The finding, published July 17, highlights a familiar pattern: decentralized and crypto-adjacent platforms are setting the pace while traditional financial infrastructure catches up. Perpetual futures contracts on GPU rental prices and prediction market forward curves are already trading, even as CME Group and the Intercontinental Exchange wait for regulatory approval to launch their own competing products, targeted for late 2026.

Perpetual futures, worth explaining briefly: these are contracts with no expiry date that use a funding rate mechanism to keep the contract price anchored to the underlying spot index. The structure originated in crypto markets, tracing back to BitMEX's launch of the first Bitcoin perpetual future in 2016, and has been central to how traders in that space manage leveraged exposure.

---

How the market got here

GPU compute has become among the most volatile cost inputs for AI companies.

Nvidia Blackwell GPU rental prices surged 48 percent between mid-February and mid-April 2026, jumping from $2.75 to $4.08 per GPU-hour, according to data from index provider Ornn. H100 chips, an older generation, have since compressed to around $1.70 per GPU-hour on the Kalshi index, though broader market quotes for H100 cloud capacity range considerably higher depending on provider and configuration.

Index provider Ornn has described a generational fragmentation problem, noting that each new chip generation creates its own fragmented pricing market. With hyperscalers including Microsoft, Google, Amazon, and Meta committing an estimated $650 billion to $700 billion in AI infrastructure spending in 2026 alone, per data cited by ICE and index provider Ornn, the stakes around compute cost exposure have grown large enough to support a dedicated derivatives market.

Academic modeling published earlier this year on ArXiv estimated that futures contracts could reduce enterprise compute cost volatility by 62 to 78 percent. That paper is a preprint and has not been peer-reviewed. It also tracked a 40-fold collapse in AI inference token prices between early 2023 and early 2025, from roughly $60 per million output tokens to under $1.50, and through simulated modeling projected a reversal as demand outpaces constrained data center supply.

---

Who moved first

Architect Financial Technologies, founded by Brett Harrison (the former president of FTX US, the now-collapsed crypto exchange whose founder Sam Bankman-Fried was convicted of fraud), launched its AX exchange in October 2025 and announced GPU compute perpetual futures on January 21, 2026. The contracts are pegged to GPU and DRAM (memory chips used alongside GPUs in compute workloads) rental prices using Ornn's index and support margin trading funded with US dollars or USD stablecoins.

Kalshi, a CFTC-regulated prediction market platform, then launched GPU compute forward curves on July 14, covering Nvidia B200, H200, and A100 rental prices. Kalshi describes these as the first market-implied forward curves ever produced for GPU compute, a claim that has not been independently verified. "For the first time, the market has a view on where compute prices are headed," the company wrote at launch.

Kalshi's CEO Tarek Mansour has been direct about the scale of the opportunity: "Compute futures will eventually dwarf oil futures."

CME Group announced its own compute futures on May 12, partnering with Silicon Data to track daily GPU benchmark rental rates; the contracts are cash-settled. ICE followed on May 19, partnering with Ornn to target Nvidia H100, H200, and B200 contracts using Asian-style settlement, which averages daily index values over the contract duration. Both products remain pending regulatory approval.

Separately, the CFTC approved the first US-regulated perpetual futures contracts on May 29, clearing Kalshi's bitcoin perp product. Analysts say this clears the path for compute perps to follow the same regulatory route.

CME Group is currently suing the CFTC to block Kalshi's perpetual futures. The legal grounds for the suit have not been publicly detailed, but the dispute reflects how much is at stake as regulated and crypto-native platforms compete for the same market.

---

The view from outside the US

For AI developers in India and across Africa, the practical relevance of these markets is significant, even if direct access remains uncertain.

India's IndiaAI Mission, which provides access to more than 38,000 GPUs, subsidizes GPU compute at roughly 65 rupees per GPU-hour (around $0.78), covering an estimated 40 to 60 percent of compute costs for participating startups. The remainder is exposed to global spot price swings. A structural limitation compounds the problem: experts cited by Business Standard warn that the program's seven-day lease cap creates operational risk for companies running longer training or inference workloads.

India's GPU market is projected to grow at a 63 percent compound annual growth rate through 2032, meaning this exposure will only increase.

Indian institutional players, particularly venture-backed AI companies, could find compute futures useful for locking in costs across project planning cycles, though domestic access to US-regulated products is not guaranteed. India's existing crypto derivatives landscape offers relevant backdrop: a 1 percent tax deducted at source on crypto transactions and a wave of exchange delistings following a 2022 regulatory crackdown drove an active offshore derivatives trading community toward platforms such as Bybit, dYdX, and Hyperliquid. That same population of offshore traders represents a natural early adopter base for compute derivatives as those instruments mature.

In Africa, the stakes are even more acute. The continent holds less than 1 percent of global data center capacity while representing nearly 18 percent of the world's population.

A $720 million AI factory buildout is underway in 2026, deploying GPU-as-a-service infrastructure intended to reduce dependence on expensive foreign cloud providers and to support the training of indigenous-language AI models covering Swahili, Zulu, Hausa, Yoruba, and Amharic. Operators of these facilities face direct revenue risk from falling GPU rental prices.

A mature, globally accessible compute derivatives market would improve the financial bankability of African data center projects and give operators tools to hedge forward capacity commitments.

For both regions, the more immediately accessible instruments are likely the crypto-native perpetual futures on decentralized exchanges. On-chain perpetual futures currently represent approximately 13.5 percent of total perps open interest globally, up from 3.6 percent a year earlier. Within that on-chain segment, Hyperliquid controls roughly 70 percent of the market and processes more than $8 billion in daily volume, with real-world assets (off-chain assets like equities and commodities) accounting for up to 44 percent of that total at recent peak.

The pattern mirrors what happened in crypto broadly: emerging-market users accessed dollar exposure through stablecoins and offshore perps long before any regulated product reached their jurisdiction.

---

What comes next

Sell-side research firm Bernstein projects total prediction market volumes across all categories could reach $70 billion in 2026, doubling from the prior year and implying roughly $1.4 billion in annual revenue for exchanges and market makers.

Whether the regulated CME and ICE products clear their respective regulatory hurdles before year-end will shape how quickly institutional capital, including from outside the US, enters this space with the legal certainty it requires. Adding further uncertainty to that timeline is CME's ongoing lawsuit against the CFTC over Kalshi's perpetual futures, the outcome of which could redraw the boundaries between regulated exchanges and crypto-native platforms across the entire asset class.