Kalshi, the prediction markets exchange, has constructed a forward curve that tracks the anticipated future cost of computing power, marking a significant step in the financialization of artificial intelligence infrastructure. This tool utilizes weekly and monthly event contracts tied to compute prices, projecting costs up to a year ahead. An algorithm then aggregates these contracts into a unified curve, intended to serve as a reference for futures, options, and other derivatives.
Udesh Jha, Kalshi’s chief risk officer, highlighted the innovation: “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.” Forward curves are a staple of commodity markets, where they plot expected future prices for crude oil, natural gas, interest rates, and more. The creation of one for GPU rental costs underscores how far computing capacity has evolved toward becoming a tradable commodity in its own right.
Understanding Forward Curves in Commodity Markets
A forward curve is a graphical representation of the expected future prices of a commodity or financial instrument over time. In traditional commodity markets such as oil, natural gas, or agricultural products, these curves provide critical information for producers, consumers, and speculators. For instance, an oil producer uses the forward curve to decide whether to lock in future sales at current prices, while a refinery uses it to plan procurement. The curve’s shape—whether contango (upward sloping) or backwardation (downward sloping)—reflects market expectations about supply, demand, storage costs, and risk premiums. Kalshi’s compute forward curve follows the same logic, but instead of barrels or bushels, the underlying asset is GPU computing power, specifically the hourly cost of renting high-performance graphics processing units used for AI training and inference.
The introduction of such a curve for computing power is a breakthrough for an industry that has grown explosively since the launch of large language models like ChatGPT. Before this, GPU rental rates were opaque, negotiated in bilateral deals between cloud providers, data center operators, and AI startups. Prices fluctuated wildly based on demand surges, supply chain bottlenecks, and technological shifts, leaving market participants with no standardized way to price or hedge future capacity needs. A functional forward curve changes that by providing a transparent, market-driven expectation of future costs.
Kalshi’s Methodology: Prediction Markets as a Price Discovery Tool
Kalshi’s approach differs from traditional futures exchanges because it leverages prediction markets—platforms where participants trade contracts that pay out based on the outcome of specific events. In this case, Kalshi lists weekly and monthly event contracts that settle on the observed price of compute for various GPU tiers (e.g., NVIDIA A100, H100, or upcoming B200). Traders buy and sell these contracts, effectively betting on where compute prices will be in the future. An algorithm then synthesizes the prices of these contracts into a continuous forward curve. This method is flexible and can be updated rapidly as new information emerges, making it particularly suited to a fast-evolving market like AI infrastructure.
Prediction markets themselves have gained credibility as accurate forecasting tools, often outperforming expert polls and surveys. By harnessing the wisdom of crowds and real-time incentives, Kalshi’s compute curve may offer more dynamic and granular insights than traditional futures, which rely on a smaller set of institutional participants. Jha emphasized that the curve is a “key enabler for hedging, risk management, and speculative activity alike,” giving stakeholders a vital tool to navigate the uncertain future of GPU supply and demand.
The Broader Race to Financialize Compute
Kalshi is not alone in this endeavor. In May 2024, the CME Group announced plans to launch compute futures, partnering with Silicon Data to create contracts linked to an index that tracks the hourly rental cost of high-end GPUs. Shortly after, Intercontinental Exchange (ICE) revealed its own cash-settled compute futures, developed in collaboration with Ornn. These three entrants—CME, ICE, and Kalshi—represent the most serious attempts to establish a benchmark contract for AI computing power, much as Brent and West Texas Intermediate (WTI) dominate crude oil pricing.
While CME and ICE are pursuing traditional futures contracts that require approval from the Commodity Futures Trading Commission (CFTC), Kalshi’s prediction market framework operates under a different regulatory regime. Kalshi is registered as a designated contract market (DCM) and already offers event contracts covering economic indicators, weather, and other topics. Its compute forward curve emerges from contracts already trading, rather than requiring a new product launch. This speed and regulatory agility could give Kalshi an early-mover advantage in capturing liquidity and establishing its curve as the industry reference.
The competition among these exchanges mirrors the historical race to standardize commodity benchmarks. In oil, Brent became the global benchmark for light sweet crude, while WTI serves the North American market. For natural gas, Henry Hub emerged as the U.S. standard. Similarly, GPU compute is likely to coalesce around one or two dominant forward curves, depending on which exchange attracts the most trading volume and open interest. The stakes are high: the exchange that wins will earn listing fees, data revenue, and a central role in the financial ecosystem of trillion-dollar AI infrastructure.
Why GPU Compute Needs Financialization
AI infrastructure spending is projected to reach trillions of dollars over the next decade. According to industry analysts, global spending on AI-specific hardware, including GPUs, data centers, and networking, could exceed $500 billion by 2025 and grow at a compound annual rate above 30%. The companies buying and selling GPU capacity have no standardized hedge against price swings, which can be violent—NVIDIA’s H100 GPU rental rates, for instance, spiked 50% in early 2023 amid supply shortages, then corrected as cloud providers expanded capacity.
The market for compute remains highly fragmented. Cloud hyperscalers like AWS, Microsoft Azure, and Google Cloud each have their own pricing models, often bundled with storage and networking. Independent GPU brokers aggregate capacity from smaller data centers and resell it at markups. This opacity creates inefficiencies: buyers may overpay or face delays, while sellers struggle to forecast revenue and invest optimally. A forward curve brings transparency, allowing both sides to agree on a future price today, reducing uncertainty and enabling capital planning.
