Nvidia, CME, and ICE Drive the Financialization of GPU Computing Power

From NVIDIA defining GPUs as collateralizable infrastructure assets, to CME and ICE building pricing infrastructure through compute futures, to the complete derivatives ecosystem encompassing OTC swaps, on-chain bulk trading, and forward curves in prediction markets, compute is being progressively transformed into a financial asset that can be priced, traded, and pledged as collateral.

On August 10, 2026, Nvidia announced memorandums of understanding with six global leading asset management firms—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to mobilize over $500 billion in third-party capital, positioning GPUs as a mortgageable and investable infrastructure asset.

The following day, August 11, 2026, CME Group announced the launch of the world’s first futures contracts tied to GPU computing power rental rates on October 5.

Not long before this, across the Pacific, the Shanghai municipal government had already issued a document on June 2 explicitly calling for preparations for the development of compute power futures; even earlier, spot trading on the Shanghai Compute Power Trading Platform had already been operational since 2023.

When these three news threads are placed side by side, a common signal emerges: hashing power is transitioning from an IT resource used via leasing to a financial asset that can be priced, traded, hedged, and pledged on public markets. And this transformation is occurring nearly simultaneously worldwide: forward contracts have already been executed, futures are lining up for listing, indices have appeared on Bloomberg terminals, and for the first time, the term “hashing power futures” has appeared in Chinese government documents.

Based on this, this article clarifies the rapidly growing industrial chain:

• How is a GPU gradually transformed into a financial asset?
• Who is trading it, and where?
• When computing power truly becomes a commodity like oil, will it give rise to a new monetary narrative—“computing power dollar”—similar to how the petrodollar emerged half a century ago?

1.1 Jensen Huang's Ambition: Attempting to Redefine the GPU

To understand why the hash rate derivatives market will experience a surge in 2026, we must first return to the most fundamental action: Nvidia’s attempt to redefine what a GPU is.

On August 10, 2026, Nvidia’s official press release announced that it has signed memorandums of understanding with six institutions—Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to jointly establish an independent "compute financing platform," aiming to mobilize over $500 billion in third-party capital in the future to support AI data centers, chip manufacturing facilities, and associated power infrastructure.

The core of this announcement is actually Huang Renxun’s proposed shift in classification: GPUs should no longer be viewed as rapidly depreciating tech hardware, but rather as investable infrastructure assets.

In this context, GPUs generate long-term, predictable cash flows similar to commercial real estate or toll roads. Jensen Huang’s reasoning is that Nvidia’s computing assets are widely adopted by nearly all cloud providers, are fungible across different customers, and can continuously benefit from performance improvements through the CUDA software ecosystem, thereby extending their useful lifespan.

This is truly the first time that technology chips have become an investable asset class.

——Jensen Huang, August 10, 2026

Whether this argument holds depends on a central question: Can the economic lifespan of a GPU align with the term required by a traditional mortgage?

Traditional collateral—commercial real estate and cargo ships—is accepted by banks because they have mature secondary markets with decades of history and relatively stable price fluctuations. In contrast, the value curve of GPUs is entirely different, determined by the pace of chip iterations rather than physical wear.

According to data from industry research firm Silicon Data and GPU trading platform GPUSmith, the H100, which sold for approximately $40,000 at the end of 2023, has seen its secondhand market price drop to between $12,000 and $22,000 by mid-2026, with some auction prices falling as low as $8,200. This cliff-like, rather than linear, revaluation represents the greatest technical challenge in this GPU assetization experiment.

1.2 A guarantee clause that has not been fully specified

To address creditors' concerns, Nvidia has proposed a key term: it is willing to provide residual value support of up to 25% for certain financing transactions, meaning that if the actual residual value of the chips at loan maturity falls below expectations, Nvidia will cover up to 25% of the shortfall, with the exact percentage assessed on a case-by-case basis.

One detail deserves attention: According to a line-by-line review by the financial analytics platform Electron Economics of the original official press release dated August 10, the text never mentioned “residual value” or “backstop”; it only stated that the company was collaborating with six institutions to establish a large-scale special fund pool offering attractive interest rates for customers, and explicitly noted that all related collaborations remain subject to the final signing of agreements. The specific phrase “up to 25% residual value” as a support mechanism was introduced later by Jensen Huang in his follow-up interview and article.

This sequence, along with the response from the credit market, is worth considering together:

• According to ICE Data Services, Nvidia’s five-year credit default swap (CDS) spread rose in July due to reports of round-trip transactions, surging 14 basis points in a single day on July 27 to a peak of 82 basis points (later retracing), marking the largest intraday gain since the contract began active trading in November 2025.
• According to Investing.com, on the day the financing plan was announced on August 10, the spread widened to approximately 77.5 basis points. The guarantee terms did not significantly cool market sentiment.

