The moment AI investing shifts its center of gravity from "ownership" to the "secondary market"
Over the past two weeks, Wall Street has sent a quiet but decisive signal. On May 12, CME Group, partnering with GPU market intelligence firm Silicon Data, announced the launch of the world's first "compute futures" market. Barely a week later, Intercontinental Exchange (ICE) followed with its own plan: a suite of GPU compute futures contracts built on OCPI (the Ornn Compute Price Index), in partnership with the compute startup Ornn.
On the surface, this reads as product-launch news. But from a market perspective, the essence of the event can be captured in a single sentence: the center of gravity in investing is moving away from the companies that make GPUs and the companies that use them, and toward the secondary market where the "compute" those GPUs produce is bought and sold.
Why the next winner is the "exchange" — not the maker, not the user
For the past three years, the grammar of AI investing was simple. Buy Nvidia, or buy the hyperscalers that had stockpiled Nvidia chips. In other words, the structure was a bet on "owning hardware" or on "those who own the hardware."
The reason this structure is breaking down is clear. When a resource is scarce and its price swings wildly, the biggest money is made not by those who mine the resource or those who consume it, but by those who trade its volatility. In the oil market, the most consistent earners are not the petrostates that own the wells or the refiners — they are NYMEX and the trading desks placing bets on top of it.
GPUs are traveling precisely the same road. The AI boom turned GPU capacity into one of the most coveted resources on the planet, and firms unable to secure enough GPUs from hyperscalers began bidding up prices on the secondary market. The interesting part comes next. Just as energy traders hoard natural gas contracts ahead of winter, startups began stockpiling GPU reservations. Securing a resource not "to use it" but "to resell it" — this is the textbook prelude to a secondary market overtaking the underlying asset itself.
The reason the exchange is the ultimate beneficiary of this trend lies in its business model. CME is a massive exchange that runs futures across interest rates, equities, commodities, and FX, and this move extends its reach beyond traditional financial-asset futures into AI infrastructure. Whether GPU prices rise or fall, the exchange earns more in volume and fees the greater the volatility. It is a structure where the revenue is the movement itself, not the direction. A position that makes money simply when AI prices move — without having to guess where they go — is the most enviable seat in the current regime.
Into this picture steps a new kind of gatekeeper: the benchmark data provider. Index operators like Silicon Data and Ornn, alongside established price references such as SemiAnalysis or TrendForce's DRAMeXchange. As persistent supply shortages, accelerating demand, and the rapid financialization of AI infrastructure converge, the value of the specialist data providers underpinning this market keeps rising. It is not the one who sells the pickaxe, but the one who quotes the price of the pickaxe, who controls the new chokepoint.
A blade that cuts volatility — and a blade that inflates the bubble
The market implication of this event is double-edged. The same mechanism produces two opposite outcomes at once.
First, the side that reduces volatility. The primary function of a futures market is hedging. Just as an airline locks in fuel costs with crude futures, an AI company can pin down its future compute costs in advance. ICE's SVP of Futures Markets noted that as AI moved rapidly from research labs to becoming a core driver of the global economy, the GPU market evolved just as quickly, and a globally accepted pricing mechanism and risk-management tool became desperately needed. Indeed, the CEO of Silicon Data diagnoses today's compute market as fragmented — with prices varying dramatically across providers, regions, and contract structures — and historically lacking any standardized reference price.
A benchmark coming into existence is not merely the appearance of a single number. On top of it, you can discount a data center's future cash flows to a present value, and collateral valuation for loans and project finance becomes possible. Standardized GPU contracts can serve as loan collateral, and a lender can value a data-center project at the market price of its forward GPU output. Bundled compute futures can become the foundation for new financial products — compute-backed notes, ETFs, structured-yield products — and compute is transformed from an exotic asset into a rateable cash-flow stream. The moment an opaque operating cost turns into a measurable asset, the risk surrounding that asset is distributed and prices move toward stabilization. That is the textbook ending.
But the very same mechanism inflates the bubble. That an asset becomes measurable and capable of being collateralized and securitized means, in short, that leverage gets attached to it. One market analyst pins down this point sharply. From a real-economy view, justifying new supply is assumed to require commensurate new demand already in hand; but from a financial-markets view, trillions of dollars in capital can be mobilized into GPUs without any new end-user revenue at all. In other words, we move from an era where real demand justified capital formation to an era where capital formation races ahead of real demand. This is efficiency and danger in the same breath.
