Pluto Wants to Let You Trade GPU Time Like Crude Oil

The Macro: Compute Is a Commodity That Doesn’t Trade Like One

The AI boom created a new commodity: GPU compute. Companies are spending tens of millions on NVIDIA H100 clusters. Cloud compute pricing fluctuates wildly based on demand. Spot instance prices on major cloud providers can swing 3-5x within a week. And yet there is no futures market, no options market, no standardized way to hedge against compute price volatility.

Think about how absurd that is. Airlines hedge jet fuel. Farmers hedge crop prices. Energy companies trade natural gas futures. Any large buyer of a volatile commodity uses financial instruments to manage price risk. AI companies buying compute just… hope prices don’t spike during their training run. The entire industry is flying without hedging.

CME Group has explored the concept of compute futures but hasn’t launched anything concrete. Akash Network offers a decentralized compute marketplace but that’s a spot market, not derivatives. Cloud provider spot markets let you buy cheap compute, but they can yank your instances with two minutes notice. There is no regulated exchange where a company can lock in compute prices for next quarter. Pluto wants to be that exchange.

The Micro: Berkeley Engineers With a Derivatives Trading Thesis

Pluto is building a regulated derivatives exchange for compute. The pitch is straightforward: if compute is a commodity, it should trade like one. Futures, options, the standard toolkit that every other commodity market has had for decades.

Ronit Jain and Aarav Patel, both co-founders, studied EE/CS at UC Berkeley and have built large-scale AI systems at Fortune 500 companies. The team of five includes former staff engineers from Meta and NVIDIA, electrical engineering PhDs, and a former CFO of ElectronX. That last hire signals they’re serious about the regulatory side. Building an exchange is 40% technology and 60% compliance. You need people who understand CFTC registration, capital requirements, and market surveillance. The ElectronX connection suggests they know this.

They’re based in New York, which makes sense for a financial infrastructure company. Part of YC’s Winter 2025 batch. The regulated part of “regulated exchange” is doing a lot of work in that description. Anyone can build a matching engine. Getting the regulatory approval to call it an exchange is the hard part.

The Verdict

I find Pluto’s thesis genuinely compelling. The idea that compute should trade like other commodities is logically sound, and the market need is real. If you’re an AI company planning a training run that will cost $20 million in compute over three months, you absolutely want the ability to lock in that price today. The counterparty is a cloud provider or GPU farm that wants revenue certainty. Both sides benefit.

The risks are significant and specific. First, regulatory approval is slow and uncertain. Getting licensed as a derivatives exchange in the US can take years. The CFTC is not known for speed. Second, the market needs liquidity to function, and liquidity is a chicken-and-egg problem. You need buyers and sellers to make prices meaningful, but buyers and sellers won’t show up until prices are meaningful. Third, the compute market is evolving fast. If NVIDIA’s next chip generation dramatically increases supply and stabilizes prices, the urgency for hedging instruments decreases.

At 30 days, the question is whether they’ve made progress on regulatory approvals. At 60 days, I’d want to see letters of intent from potential market participants. Are AI companies and cloud providers actually willing to trade compute derivatives, or is this a solution to a problem that people prefer to handle informally? At 90 days, any early trading activity or pilot program would be a strong signal. If they can show even a small group of sophisticated buyers and sellers transacting on the platform, the concept is validated. This is either a massive infrastructure play that creates an entirely new financial market, or a good idea that arrived five years too early. I lean toward the former, but the execution path is long and uncertain.