The pricing in the above credit market, to some extent, more honestly reflects the current state of this mechanism than the announcement itself: it is still being continuously tested by the market and has not yet fully stabilized.

2.1 Why Assetization Cannot Exist Without Futures

For an asset to be accepted as eligible collateral on a bank’s balance sheet, it typically must meet three criteria: relatively stable value, a mature secondary market, and an authoritative and reliable pricing benchmark. A single physical GPU struggles to satisfy all three, which is precisely why established exchanges like CME and ICE are entering the space intensively in the first half of 2026—they aim to establish a widely recognized pricing benchmark for this emerging asset class.

On May 12, 2026, CME Group and GPU market data company Silicon Data jointly announced plans to launch the first-ever compute futures contracts this year. On August 11, the two parties further clarified that they will introduce two futures contracts on October 5: the Silicon Data H100 Rental Index Futures and the Silicon Data B200 Rental Index Futures, which will track the hourly rental index for H100 and the next-generation Blackwell B200 chips, respectively. If successfully launched, these contracts will be listed on the New York Mercantile Exchange (NYMEX). Pete Keavey, Global Head of Energy and Environmental Products at CME, compared them to crude oil futures: "Compute has become the currency of the AI era. Just as oil powered the 20th-century economy and evolved from spot trading into a global derivatives market, our futures contracts will transform compute into a standardized, tradable commodity."

Just one week later, on May 19, 2026, Intercontinental Exchange (ICE), CME’s longtime rival, announced a partnership with another compute data provider, Ornn, to launch GPU compute futures contracts based on the Ornn Compute Price Index (OCPI), covering products ranging from enterprise-grade H100 and H200 GPUs down to consumer-grade RTX 5090 cards. The nearly simultaneous entry of both exchanges underscores that the battle for pricing control is unfolding faster than externally anticipated.

2.2 Contract Design: Two Different Paths

From a contract design perspective, global computing power futures have currently diverged into two main pathways:

• One is a computing power leasing pathway, with the GPU hour rental price index as its underlying asset, essentially pricing from the supply side and hardware level;
• The other is the Token pathway, which anchors to the demand side of computing power and is priced based on the actual number of Tokens consumed by large models, better reflecting the real cost perception of downstream AI application developers and end users.

CME and ICE are pursuing the first path, while, according to financial media reports, China’s Shanghai Futures Exchange is currently developing a computing power futures scheme that follows the second path—the AI Token futures route—representing a divergence from the U.S. technological approach and a point worth continued attention.

It is worth noting that the claim that CME offers the world’s first hashrate futures requires a qualifier: it is the first hashrate futures contract to list on a regulated, mainstream exchange. The world’s first hashrate derivative was actually traded elsewhere—before CME’s official listing, several transactions had already taken place on over-the-counter and prediction market platforms.

Before the exchange’s standardized futures were officially listed, the hashpower derivatives market had already developed a complete industrial chain—from over-the-counter forwards, to on-chain bulk trading, to forward curves in prediction markets—each segment had real trading activity, not just theoretical concepts.

3.1 Without an index, there can be no derivatives.

All derivative trades analyzed below—whether FalconX’s over-the-counter swaps, Polymarket’s on-chain trades, or the upcoming futures from CME and ICE—ultimately rely on the same thing: a trusted price index.

Without an index, there is no pricing benchmark; without a pricing benchmark, there is no room for standardized derivatives.

Currently, two parallel index pathways exist in the market:

• Ornn Computing Power Price Index (OCPI): Published by computing power company Ornn, its defining feature is being the world’s first computing power index built solely on actual transaction records, rather than quoted or advertised prices. In April 2026, OCPI was launched on the Bloomberg Terminal, covering models such as H100, H200, A100, B200, and RTX 5090, and serves as the settlement benchmark for ICE computing power futures.
• Silicon Data Index: Serves as the benchmark index for CME hash rate futures, tracking daily rental prices of GPU chips in the market. CME currently plans to launch two separate contracts: the H100 Rental Index Futures and the B200 Rental Index Futures. Both contracts are priced independently based on the hourly rental rates of their respective GPU models, without converting different models into a standardized unit of computational power.

Only with price indices like WTI crude oil and Brent crude oil does computing power truly meet the basic conditions to become a commodity.