The overheating of the secondary market has already begun. The more dangerous signal is that the thing to be traded does not yet exist. On May 21, Bloomberg ETF analyst James Seyffart flagged that the "OK Computer Power ETF" had become the third to file for a compute futures ETF — joining Roundhill's Compute ETF (proposed ticker GPUX) and ProShares' AI Computing Power ETF — and the catch is that these futures contracts are still awaiting regulatory approval. The phrase "Wall Street has a habit of laying the financial plumbing before the water is even turned on" sums up this moment precisely. Not a single unit of the underlying asset has been transacted, yet three derivative ETFs are already lined up on top of it.
For retail investors, this is both an opportunity and a trap. Roundhill's filed GPUX is an actively managed ETF holding futures contracts pegged to the price of compute, including the cloud computing, HPC, and storage needed to train AI and machine-learning models. At last, a path opens for ordinary investors to bet directly on compute prices. But the structural pitfall of futures-based ETFs is well known: the risk is that contango and roll costs quietly erode returns over time — a lesson holders of early oil and volatility ETFs learned the hard way. It also remains an open question whether the compute futures market is liquid and robust enough to support an ETF; just as the Bitcoin futures market had CME behind it, compute futures will need comparable institutional infrastructure to operate at scale.
Bottom line: not the chip price, but the "compute price" becomes the thermometer of the AI economy
The metric investors must track is changing. Until now, Nvidia's share price and HBM shipments were the proxy indicators for the AI cycle. From here on, the slope of a compute futures curve like OCPI — contango or backwardation — will be the real-time thermometer of the AI economy. A steeply upward-sloping forward curve means the market is pricing in future supply shortages; a flat or downward-sloping one may be a signal of overinvestment.
The trend of the secondary market overtaking the underlying already looks like a settled direction. Exchanges and index operators are the quietest and surest winners. Whether that market becomes a tool that tames volatility, or a self-replicating bubble engine decoupled from real demand, however — frankly — cannot be declared by anyone, precisely because the same mechanism holds both outcomes, until we can see for ourselves the volume, open-interest, and roll-cost data over the next six months as the first contracts are struck and the first ETFs are listed.
References
CME Group Official Press Release — CME Group and Silicon Data Partner to Launch First Compute Futures (May 12, 2026) https://www.cmegroup.com/media-room/press-releases/2026/5/12/cme_group_and_silicondatapartnertolaunchfirstcomputefutures.html
ICE Official Press Release — ICE and Ornn to Launch GPU Compute Futures Contracts https://ir.theice.com/press/news-details/2026/ICE-and-Ornn-to-Launch-GPU-Compute-Futures-Contracts/default.aspx
Yahoo Finance — ICE, Ornn to Launch GPU Compute Futures (OCPI; H100/H200/B200/RTX 5090 contract details) https://finance.yahoo.com/markets/options/articles/ice-ornn-launch-gpu-compute-142609556.html
Asymmetrix Intelligence — CME and Silicon Data Partner to Launch First Compute Futures (benchmark data provider analysis) https://asymmetrixintelligence.substack.com/p/cme-and-silicon-data-partner-to-launch
Cryptobriefing — Roundhill Investments Files for ETF That Tracks Raw Computing Power (GPUX; secondary market & contango risk) https://cryptobriefing.com/roundhill-compute-etf-filing/
Cryptobriefing — OK Computer Power ETF Files Third Application for Compute Futures That Don't Exist Yet (May 21, 2026) https://cryptobriefing.com/ok-computer-power-etf-compute-futures/
Bloomberg — The Race to Offer Compute Futures to Masses Has Already Started (GPUX prospectus details) https://www.bloomberg.com/news/articles/2026-05-20/the-race-to-offer-compute-futures-to-masses-has-already-started
Dave Friedman — The Birth of the AI Compute Market (collateralization, securitization & risk-transfer mechanisms) https://davefriedman.substack.com/p/the-birth-of-the-ai-compute-market
Dave Friedman — The Future of AI Compute Looks Like Oil Futures Trading (the financialization logic of capital formation outrunning real demand) https://davefriedman.substack.com/p/the-future-of-ai-compute-looks-like
Yahoo Finance / Simply Wall St — CME Group Enters AI Compute Futures As Valuation Looks Stretched (CME valuation & execution risk) https://finance.yahoo.com/markets/options/articles/cme-group-enters-ai-compute-010930223.html