When computing power itself becomes a tradable asset: From compute futures to Compute Dollar

From Nvidia defining GPUs as collateralizable infrastructure assets, to CME and ICE building pricing infrastructure like compute futures, to a complete derivatives industry chain encompassing OTC swaps, on-chain block trades, and prediction market forward curves, compute power is being progressively transformed into a financial asset that can be priced, traded, and collateralized.

On August 10, 2026, Nvidia announced it had signed memorandums of understanding with six of the world's top asset management institutions—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—aiming to mobilize over $500 billion in third-party capital and define GPUs as a collateralizable, investable infrastructure asset class.

The next day, on August 11, 2026, CME Group announced it would launch the world's first futures contracts linked to GPU compute rental prices on October 5.

Not long before this, across the Pacific, the Shanghai municipal government had already taken the lead on June 2, issuing a document that explicitly called for preparatory research on compute futures; and even earlier, the spot trading on the Shanghai Compute Trading Platform had been operational since 2023.

When these three news lines are placed side by side, a common signal emerges: compute power is shifting from being an IT resource accessed through leasing to a financial asset that can be priced, traded, hedged, and collateralized on public markets. Moreover, this shift is happening almost simultaneously on a global scale: forwards have already been executed, futures are queuing up for listing, indices have already appeared on Bloomberg terminals, and for the first time, the term "compute futures" has appeared in Chinese government documents.

With that in mind, this article lays out the rapidly growing industry chain clearly:

• How exactly have GPUs been transformed into financial assets?
• Who is trading them, and where?
• If compute power truly becomes a commodity like oil, will it give rise to a new monetary narrative—"petrocompute" or a "compute dollar"—just as petrodollars emerged half a century ago?

1. The Starting Point: How GPUs Were Transformed into Collateral

1.1 Jensen Huang's Ambition: Redefining the GPU

To understand why the compute derivatives market erupted in 2026, we must 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 had signed memorandums of understanding with six institutions—Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to jointly build an independent "compute financing platform," with the goal of mobilizing over $500 billion in third-party capital to support the construction of AI data centers, chip factories, and supporting power infrastructure.

At the core of this announcement is a taxonomic shift proposed by Jensen Huang: GPUs should no longer be viewed as rapidly depreciating technology hardware, but rather as investable infrastructure assets.

In this context, GPUs possess long-term, predictable cash flows similar to commercial real estate or toll roads. Huang's rationale: Nvidia's compute assets are widely adopted by virtually all cloud service providers, can be transferred among different customers (fungible), and can continuously gain performance improvements through the CUDA software ecosystem, thereby extending their useful life.

"This is truly the first time a tech chip has become an investable asset class."

— Jensen Huang, August 10, 2026

The core controversy over whether this narrative holds lies in one question: Can the economic lifespan of a GPU match the maturities required by traditional collateralized lending?

Traditional collateral—commercial real estate, cargo ships—is accepted by banks because these assets have mature secondary markets spanning decades, with relatively moderate price fluctuations. GPU value curves are completely different, driven by chip iteration cycles rather than physical depreciation.

According to data from industry research firm Silicon Data and GPU trading platform GPUSmith, an H100 that sold for approximately $40,000 in late 2023 had fallen to a resale price range of $12,000 to $22,000 by mid-2026, with some auction prices dropping as low as $8,200. This cliff-edge rather than linear value reset is the biggest technical challenge in the GPU assetization experiment.

1.2 A Guarantee Clause Not Yet Fully Written

To alleviate creditors' concerns, Nvidia proposed a key provision: a willingness to provide up to 25% residual value support for certain financing transactions—meaning that if a chip's actual residual value falls below expectations at loan maturity, Nvidia would cover up to 25% of the shortfall, with the exact percentage assessed on a case-by-case basis.

One detail worth noting: according to a sentence-by-sentence review of the official August 10 press release by financial analytics platform Electron Economics, the terms "residual value" or "backstop" did not actually appear in the original announcement. It merely stated cooperation with six institutions to establish a large-scale dedicated capital pool with attractive interest rates for customers, explicitly noting that the cooperation remains subject to final agreement execution. The specific articulation of a "up to 25% residual value" support mechanism came from Huang's supplementary comments in interviews and articles the following day.