3.2 Over-the-Counter Swap: FalconX's First Trade

On May 27, 2026, digital asset broker FalconX announced the completion of the world’s first over-the-counter hash rate forward swap transaction. The counterparty was Robert Leshner, founder of the digital asset platform Superstate, with the underlying tied to the forward price of H100 in the Ornn Hash Rate Price Index (OCPI), with FalconX acting as the dealer.

Ornn CEO Kush Bavaria described it as turning a volatile, unpredictable market into a measurable, tradable, and hedgeable commodity.

The transaction amount itself is not large, but it demonstrates that institutional investors are already using over-the-counter derivatives to hedge against computing power price risks, even while the exchange's futures are still pending approval.

3.3 Polymarket: On-chain Institutional-grade Block Trading

A few days later, on June 2, 2026, the prediction market platform Polymarket announced the completion of its first institutional-grade on-chain block trade.

The counterparties to the trade are FalconX and AneraLabs, an AI-driven risk clearing startup, with settlement based on Ornn’s OCPI index, in a transaction valued at six-figure U.S. dollars, recorded on the Polygon blockchain.

This risky trade was itself structured as a prediction market position on Polymarket, which provides both a trading venue and on-chain settlement.

According to CNBC, this is also the first institutional block trade in the prediction market industry explicitly tied to hashing power pricing (H100 OCPI Index)—just a month earlier, Kalshi completed its first institutional block trade, but it was based on California carbon allowance auction prices. Together, these two events demonstrate that prediction market platforms are simultaneously enhancing their institutional-grade trading capabilities, with hashing power being one specific asset class.

This transaction includes an additional layer of on-chain execution and settlement infrastructure compared to traditional over-the-counter swaps: after FalconX and AneraLabs reach a bilateral agreement, the trade is executed through Polymarket’s international platform and ultimately recorded on the Polygon blockchain, rather than relying entirely on the back-office clearing systems of traditional financial institutions.

There is an easily overlooked distinction: Polymarket actually operates two separate, independent platforms.

• One is the international platform where this hashpower transaction takes place, built on Polygon, settled in USDC, requires no identity verification, and has geographic restrictions blocking users from the United States;
• The other is Polymarket US, launching by the end of 2025, operated by the acquired QCX entity, regulated by the CFTC, settled in USD, and requiring full identity verification.

This GPU computing power transaction occurred in the former and is outside the CFTC’s regulatory framework, which is fundamentally different from Kalshi, as discussed in the next section.

3.4 Kalshi: From Prediction Markets to Forward Curves

On July 14, 2026, Kalshi, a CFTC-regulated prediction market platform, launched an AI compute forward curve.

According to Kalshi’s official press release, this curve currently covers the Nvidia B200, H200, and A100 chips; broader compute-related contracts on the Kalshi platform also include the H100 and RTX 5090.

Kalshi’s Chief Risk Officer, who previously worked at CME Group for about 16 years, said in an interview with Bloomberg that Kalshi is using prediction markets to build a forward curve for GPU computing power, viewing it as the foundation for future futures, options, and other products; he also noted that just in 2026, capital expenditure commitments by hyperscale cloud providers have already reached the level of $500 billion to $600 billion.

One detail to note is that Kalshi explicitly states that the forward curve itself is not a tradable asset, but rather a reference price used for pricing over-the-counter swaps and structured products; the truly tradable assets are the underlying prediction market contracts on the Kalshi exchange.

The data source for this curve is also not entirely independent of the Ornn system—part of Kalshi’s computational power contracts are settled based on real-time pricing data provided by Ornn. This indicates that, although there are many participants in this产业链, the underlying price index providers are highly concentrated between Ornn and Silicon Data.

3.5 Summary and Comparison: Four Transactions, Four Roles

These four transactions are all hash rate derivatives, but each has a different positioning.

• FalconX’s over-the-counter swap and Polymarket’s on-chain block trade both employ an inter-institutional negotiated trading model, but their participation structures differ: the former involves FalconX acting as the dealer and Robert Leshner as the counterparty, with FalconX effectively market-making to facilitate Leshner’s hedging transaction; the latter features FalconX and AneraLabs as mutual counterparties, with Polymarket providing the on-chain trading infrastructure (trade execution functionality) and settling the transaction on Polygon.
• Kalshi’s forward curve is entirely different—it is not a contract itself, but rather a reference price derived by aggregating the prices of numerous real, small-scale prediction contracts traded on the platform. What you can actually buy and sell are the underlying contracts themselves; the curve simply connects their prices into a single line.
• The futures offered by CME and ICE are closest to traditional commodity futures, featuring standardized terms, exchange listing, and centralized clearing, similar to the structure of crude oil futures.

