How Kalshi Turns AI Computing Power into a Financial Instrument

Discover how the Kalshi platform is creating new financial markets centered around the cost of computing resources for artificial intelligence.

In recent years, the rapid growth of artificial intelligence (AI) has created demand for a new resource, previously little-known to financial markets — computing power, or “compute.” This concept encompasses the volume and quality of data processing required to train and operate modern AI models. The Kalshi platform is attempting to turn the cost of computing resources into a full-fledged financial instrument, opening new opportunities for traders and companies working in the AI sector. Computing Power as a New Financial Asset Traditionally, prediction markets have focused on political, economic, and sporting events. Kalshi goes beyond these boundaries, offering contracts tied to the hourly rental of Nvidia graphics cards — the key hardware for artificial intelligence. This not only expands the range of events available for trading but also helps build a financial infrastructure around the AI economy. Kalshi co-founder and CEO Tarek Mansour emphasized in an interview with The New York Times that compute markets are one of the most exciting areas for the company. According to him, compute is becoming a commodity comparable to oil, as businesses need mechanisms to assess and transfer future risks associated with this resource. However, unlike oil, which is standardized by grade, field, and delivery conditions, computing resources are much less homogeneous. Their value depends on the chip model, hardware configuration, data center location, software, availability, and contract terms. Therefore, standardizing the compute market remains extremely challenging. What Is Compute and How Is It Measured? The term “compute” in the context of AI usually means access to graphics processing units (GPUs) and the associated data center infrastructure. Computing power is needed for two main processes: Training — the process by which an AI model learns patterns from large volumes of data. Inference — using a trained model to generate answers or perform tasks based on new queries. Both processes require significant computing resources, but their load and cost structure differ. Graphics processing units (GPUs) were originally designed for computer graphics, but their architecture with many parallel computing blocks is ideal for AI tasks. Nvidia notes that GPUs can perform many operations simultaneously and scale up to large computing systems, which is critical for modern models. The cost of computing is often expressed in GPU-hours — a unit meaning the rental of a specific GPU for one hour. However, a GPU-hour is only a measure of volume, not an equivalent economic unit: one hour on an Nvidia A100 differs in performance and price from an hour on an H200 or B200. The rental price is influenced by demand from AI labs, the emergence of new chips, power and data center capacity constraints, regional availability, export restrictions, and the introduction of more efficient equipment. Additionally, computing power is a “perishable” resource: an unused GPU-hour cannot be stored and sold later, forcing providers to lower prices when utilization drops. How Kalshi Compute Markets Work Currently, Kalshi offers contracts tied to the rental prices of Nvidia A100, H100, H200, B200, and RTX 5090. These contracts cover both hourly values and average rates over a month or longer periods. These are binary (two-sided) event contracts. The trader does not buy computing power directly and does not reserve a GPU — instead, they buy a “Yes” or “No” contract on whether a certain price index will exceed a set threshold by the time of expiration. Example: a contract asks whether the hourly rental price of the H200 will be above $4.50 by the end of August. If, according to the contract rules, the official figure exceeds $4.50, the “Yes” contract pays out $1, otherwise — $0. The “No” contract works the opposite way. The result is confirmed using the Ornn Compute Price Index (OCPI). This is a family of indices calculated based on actual GPU rental transactions, expressed in dollars per GPU-hour. The index reflects completed deals, not advertised prices from cloud providers. Each Kalshi contract clearly specifies the calculation rules, observation window, and expiration time. This means a trader may see one price on a major cloud platform, while the Ornn index shows another, as they reflect different market segments. The contract is always settled strictly according to its rules, regardless of which price seems more “market” to the trader. Brief Overview of How Kalshi Binary Contracts Work: Select the chip of interest (for example, Nvidia H200). Formulate the condition (for example, “Hourly price above $4.50 by August 31”). The trader buys a “Yes” or “No” contract. At expiration, the official index value is compared to the threshold. Payout: $1 for a correct prediction, $0 for an incorrect one. What Is the Forward Curve for Compute? In July, Kalshi introduced a forward curve for Nvidia B200, H200, and A100. The forward curve is a chart reflecting the market’s expectations for future rental prices of a specific GPU. It answers the question: how much do traders think a GPU-hour will cost in a week, a month, or at the end of the year? The curve is built based on the probabilities “embedded” in binary contracts with different expiration dates and price thresholds. If the probabilities of exceeding various price levels are known for several contracts, their combination allows for an estimate of the expected price distribution and the calculation of the