Wall Street turns Nvidia AI chips into a new futures market

Starting October 5, 2026, Wall Street will, for the first time, get regulated futures that let companies and investors trade on the future rental price of Nvidia’s AI chips. CME Group plans to launch contracts for the Nvidia H100 and Blackwell B200, signaling how AI compute is shifting from pure infrastructure to a tradable economic commodity.

The move lands as multi-billion-dollar data center buildouts make GPU costs a critical variable for AI firms. CME Group says the contracts still depend on completing required regulatory approvals.

The products are officially called Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures. They track Silicon Data’s benchmarks for the hourly rental price of Nvidia GPUs.

Each contract represents a month of rental costs for the H100 or B200, according to CME. Listed under NYMEX rules, they target AI developers, cloud providers, and institutional investors.

Crucially, buyers don’t receive a Nvidia chip nor do they reserve actual GPU capacity. These are cash-settled instruments tied to a price index.

Yahoo Finance reported Saturday that Silicon Data’s benchmark was about $2.68 per GPU-hour for the H100 and roughly $5.66 for the B200. Prices can swing sharply—and not always in the same direction.

For AI companies, GPU capacity has become a major, volatile, and hard-to-forecast expense. Teams training large models or serving millions of AI requests often rent massive compute blocks from hyperscalers and specialized GPU clouds.

Rental rates depend on more than the chip. Availability, region, cloud provider, networking, and contract length all shape the final price.

Then there’s volatility. Silicon Data’s historical indices show H100 rental prices fell 45.3% from the start of its index to February 2025 before rebounding. The B200 saw nearly a 20% drop from a prior peak, followed by a rebound of more than 40%.

For companies planning billions in AI infrastructure, such swings make future costs harder to predict.

Futures can partially hedge that risk. An AI firm expecting heavy H100 usage in a few months can insure against rising rental prices. A GPU cloud provider can hedge the opposite—falling prices.

CME Group is drawing a direct parallel to commodities markets. Pete Keavey, who oversees energy and environmental products at CME, calls compute the “currency of the AI age.” A standardized futures market, CME argues, can help businesses manage price risk around AI infrastructure.

The analogy to oil isn’t perfect. You can’t store a GPU-hour and deliver it later. Data centers also vary widely by location, networks, availability, and performance.

Still, the same economic need is emerging: companies want visibility today into what a critical input will cost tomorrow.

That gives compute a new financial layer. GPU capacity is no longer just a technical resource bought from a cloud provider—it now carries a tradable price signal on top of the physical infrastructure.

A liquid futures market could do more than hedge risk. Forward prices may become a new barometer for expectations across the AI economy.

Silicon Data already publishes forward curves for multiple GPU generations. In July, those prices pointed to lower long-term rates for chips like the H100 and B200, with 36‑month rates below then-current spot prices.

That tracks with fast-moving hardware cycles. New chips typically deliver more performance, while data center buildouts expand supply. An H100 can remain technically strong even as its economic value gets squeezed by newer generations.

A mature futures market would make those expectations more visible. Rising forward prices for GPU capacity could signal looming scarcity or surging AI demand. Falling futures might point to more supply, more efficient models, or faster tech obsolescence.

For Nvidia, the twist is notable: its hardware effectively becomes the underlying reference for a new class of financial products.

That fits Nvidia’s exceptional standing in the AI economy. The company last week reported quarterly revenue of $96.22 billion, with roughly $89 billion coming from its data center business. It expects about $108 billion in revenue this quarter.

The new futures don’t give Nvidia direct control over the market. The contracts are offered by CME and track independent rental price benchmarks from Silicon Data.

But focusing specifically on the H100 and B200 underscores how Nvidia’s hardware has become a de facto economic benchmark.

The biggest shift: the price of AI compute becomes more measurable, comparable, and hedgeable. That’s especially relevant for companies locking in large GPU capacity over longer periods.

For AI startups, a more transparent market helps forecast infrastructure costs. Cloud providers may get new ways to hedge pricing risk. Investors, meanwhile, gain a tool to trade directly on expectations for AI compute prices—without buying Nvidia stock.

CME has already tuned its systems for the new category. The exchange is even introducing a new unit of account: GPU-H, or GPU-hours. Pending regulatory review, contracts are slated to go live on CME Globex and via CME ClearPort starting October 5.

It’s a striking new chapter in the AI boom. After chips, data centers, energy contracts, and multibillion-dollar funding rounds, there’s now a financial market for the price of compute itself.

If enough players participate, GPU pricing could evolve into a key economic barometer for AI—not just how many Nvidia chips companies buy, but what the market believes an hour of AI compute will be worth months or years from now.