Taken together, these elements illustrate a clear division of labor across the supply chain: OTC forwards absorb risk management demands during gaps in exchange approvals, prediction markets use high-frequency, small-volume trades to establish a widely accepted market price, and exchange futures standardize this price into tradable contracts accessible to all.

4.1 Shanghai: From Pilot Operation to Inclusion in Government Documents

Few have noticed that China began building infrastructure for spot computing power trading well before the recent wave of futures enthusiasm in the U.S., and it did so along multiple parallel tracks.

At the platform level, Shanghai has taken the lead. The Shanghai Computing Power Trading Platform launched its pilot operation in April 2023, rolled out version 2.0 in December of the same year, and was upgraded in May 2025 to the China Computing Power Platform (Shanghai). However, this is merely a local node; national-level expansion is also gradually underway: the China Computing Power Platform, led by the China Academy of Information and Communications Technology, achieved full connectivity at the China Computing Power Conference in August 2025, with ten provincial and municipal platforms—Shanxi, Liaoning, Shanghai, Jiangsu, Zhejiang, Shandong, Henan, Qinghai, Ningxia, and Xinjiang—successfully integrated. Shanghai is one node within this national network, not the sole entity behind this achievement.

At the pricing benchmark level, China Securities Index Company released the CSI Intelligent Computing Power Supply Index Series on December 24, 2025, comprising 16 indices, including one total index, seven regional indices, and eight national computing power hub node indices.

At the national policy level, the turning point occurred later and unfolded more clearly in two distinct steps. In April 2026, the Ministry of Industry and Information Technology first proposed exploring innovative business models such as computing power banks and computing power supermarkets, marking a conceptual loosening.

The real breakthrough came at the end of May: On May 28, the General Office of the Shanghai Municipal Government issued the "Several Opinions on Deepening the Construction of Shanghai as a Global Asset Management Center," which was publicly released on June 2. The document explicitly included preparing for the research and development of power futures and computing power futures. — Thus, the term "computing power futures" appeared for the first time in an official public document issued by a provincial-level government in China.

Prepare for the development of power futures and computing power futures, and create more new futures products that represent the direction of advanced productive forces. — Shanghai Municipal Government, "Several Opinions on Deepening the Construction of Shanghai as a Global Asset Management Center," June 2, 2026

An interesting point is that the document uses the term “research and development preparation” rather than “upcoming listing.” This reflects a noticeably more cautious and earlier-stage timeline compared to specific listing dates like CME’s October 5. The Shanghai document also sets a broader goal: aiming to reach RMB 55 trillion in asset management scale by 2030, accounting for one-third of the national total.

4.2 A previously rarely mentioned technical roadmap fork

According to a Reuters exclusive report, the Shanghai Futures Exchange is developing a futures product linked to AI tokens, with the underlying asset being the smallest unit of information processed by large models—used for pricing AI services—differing entirely from the GPU rental-by-the-hour model adopted by CME and ICE. The report also notes that this research is still in its early stages: it has not disclosed whether pricing will be based on the number of tokens or the price per token, nor has it outlined a timeline for regulatory approval.

Even so, the two paths for computing power futures in the U.S. and China diverged from the outset: the U.S. anchors its pricing to the rental cost of hardware supply, while China explores pricing models tied to demand-side and Token consumption.

This fork is easy to understand—CME and ICE primarily serve data center operators and cloud service providers, who care about the hourly rental cost of a single card; token pricing, on the other hand, better reflects the real cost perception of downstream AI application developers, who care about how many tokens are consumed and how much it costs to complete a task.

It remains unclear which path is superior, but if China successfully implements token futures, it would unlock a downstream application market far larger than the card-selling and rental model, with significant potential for product innovation—such as transferability and convenience—though substantial challenges remain in areas like standardization.

Once computing power truly begins to be priced, traded, and pledged like oil, a natural question arises: Since oil trade has been uniformly priced and settled in U.S. dollars, giving rise to the petrodollar system that has lasted for half a century, could computing power follow the same path?

5.1 Proposal of the "Hash Power Dollar" Concept

This analogy is not original to this article. Over the past six months, multiple institutional research bodies have proposed similar frameworks, with one of the most influential being the commentary article “Turning the AI Revolution into Dollar Dominance,” published on December 8, 2025, by Navin Girishankar, Head of the Economic Security and Technology Division at the Center for Strategic and International Studies (CSIS).

