.