GPU markets are still maturing, but given the wide range of companies using, selling and renting GPUs, there’s a robust market for spot prices on GPU rental, typically charged by the hour. According to data from AI Mining Co., which tracks daily GPU rental pricing across 28 marketplaces and cloud providers, median prices for Nvidia H100 GPUs ranged from $1.40 to $4.27 per hour across 13 marketplaces, while the average price for H200 GPUs were between $2.34 and $5 per hour across 10 marketplaces. And just over the past seven days, average H100 prices ranged from $2.79 to $3.33.

But while mature markets exist for GPUs, there’s less infrastructure around tokens themselves — the fundamental building blocks of contemporary AI models. Enterprise plans for major AI companies are commonly denominated in tokens: OpenAI, for example, charges $5 per million input tokens, and $30 per million output tokens if you want to use the API for its latest GPT-5.5 model. Even cloud providers are increasingly offering the opportunity to charge per-token, as in Amazon’s Bedrock system.

The effort comes amid an unprecedented buildout of AI infrastructure. Cloud service providers, private equity firms and infrastructure players alike have poured hundreds of billions into building data centers, anticipating that demand for GPUs and compute will continue to rise. An emerging crop of global neocloud companies is also vying for a piece of this demand. Some of these new entrants are specializing, focusing on inference, while others are competing with cloud giants like Oracle, AWS and Google Cloud to offer their services to AI companies.

By targeting AI tokens, the Shanghai exchange’s derivative product would be tied to how AI companies price their services, giving businesses, investors and data center operators a way to hedge against the cost of compute.