This sequence of events, combined with credit market reactions, is worth examining together:

• According to ICE Data Services, Nvidia's five-year credit default swap (CDS) spread widened in July amid reports of circular trading, jumping 14 basis points in a single day on July 27 to reach 82 basis points (before subsequently pulling back)—the largest single-day intraday gain for the contract since it became actively traded in November 2025;
• Investing.com analysis shows that on August 10, the day of the financing plan announcement, the spread widened to approximately 77.5 basis points. The guarantee clause did not significantly cool market sentiment.

The pricing in the credit market above is, in some ways, a more honest reflection of where this mechanism currently stands than the announcement itself: still being continuously tested by the market, not yet fully finalized.

2. Collateral Needs a Yardstick: Exchange Entry

2.1 Why Assetization Cannot Happen Without Futures

For an asset to truly become acceptable collateral on bank balance sheets, it typically needs to satisfy three conditions: relatively stable value, an established secondary market, and an authoritative, credible pricing benchmark. A single physical GPU struggles to satisfy all three simultaneously—which is precisely why established exchanges like CME and ICE moved in rapidly during the first half of 2026. Their goal: to establish a 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 batch of compute futures contracts within the year. On August 11, the two parties further clarified that they would list two futures contracts on October 5—the Silicon Data H100 Rental Index Futures and the Silicon Data B200 Rental Index Futures—tracking hourly rental indices for the H100 and the next-generation Blackwell B200 chips respectively. If successfully launched, they would be listed on the New York Mercantile Exchange (NYMEX). Pete Keavey, CME Group's Global Head of Energy and Environmental Products, drew an analogy 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, CME's rival Intercontinental Exchange (ICE) announced it would partner with another compute data provider, Ornn, to launch GPU compute futures based on the Ornn Compute Price Index (OCPI), covering everything from enterprise-grade H100 and H200 to the consumer-grade RTX 5090. The near-simultaneous entry of the two major exchanges indicates that the battle for pricing power is moving faster than outsiders expected.

2.2 Contract Design: Two Divergent Paths

From a contract design perspective, global compute futures are currently diverging along two main paths:

• One is the compute rental path, benchmarked to GPU hourly rental price indices, essentially pricing from the supply side at the hardware level;
• The other is the Token path, anchored to the demand side of compute, priced according to the actual number of Tokens consumed by large language models, closer to the real cost experience of downstream AI application developers and end users.

Both CME and ICE follow the former path, while according to financial media reports, the compute futures scheme currently under development at China's Shanghai Futures Exchange follows the latter—the AI Token futures route. This divergence from the US technical approach is a fork worth continuous attention.

It's worth noting that the claim of CME launching the world's first compute futures requires a qualifier: it is the first to enter the regulated mainstream exchange track. The world's first compute derivative actually belongs to someone else—before CME's official listing, OTC markets and prediction market platforms had already completed several transactions.

3. Beyond Exchanges: An Industry Chain Integrated with On-Chain Finance

Before standardized exchange futures officially listed, the compute derivatives market had already developed an entire industry chain—from OTC forwards to on-chain block trades to prediction market forward curves—with real transaction records at every stage, not just停留在 conceptual stages.

3.1 No Index, No Derivatives

All the derivatives transactions to be analyzed below—whether FalconX's OTC swaps, Polymarket's on-chain trading, or the upcoming CME and ICE futures—ultimately depend on the same thing: a credible price index.

Without an index, there is no pricing benchmark; without a pricing benchmark, there is no room for any standardized derivative to exist.

Currently, two index tracks operate in parallel in the market:

• Ornn Compute Price Index (OCPI): Published by compute company Ornn, its defining feature is being the world's first compute index constructed solely from actual transaction records, rather than quotes or advertised prices. In April 2026, OCPI went live on Bloomberg terminals, covering models including H100, H200, A100, B200, and RTX 5090, and serves as the settlement benchmark for ICE compute futures;
• Silicon Data Index: Provides the index benchmark for CME compute futures, tracking daily rental prices of GPU chips in the market. CME currently plans to list H100 Rental Index Futures and B200 Rental Index Futures as two separate contracts, with each priced independently based on the hourly rental price of the corresponding GPU model, without converting different models into a unified standardized compute unit.

With price indices like WTI crude or Brent crude, compute power has truly acquired the fundamental conditions to become a commodity.

