CFTC Explores AI Compute Futures as CME Plans Launch in 2026

The true significance of hash power futures may not be creating a new trading instrument, but rather establishing the first "resource price map" for the AI era.

Author and source: 0x9999in1, ME News



• The U.S. Commodity Futures Trading Commission (CFTC) has begun soliciting public comments on AI computing power derivatives, signaling that the financialization of computing power has moved from corporate experimentation into the realm of regulatory exploration.
• CME plans to launch its first AI computing power futures products on October 5, 2026, with underlying assets being the rental prices of NVIDIA H100 and B200 GPUs, subject to regulatory review.
• The core value of compute futures is not to enable investors to "trade GPUs," but to allow AI companies to proactively manage their future compute costs, transforming compute from an implicit infrastructure expense into a transparent, priced commodity.
• However, computing power is not oil; GPUs cannot be stored, transported, or settled uniformly like crude oil, and price standardization, liquidity, and index reliability remain the greatest challenges.
• The U.S. approach focuses on GPU rental fees, while China-related explorations emphasize the token value consumed by AI services—two paths representing distinct pricing logics for the AI economy.
• The true significance of hash power futures may not be creating a new trading instrument, but rather establishing the first "resource price map" for the AI era.

Today, as AI has evolved, a growing truth is emerging: the core of future competition is not just model capability or algorithmic innovation, but who can secure cheaper, more stable, and larger-scale computing resources.

Over the past few decades, human economies have developed mature pricing systems centered around energy. Oil has futures contracts, natural gas has forward agreements, and metals have trading markets. Companies understand approximately where future costs will stand and can use financial instruments to manage volatility.

Yet, the most critical resource in the AI era—computing power—has long been in an awkward state.

A company looking to train large models needs to purchase or lease a large number of GPUs. The problem is that the same NVIDIA H100 GPU can have vastly different prices depending on the cloud provider, region, and contract duration. While companies know how much they are paying, it is difficult for them to determine the true market price.

What does this mean?

This means the AI industry is expanding rapidly, but its most important production asset lacks a mature public pricing system.

The gap is precisely what U.S. regulators are now focusing on.

The CFTC is seeking public comment on AI computing power derivatives not merely to approve a new financial product, but to explore a more fundamental question: whether computing power has met the conditions to become a financial commodity. Regulators are closely examining whether price indices are susceptible to manipulation, whether the market has sufficient liquidity, and whether ordinary investors could face risks upon participation. (Reuters)

This step is similar to the historical development of the energy market.

Oil did not naturally become a global financial asset. It went through dispersed production, chaotic pricing, and a lack of trading standards before gradually developing into global pricing hubs such as New York and London.

AI computing power is currently undergoing a similar phase.

The difference is that oil powered 20th-century industry, while computing power may determine the foundational efficiency of 21st-century intelligent industries.

The first products planned by CME are actually more "traditional" than many people imagine.

It is not about trading physical GPUs or having investors buy batches of H100 chips in anticipation of appreciation.

According to CME's announcement, the initial products include the Silicon Data H100 Rental Index Futures and the Silicon Data B200 Rental Index Futures, which track hourly rental price indices for two types of GPUs, with each contract representing one month's rental cost and settled in cash. (CME Group Inc.)

In simple terms:

If AI companies are concerned about rising GPU rental prices in the future, they can hedge their costs by purchasing futures contracts in advance; similarly, cloud computing companies worried about falling rental rates in the future can also use futures to manage risk.

This logic is not unfamiliar.

Airlines manage oil price risk through fuel futures; manufacturers manage raw material costs through metal futures. So why can’t AI companies manage GPU costs through compute futures?

Especially against the backdrop of rapid expansion in AI infrastructure investment, this demand is becoming increasingly realistic.

Over the past few years, global tech giants have consistently increased their AI capital expenditures. Companies such as Microsoft, Google, Amazon, and Meta have continued investing in data centers, GPU clusters, and AI infrastructure. Computing power is no longer merely an issue for technology departments—it has gradually become a cost variable that finance departments must manage.

If an AI company needs to invest billions of dollars annually in computing power, a few percentage points in computing power price fluctuations could impact its profit model.

Financial markets are best suited to address this kind of "uncertainty."

So, what CME really wants to trade is not GPUs, but the future cost expectations of the AI industry.

