Nvidia’s AI Chips Become Wall Street’s Next Big Bet: Inside the New Compute Futures Market

Investment in artificial intelligence continues to climb at a remarkable pace. Yet establishing a reliable pricing framework for that investment remains a complex challenge.

Nvidia (NVDA) underscored the magnitude of this surge recently, projecting approximately 70% revenue growth for fiscal 2028, even as supply constraints persist.

The company is evolving beyond chip manufacturing, offering financing to customers and sharing in rental revenue. This strategic pivot has led many on Wall Street to describe Nvidia as the central bank of AI.

CEO Jensen Huang captured the transformation succinctly: “Now, compute is revenue.”

Financial markets are preparing to assign a tradable value to that computing power.

The CME Group is set to introduce futures contracts on October 5, linked to the hourly rental rates of Nvidia’s H100 and B200 graphics processing units, pending regulatory clearance. These contracts will settle based on benchmarks from Silicon Data, which monitors chip rental prices across the industry.

Currently, Silicon Data’s H100 benchmark stands at approximately $2.68 per GPU-hour, while the newer B200 benchmark is around $5.66. Both metrics have fluctuated significantly over the past year, often independently of one another.

Consider the analogy of hotel accommodations. A room at a luxury resort and a budget motel both provide a place to sleep, but their prices differ substantially.

The same principle applies to computing resources. While the chip itself is important, factors such as the provider, geographic location, lease duration, networking capabilities, and availability all influence pricing.

A futures contract, however, does not secure physical access to the chips. Instead, it derives its value from movements in the benchmark price.

If a hotel is fully booked, rising room rates will not guarantee you a place to stay.

Wall Street’s Track Record with Emerging Markets

The DRAM initiative encountered a familiar obstacle. “The biggest hurdle was getting agreement within the industry about what the standard chip would be,” former exchange executive Charles Rose later reflected.

Weather derivatives faced a distinct challenge: individual corporate exposure often proves too specific to be captured effectively by a single standardized contract.

Bandwidth trading offers perhaps the most direct precedent. During the late-1990s fiber expansion, Enron attempted to commoditize network capacity, enabling it to trade similarly to energy. That market ultimately failed to gain traction. The resulting fiber surplus has since served as a cautionary tale for the current AI infrastructure surge, where overcapacity has already begun pressuring prices downward.

Iron ore illustrates a more successful trajectory. Following decades of private annual negotiations, the market transitioned to published daily indexes around 2009 and 2010. Futures and swaps subsequently flourished, and these indexes eventually became embedded in physical contracts as well.

That evolutionary path is precisely what the compute market would need to follow.

The U.S. Commodity Futures Trading Commission (CFTC) is not taking that progression for granted.

CFTC Chair Michael Selig has emphasized that “America cannot win the AI race without a robust derivatives market for compute.” Simultaneously, however, the agency is evaluating whether the underlying market possesses sufficient standardization and transparency to support such derivatives.

The chair envisions the racetrack being constructed. His agency remains focused on whether all participants can agree on the vehicles.

The CME still requires regulatory approval, and the CFTC’s broader public comment period extends through October 20, following the planned October 5 launch.

Competing benchmark provider Yggdrasil Financial Technologies has cautioned the CFTC that conflicts surrounding privately produced benchmarks could replicate “the LIBOR dynamic … in miniature.”

Libor once served as a global borrowing benchmark until a manipulation scandal exposed the risks inherent in relying on a rate controlled by a small group of institutions.

For investors, however, the market could prove valuable even without direct participation.

Jessica Inskip, director of investor research at StockBrokers.com, is particularly attentive to what the futures curve might reveal about expectations months into the future.

“A stock or even a commodity is two-dimensional. Price, up or down. A derivatives curve adds a third dimension: time,” she explained. “Compute can’t be stored. An idle GPU hour is gone forever.”

This characteristic could transform the curve into a fresh indicator of the AI build-out itself.

“Nvidia’s revenue and hyperscaler capex tell you what’s been booked; the compute curve tells you what’s actually being consumed and what someone will pay for it a year out,” Inskip noted.

The divergence she intends to monitor is semiconductor stocks and spending plans climbing while H100 and B200 rental prices soften, an indicator that capacity may be arriving ahead of genuine demand.

However, she would only place confidence in that signal once genuine buyers and sellers participate actively.

“Participation decides whether this is a signal or a sentiment index,” Inskip concluded. “If the open interest is all managed money, we’ve built a noisier way to be long or short Nvidia.”

Source link