The core idea of the article is that the United States is exporting advanced AI chips to allies and partners to help them build computing infrastructure, which will continuously generate AI services exported globally. Whoever controls the settlement currency for these AI service exports will replicate the status once held by the petrodollar. (Note: This is a commentary piece reflecting the author’s personal views; CSIS officially states that its research is nonpartisan and does not represent the institution’s position, nor is it U.S. government policy.)

5.2 Core Flaw in the Concept: Hashrate Has Not Been Strongly Tied to the US Dollar

The biggest flaw in this concept, as the author themselves pointed out, is that these chip export agreements make no requirement for recipient countries to settle their AI service export revenues in U.S. dollars.

He outlined a calculation logic: a country might spend only $10 billion once to build data center infrastructure, yet the chips installed could subsequently generate annual AI service export revenues of $50 to $100 billion—there are currently no constraints on what currency these ongoing, long-term revenues would be settled in. The petrodollar system relies on implicit pricing agreements; the "computing dollar" hasn't even taken this first step yet.

To address this vulnerability, the author proposes three policy recommendations:

• Link the chip export license to a commitment to settle in U.S. dollars or USD-stablecoins;
• The GENIUS Act was signed into law in July 2025, authorizing USD-backed stablecoins as a payment and settlement instrument.
• Provide an economic security umbrella as the modern counterpart to the defense umbrella of the Cold War era.

Based on this, the Hash Power Dollar is more accurately positioned as a policy proposal outlined in a think tank commentary article, rather than a fully built and operational monetary system.

5.3 Historical Reference: The Cycle from Proposal to Validation of New Narratives

It typically takes a full cycle for a new asset narrative to move from proposal to genuine market validation. The two bubbles around the year 2000 serve as useful references: one is the well-known internet stock bubble; the other is the telecommunications equipment bubble, which structurally resembles today’s compute narrative—companies like Cisco, Lucent, and Nortel provided large-scale vendor financing to cash-strapped telecom operators and internet service providers, enabling customers to purchase equipment, creating a self-reinforcing loop where loans drove purchases, and purchases pushed stock prices higher.

The entire industry expanded to over $1.5 trillion before the bubble burst, with approximately $1 trillion in accumulated debt and more than $500 billion spent on infrastructure such as fiber optic cables. For example, in the year 2000 alone, Nortel Networks completed $19.7 billion in acquisitions, nearly all paid for with stock.

More interestingly, someone once tried to turn bandwidth into a tradable commodity—and failed. In December 1999, Enron completed the first-ever bandwidth trade, targeting a monthly incremental DS-3 contract on Global Crossing’s network between New York and Los Angeles, explicitly aiming to establish this contract as a benchmark and proposing a series of subsequent expansion plans. This approach is identical to the methodology used today by Ornn and Silicon Data to create indices, and by CME to develop standardized contracts.

The outcome was: the bandwidth market collapsed before achieving sufficient liquidity; following Enron’s 2001 bankruptcy, subsequent players such as Williams, Dynegy, and El Paso shut down their bandwidth trading divisions, and independent bandwidth trading platforms like RateXchange exited entirely.

Viewing these two bubbles together reveals the same truth: just because a story is told early doesn’t mean the demand is fake. Internet traffic did indeed grow as expected, and the overbuilding of fiber optics actually accelerated the decline in costs for later broadband, video, and cloud computing services. The bubble’s burst punished those who used leverage and saw stock prices outpace real demand—not the underlying direction itself. Today, the demand signals for AI computing power are equally strong, though whether they will translate into sustainable profitability remains uncertain. Moreover, historical regulatory reforms emerged only after bubbles burst; now, institutions like the U.S. CFTC and the Bank of England are already proactively monitoring the space—a stark contrast to the nearly nonexistent regulation before 2000.

The rapid development of产业链 such as computing power trading does not mean the system has been fully validated. Long-term transactions, index launches, and futures awaiting listing may create the impression that computing power assetization is already a done deal. However, whether this system can truly stand depends on three unanswered questions: Can demand support the leverage? Can regulation keep pace with innovation? And will different pathways ultimately converge into a single standard?

Whether computing power becomes assetized will repeat the 2000 internet finance bubble depends on several unclear variables: whether the residual value assumptions for GPUs can withstand the impact of each new generation of chips, whether computing power denominated in USD can truly be encoded into protocol terms, and whether China’s token futures and U.S. GPU mini-futures will ultimately converge on the same standards.

Computing power is being transformed into a financial asset, and this direction is likely irreversible. As for who will ultimately price this new system, which currency will be used for settlement, and what standards will govern its operation—the answers are still on the way.

This is a new story.