average expected price for a specific date. It’s important to understand that the curve itself is not a trading instrument — it is only an analytical reference reflecting current market expectations. The reliability of the forward curve depends on liquidity, the awareness of participants, and the quality of contract calculation rules. In July, a Kalshi Research study showed that the B200 curve was in backwardation (a downward trend), meaning expected prices were falling over time. Curves for older chips were relatively flat. This pattern reflected expectations of increased B200 supply after an initial period of shortage. However, the forward curve is not a guaranteed forecast, but merely a snapshot of current market expectations. Example of Building a Forward Curve: Binary contracts are taken for different dates (for example, September, October, November) and different price thresholds. The probabilities of these contracts being fulfilled are used to build the expected price distribution. The resulting curve shows how the market assesses the dynamics of GPU-hour rental prices in the future. Other Players: MNX and the Formation of AI Derivatives Markets Kalshi is not the only company creating markets for trading AI infrastructure. The startup MNX, founded by Manifold Markets co-founders Steven Gruszett and Ian Phillips, is developing a decentralized futures exchange focused on AI. The MNX model differs from Kalshi. While Kalshi offers regulated binary contracts with fixed dates and outcomes, MNX plans to launch perpetual contracts (open-ended derivatives) with leverage, tracking, for example, H100 rental rates without a fixed expiration date. In addition, MNX intends to list products tied to private AI company valuations, public AI company stocks, and model performance benchmarks. Steven Gruszett notes that the AI economy lacks a single platform where participants can trade or hedge risks across the entire value chain. Compute is a key link connecting chip manufacturers, cloud platforms, data centers, AI labs, and end users. The Main Problem — Liquidity From an economic perspective, the emergence of derivatives on computing power makes sense. AI labs are interested in predictable expenses, infrastructure providers in stable revenue, and lenders in having benchmark prices for evaluating equipment and long-term contracts. However, the key question is whether such markets can attract enough buyers, sellers, and speculators to form reliable prices. Computing power remains heterogeneous, and a global index may not reflect the actual costs of a specific company in a specific region. With low liquidity, order books become “thin,” spreads widen, and contract fulfillment probabilities become unstable. Competition in this segment is increasing: in addition to Kalshi and MNX, the sector includes traditional derivatives exchanges and specialized index providers. So far, no platform has achieved the depth and stability characteristic of mature energy or financial futures markets. Nevertheless, the formation of these markets is an important stage in the development of the AI economy. Not only are new chips and data centers emerging, but also financial instruments for managing their risks. For traders, Kalshi contracts are an accessible way to express their views on GPU rental price dynamics. For the market as a whole, their significance lies in the ability to form indicative forward prices for one of the most important resources of the digital economy. Table: Kalshi Contracts for GPU Rental GPU Model Contract Type Period Threshold Value Calculation Index Nvidia A100 Binary Hour/Month/Long-term Specified in contract Ornn Compute Price Index Nvidia H100 Binary Hour/Month/Long-term Specified in contract Ornn Compute Price Index Nvidia H200 Binary Hour/Month/Long-term Specified in contract Ornn Compute Price Index Nvidia B200 Binary Hour/Month/Long-term Specified in contract Ornn Compute Price Index RTX 5090 Binary Hour/Month/Long-term Specified in contract Ornn Compute Price Index Risks and Legal Aspects of Participating in Prediction Markets Trading on prediction markets involves risk and is not suitable for everyone. Even the best platforms only provide tools for decision-making and do not guarantee outcomes. Users should only risk what they are prepared to lose and always follow responsible trading principles. Additionally, platform availability and legal status vary by region. Before participating, it is important to independently check local laws and ensure the legality of using the chosen platform. What This Means for PPPoker Club Players For PPPoker club members, the emergence of compute markets is a clear example of how modern technology and financial instruments are penetrating all kinds of fields. Just like in poker, analysis, probability assessment, and risk management are crucial here. By studying new markets, players can develop skills useful both for trading and for decision-making at the poker table. The second important takeaway is the need to constantly monitor new trends and tools. As soon as new markets or trading formats appear, attentive players have a chance to gain an edge through a better understanding of product structure and calculation specifics. This applies not only to finance but also to poker formats, where innovations often bring extra benefits to those who master them first. Finally, experience trading on prediction markets can help PPPoker players gain a deeper understanding of probabilistic thinking and bankroll management. A responsible approach to risk, information analysis, and the ability to react quickly to changes are key skills for both successful trading and winning in poker clubs.