3.2 OTC Swaps: FalconX's First Trade

On May 27, 2026, digital asset broker FalconX announced the completion of the world's first OTC compute forward price swap transaction. The counterparty was Robert Leshner, founder of digital asset platform Superstate, with the underlying linked to the forward price of H100 in the Ornn Compute Price Index (OCPI). FalconX acted as the dealer.

Ornn CEO Kush Bavaria described it as transforming a volatile, unpredictable market into a commodity that can be measured, traded, and hedged.

The transaction amount itself was not large, but it proved one thing: while exchange futures were still awaiting approval, institutional investors had already begun positioning for compute price risk management through OTC derivatives.

3.3 Polymarket: On-Chain Institutional Block Trades

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

The two parties were FalconX and AneraLabs, an AI risk clearing house startup, with the underlying settled against Ornn's OCPI index. The transaction size reached six figures in US dollars, with settlement records on the Polygon blockchain.

This risk transaction itself was structured as a prediction market position on Polymarket, with Polymarket providing both the trading venue and on-chain settlement mechanism.

According to CNBC, this was also the first institutional block trade in the prediction market industry explicitly linked to compute prices (the H100 OCPI index)—one month earlier, Kalshi had completed its first institutional block trade, but with California carbon allowance auction prices as the underlying. Viewed together, these events demonstrate that prediction market platforms are simultaneously building out institutional-grade trading capabilities, with compute as one specific category.

Compared to traditional OTC swaps, this transaction added a layer of on-chain execution and settlement infrastructure: after FalconX and AneraLabs completed bilateral negotiations, the trade was executed through Polymarket's international platform and ultimately recorded on the Polygon blockchain, rather than relying entirely on traditional financial institutions' back-office clearing systems.

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

• One is the international site where this compute trade took place, built on Polygon, settled in USDC, requiring no identity verification, and geo-blocked for US users;
• The other is Polymarket US, launched in late 2025, operated by the acquired QCX entity, regulated by the CFTC, settled in US dollars, and requiring full identity verification.

This GPU compute trade occurred on the former, outside the CFTC regulatory framework—a fundamental difference from Kalshi, which will be discussed in the next section.

3.4 Kalshi: From Prediction Market to Forward Curve

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

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

Kalshi's Chief Risk Officer, who previously spent approximately 16 years at CME Group, told Bloomberg in an interview that Kalshi is using prediction markets to build a forward curve for GPU compute, viewing it as the foundation for future products such as futures and options. He also noted that hyperscale cloud service providers' capital expenditure commitments for 2026 alone had already reached the $500–600 billion level.

One detail worth clarifying: Kalshi has explicitly stated that the forward curve itself is not a tradable asset, but rather a reference price used as a pricing basis for OTC swaps and structured products. What is actually tradable are the underlying prediction market contracts on the Kalshi exchange itself.

The data sources for this curve are also not entirely independent of the Ornn ecosystem—some of Kalshi's compute contracts are likewise settled against real-time pricing data provided by Ornn. This indicates that while the current industry chain has many participants, the underlying price index providers are highly concentrated between Ornn and Silicon Data.

3.5 Summary Comparison: Four Transactions, Four Roles

All four transactions are compute derivatives, but their positioning differs.

• FalconX's OTC swap and Polymarket's on-chain block trade both took the form of institutional negotiated transactions, but their participation structures differ: in the former, FalconX acted as dealer with Robert Leshner as counterparty—effectively FalconX market-making to complete a hedging transaction for Leshner; in the latter, FalconX and AneraLabs were mutual counterparties, with Polymarket providing the on-chain trading infrastructure (execution function) and settling on-chain via Polygon.
• Kalshi's forward curve is an entirely different matter—it's not a contract, but a reference price reverse-engineered from a large number of real small-value prediction contracts on the platform. What can actually be bought and sold are those underlying contracts themselves; the curve merely stitches their prices into a line.
• CME and ICE futures are closest to traditional commodity futures—standardized terms, exchange listing, central clearing, structurally identical to crude oil futures.

Viewed together, a clear industry chain division of labor emerges: OTC forwards catch risk management demand during the exchange approval gap; prediction markets use high-frequency small-value trading to probe out a market-recognized price level; exchange futures standardize that price into contracts anyone can place orders on.

4. The China Side: Earlier on Spot, Slower on Futures

4.1