If compute futures represent the AI industry's maturation, they also face a very real challenge:

Can computing power really be standardized like oil?

The answer is not simple.

Although oil comes from different sources, it can be standardized through quality grades, delivery locations, and other factors. GPU computing power, however, is more complex.

The value of an H100 depends not only on the chip itself but also on the region, electricity costs, network conditions, server configuration, software environment, and the cloud provider's service capabilities.

One hour of H100 computing time in a data center in the Eastern United States may not be exactly the same as one hour in a data center in a region in Asia.

More importantly, computing power has a characteristic that traditional commodities do not: it cannot be stored.

The GPU capacity you idle today cannot be resold tomorrow.

A barrel of oil can be stored in a tank and waited for price increases, but unused computing power for an hour may lose its value entirely.

This will make the pricing of hash rate futures more complex.

In addition, the current GPU rental market remains highly fragmented. There is no standardized pricing across different providers, and a large volume of transactions occur through private corporate negotiations, resulting in limited public market data.

Therefore, the questions raised by the CFTC are not obstacles to innovation, but rather the questions that must be answered when financial markets enter new territories:

Where does the price come from?

Who provides liquidity?

How can you prevent a small number of participants from influencing the index?

If these issues cannot be resolved, hash rate futures may remain just a conceptually appealing but poorly traded product.

It is worth noting that the financialization of AI computing power is not being explored by the United States alone.

The Chinese market is also researching AI-related financial products; for example, the Shanghai Futures Exchange is exploring AI token futures, aiming to focus on the consumption and utilitarian value of AI services.

The main difference between the two is:

The U.S. is more focused on the prices of means of production, namely GPU and computing resource costs.

China's exploration is more aligned with "consumption-side" value, namely AI service usage and token consumption.

This actually reflects two different understandings.

The U.S. believes that the AI industry is first and foremost a competition over infrastructure. Whoever controls computing power resources holds the pricing power at the foundational level of the industry.

China's exploration focuses more on the demand value generated by AI applications, aiming to capture economic activities driven by model invocation and service consumption.

It is still unclear which path will ultimately succeed.

Because the AI economy is still in its early stages.

In the internet era, people once believed bandwidth would become the core resource; in the mobile internet era, attention shifted to data traffic; today, all eyes are on computing power.

But history shows that the most important resources ultimately go through a process: from scarcity, to standardization, to financialization.

Many people, upon seeing CME launch hash rate futures, assume it's merely Wall Street creating another new trading product.

But more importantly, it signifies that the AI industry is transitioning from a technology race to an industry-wide competition.

Early AI competition focused on model parameters, algorithm efficiency, and talent reserves.

The next phase will be about energy, chips, data centers, capital efficiency, and how to manage these resources.

When hashing power begins to have a public price, when companies can see their future cost curves, and when investors can assess the return on AI infrastructure, the entire industry will enter a more transparent phase.

Of course, this does not mean that hash rate futures will necessarily succeed.

Financial markets do not automatically thrive simply because of a popular concept. Many so-called "future assets" in the past have failed due to a lack of real demand.

Whether computing power futures can become a critical infrastructure in the AI era depends on three conditions:

First, do AI companies truly need risk management tools?

Second, whether the market can generate sufficient trading volume;

Third, can regulators establish a trustworthy and fair pricing mechanism?

But regardless of the outcome, the fact that U.S. regulators have begun examining hash rate derivatives itself sends a very clear signal:

AI is moving from laboratories into economic systems.

In the past, discussions about AI always focused on how powerful the models were and how many parameters they had.

In the future, people may discuss another issue more frequently:

How much is one hour of intelligence really worth?

AI truly gains its own economic coordinates when this question can be answered by the market.

1. U.S. Commodity Futures Trading Commission, Public Comments and Regulatory Materials, 2026.
2. CME Group, “CME Group and Silicon Data to Launch Compute Futures on October 5 to Unlock New Way to Hedge AI Risks,” August 2026.
3. CME Group, "Compute Futures," 2026.
4. Reuters, “US CFTC Seeks Comment on Compute Derivatives as AI Demand Grows,” August 2026.
5. Financial Times, “CME plans to launch futures market for AI computing power”, 2026.
6. Barron’s, “Bleeding-Edge AI Meets Cutting-Edge Finance. Is This Peak AI?”, 2026.