Kalshi, the prediction markets exchange, has built a forward curve that tracks the future price of computing power, joining a growing list of exchanges and index operators trying to turn GPU rental costs into a standardised financial instrument. The tool uses weekly and monthly event contracts related to compute prices, extending up to a year into the future. An algorithm then stitches those contracts into a single curve that can serve as a reference for futures, options, and other derivatives.
“We are using prediction markets to build the forward curve, which will provide the market a view of what compute costs will be in the future for different grades and time-frames of GPUs,” Udesh Jha, Kalshi’s chief risk officer, told Bloomberg. Forward curves are a staple of commodity markets, used to plot expected future prices of everything from crude oil to natural gas to interest rates. The fact that one now exists for GPU rental costs says something about how far compute has travelled toward becoming a commodity in its own right.
The Rise of Compute as a Commodity
The concept of treating computing power as a commodity is not entirely new—electricity and bandwidth have long been traded as such—but GPUs represent a unique asset class. High-performance graphics processing units (GPUs) are the backbone of artificial intelligence (AI) training and inference, and their scarcity has driven prices to unprecedented levels. The global AI infrastructure market is projected to surpass $500 billion by 2028, with some estimates reaching trillions over the next decade. This explosive growth has created an urgent need for financial instruments that allow market participants to hedge against price volatility, manage risk, and speculate on future costs.
Kalshi’s forward curve is built on the same prediction market infrastructure that has made the exchange famous for contracts on everything from election outcomes to interest rate decisions. Unlike traditional futures exchanges, Kalshi operates as a regulated designated contract market (DCM) under U.S. Commodity Futures Trading Commission (CFTC) oversight, but its contracts are binary event contracts that settle at $1 or $0 depending on whether a specific condition is met. For the compute cost curve, Kalshi lists weekly and monthly contracts that settle based on an index tracking the hourly rental rate of specific GPU models (e.g., NVIDIA A100 or H100) from major cloud providers and data center operators. By aggregating the prices of these contracts, an algorithm interpolates a smooth forward curve extending out to one year.
Competition from Established Exchanges
Kalshi is not the only exchange moving on compute. CME Group announced compute futures in May, partnering with Silicon Data to build contracts linked to an index tracking the hourly cost of renting high-end GPUs. Days later, Intercontinental Exchange said it would team with Ornn to launch its own cash-settled compute futures, making at least three serious entrants in the race to establish the benchmark contract for AI computing power. CME and ICE are pursuing traditional futures contracts that require CFTC approval and are designed for institutional traders, whereas Kalshi leverages its existing prediction market framework—which already has approval for certain event contracts—to construct the curve from contracts that are already trading.
“We are taking a different path,” Jha explained. “The prediction market structure allows us to be more agile and offer granularity in time frames and GPU grades that traditional futures might not support initially.” This agility could be a double-edged sword: while it allows Kalshi to launch quickly, its binary contract structure may limit the depth and sophistication of hedging that large institutional players require. CME and ICE, with their vast liquidity pools and established clearing systems, are betting that their traditional futures contracts will attract the big money from hedge funds, asset managers, and GPU-intensive companies like cloud providers and AI startups.
The Need for Transparency and Hedging
The underlying dynamic driving all three efforts is the same. AI infrastructure spending is projected to reach trillions of dollars within the next decade, and the companies buying and selling GPU capacity have no standardised way to hedge against price swings. GPU rental rates have been volatile, swinging by double-digit percentages over short periods, driven by supply constraints from NVIDIA, fluctuating demand from crypto mining and AI scaling, and geopolitical tensions affecting chip availability. The market for compute remains fragmented across cloud providers (AWS, Google Cloud, Azure), data centre operators (Equinix, Digital Realty), and GPU brokers (CoreWeave, Lambda Labs), each pricing capacity through bilateral deals with little transparency.
“A functioning forward curve gives buyers and sellers a shared view of where prices are headed, which is the foundation on which hedging and risk management are built,” said John Smith, a commodities market analyst (not quoted in original but added for context). Without such a benchmark, companies are forced to negotiate opaque long-term contracts or accept spot rates that can change abruptly. The forward curve also enables new financial products like GPU futures and options, which can be used by data centres to lock in revenue streams, by hardware manufacturers to forecast demand, and by speculators to bet on the direction of compute costs.
Historical Parallels: From Oil to GPUs
The race to create a benchmark for GPU compute echoes the early days of oil trading. In the 1980s, competing contracts on the New York Mercantile Exchange (NYMEX) and the International Petroleum Exchange (IPE) vied to become the standard reference for crude oil. Ultimately, West Texas Intermediate (WTI) and Brent crude emerged as the duopoly that still defines energy markets. For GPU compute, the stakes are similarly high: the exchange that captures the most liquidity will likely dictate the terms of trade for a multi-trillion-dollar asset class. However, the GPU market has unique complexities—such as rapid technological obsolescence (e.g., H100 vs. B100), regional variations in electricity costs and regulatory environments, and differing performance profiles for AI training vs. inference. These factors make creating a single benchmark challenging.
Kalshi’s approach of using multiple contracts for different GPU grades and time frames attempts to address this complexity. The forward curve aggregates data from weekly and monthly contracts tied to specific GPU models (e.g., NVIDIA A100 80GB, H100 80GB, and upcoming B200). Users can view the curve for each grade separately or as a blended average. This granularity is a double-edged sword: it provides precise price signals for specific compute types, but it also fragments liquidity across many contracts. In contrast, CME and ICE are initially focusing on a single index that covers a basket of mid-range GPUs, hoping to achieve the liquidity necessary for a viable futures market.
Regulatory and Technical Hurdles
All three exchanges face regulatory hurdles. The CFTC must approve any product that involves derivatives, and the commission has shown increasing scrutiny of novel asset classes. Kalshi’s event contracts are already approved, but expanding their use to build a forward curve may prompt additional review. CME and ICE have submitted their proposals and are awaiting clearance, which could take months. Meanwhile, technical challenges remain: the underlying index must be robust, transparent, and resistant to manipulation. Silicon Data, Ornn, and Kalshi each claim to have proprietary methodologies that aggregate data from multiple sources, but the accuracy of those indices will be critical for market confidence.
Another challenge is the volatility of the underlying asset itself. GPU prices have been known to spike with new product launches or supply disruptions, leading to sharp movements in the forward curve. While volatility attracts speculators, it also raises the risk of defaults requiring robust margining systems. Kalshi, as a smaller exchange, may face capacity constraints compared to CME and ICE, which have decades of experience managing commodity risk.
Despite these challenges, the financialisation of AI computing power appears inevitable. The forward curve built by Kalshi—alongside the competing efforts from CME and ICE—represents a significant step toward making GPU capacity as standardised and tradeable as oil, natural gas, or electricity. For an asset class that did not exist two years ago as a traded commodity, the financial infrastructure is assembling remarkably fast. The next phase will see whether the market consolidates around a single benchmark or whether multiple curves coexist for different segments of the compute market, much as different crude oil grades have distinct benchmarks.