For example, a startup training a large AI model might need 1,000 H100 GPUs for three months starting next quarter. Without a forward curve, the startup must negotiate with multiple cloud providers or brokers, receiving disparate quotes with varying terms. With a curve, the startup can lock in a price today, perhaps at a premium or discount to the spot rate, and budget accordingly. Conversely, a data center operator with idle capacity can sell future compute at a guaranteed price, securing revenue and managing utilization risk.
Financialization also opens the door to derivative products that further refine risk management. Once a forward curve exists, exchanges can list futures, options, swaps, and structured notes. For instance, a hedge fund might short compute futures if it expects a glut of new GPU supply, while an AI lab could buy call options to cap its costs. These instruments attract speculative capital, deepen liquidity, and improve price discovery—all feedback loops that make the benchmark more robust.
Challenges and Hurdles to Compute Financialization
Despite the promise, turning GPU compute into a standardized commodity faces several obstacles. First, the underlying asset is not homogenous. GPUs come in different architectures (e.g., NVIDIA’s Ampere, Hopper, Blackwell), each with varying performance, memory, and power efficiency. A forward curve must account for these grades, just as crude oil benchmarks differentiate between light sweet and heavy sour. Kalshi’s curve claims to cover “different grades and time-frames,” but the market may require multiple curves for different GPU types.
Second, the spot price of compute is not as transparent as oil or wheat. The index used for contracts must be based on observable transaction prices from multiple sources, which are often confidential. Both CME’s Silicon Data index and ICE’s Ornn partnership aim to aggregate private data, but the accuracy and representativeness of these indices are unproven. Kalshi’s prediction market approach bypasses the need for a spot index by using event contracts settled on a yet-to-be-disclosed benchmark, but the exchange must still ensure that settlement prices are reliable and manipulation-resistant.
Third, regulatory oversight is evolving. The CFTC has not yet approved compute futures for CME or ICE, and while Kalshi operates under its DCM license for event contracts, the expansion into pure commodity pricing could attract additional scrutiny. Certain market participants worry that the financialization of compute could introduce speculative bubbles or volatility that harms the real economy—similar to critiques of oil futures in the 2000s.
On the technology side, the physical nature of computing power is more complex than traditional commodities. Electricity, cooling, networking, and software stacks all influence the effective cost of compute. A standardized contract must define exactly what is being delivered: raw GPU cycles? Including memory bandwidth? Excluding power? These details matter for pricing and settlement. Both CME and ICE are working with specialized partners to define their contracts precisely, and Kalshi’s prediction market participants will implicitly price these nuances through their trading.
Finally, the adoption of compute derivatives requires education and trust. Most participants in the AI ecosystem are engineers or entrepreneurs, not financial traders. They may be unfamiliar with futures, curves, and hedging strategies. Exchanges will need to provide educational resources, simplified interfaces, and perhaps team up with brokers to onboard new users. The success of these products depends on attracting not just hedge funds but also the actual producers and consumers of compute—data center operators, cloud providers, and AI firms.
Broader Implications for AI and Commodity Markets
The emergence of GPU forward curves signals a maturation of the AI industry. In its early years, AI was driven by academic research and venture capital, with little need for financial risk management. Today, AI is an industrial-scale enterprise requiring billions in capital expenditure. Just as the rise of oil in the 19th century led to futures markets on the Chicago Board of Trade in the 1860s, and the dot-com boom spawned derivatives for technology stocks, the AI boom is spawning derivatives for its core input: computing power.
If compute becomes a widely traded commodity, it could reshape how technology companies budget and invest. Currently, tech giants like Google, Microsoft, and Amazon dominate GPU procurement through their own cloud arms, enjoying preferential pricing. A transparent forward curve could level the playing field for smaller players, enabling them to hedge and plan with the same visibility as the hyperscalers. This could accelerate innovation by reducing the capital risk for startups.
Moreover, a compute benchmark could attract sovereign wealth funds, pension funds, and other institutional investors seeking exposure to the AI theme without buying individual stocks like NVIDIA or Microsoft. They could take positions in compute futures, effectively betting on the long-term demand for AI processing. This would deepen the capital pool available for AI infrastructure financing, potentially lowering the cost of capital for data centers and GPU manufacturers.
The forward curve also provides a barometer for the broader economy. Compute prices reflect AI adoption rates, supply chain health, and energy costs. A rising curve could signal booming demand and tight supply, while a falling curve might indicate overcapacity or technological breakthroughs that reduce GPU costs. Analysts and policymakers could use compute futures as a leading indicator for productivity and economic growth, much as they use copper or lumber prices today.
In the nearer term, the race among Kalshi, CME, and ICE will determine which forward curve becomes the industry benchmark. Each competitor brings distinct strengths: CME has unparalleled liquidity and regulatory approval in futures, ICE has experience with complex commodities like carbon credits and electricity, and Kalshi has speed and a retail-friendly prediction market model. The winner will likely be the one that attracts the most genuine hedging demand from compute buyers and sellers, not just speculators. As Jha noted, the curve is a tool for “hedging, risk management, and speculative activity,” but lasting benchmarks are built on real economic need.
For now, Kalshi has taken the first public step, demonstrating that a forward curve for compute is feasible. The market’s response will be closely watched. If trading volumes grow and the curve proves reliable, it will validate the concept and pressure CME and ICE to accelerate their launches. Even if Kalshi does not ultimately dominate, its initiative has already forced the industry to confront the question of how to price the future of AI—a question that will only become more pressing as the world invests trillions into machines that think.