Wen Hongjun Stablecoin New Finance-25: Hashrate Tokenization and Hashrate USD Stablecoin

The Battle for AI Dominance Extends to Computing Power Pricing

The intensifying competition between the United States and China for artificial intelligence (AI) leadership has now extended to the pricing power of computing resources. While the Chicago Mercantile Exchange’s (CME) GPU rental futures await review by the Commodity Futures Trading Commission (CFTC), perpetual contracts for H100 rentals, settled in USDC, have already emerged on-chain. Financial commentator Wen Hongjun believes that computing power will become the third anchor for the U.S. dollar, following gold and oil. By layering measurement, asset, and settlement, a computing power-backed stablecoin will naturally form, and Taiwan should strive to secure settlement rights.

From Gold to Oil, and Now to Computing Power

In a previous discussion, we explored how U.S. dollar stablecoins are becoming an API for the banking system. This article delves deeper into a more significant question: if stablecoins are the new generation of the dollar, what will they be pegged to?

The answer is “computing power.”

At a White House luncheon in late September, the opening remark of a speech by former President Trump was: “Whoever wins superintelligence wins.” This statement succinctly captures the essence of technological competition: the U.S. and China are not just competing on the intelligence of their models, but on who can convert “electricity” into “intelligence” faster and more affordably.

Simultaneously, the AI agent economy is rapidly taking shape. Stripe is reportedly acquiring AI model routing platform OpenRouter for over $7 billion. On the same day, Ramp launched Router.com, citing data from Ramp’s corporate cards showing a 20.7-fold increase in enterprise AI spending since June 2025.

Two of the most astute fintech companies have reached the same conclusion: “helping enterprises buy intelligence” is itself a financial services business. Intelligence can be bought and sold, making the underlying computing power the most critical means of production. And all means of production are ultimately financialized.

According to the latest financial forecasts from major tech companies, Microsoft’s capital expenditures are projected to be around $175 billion in 2026, Amazon’s $220 billion, Alphabet’s $195 billion to $205 billion, Meta’s $130 billion to $145 billion, and Oracle’s approximately $70 billion. Combined, these five companies will spend close to $800 billion. Goldman Sachs estimates that total capital expenditure for computing power, data centers, and electricity between 2026 and 2031 will be around $7.6 trillion in a baseline scenario. Tomasz Tunguz estimates that data center construction will require $4 trillion in debt within five years, equivalent to a 34% expansion of the U.S. corporate bond market, exceeding the entire global private credit market.

SemiAnalysis estimates that the basic annual operating cost per MW of computing power is between $10 million and $15 million, while Anthropic’s latest generation of top-tier models can generate up to $50 million per MW. This is a genuinely profitable business, not just a speculative narrative.

Capital expenditures by hyperscale cloud providers are expected to surpass operating cash flow this quarter, indicating a funding shortfall. More critically, Bank of America believes that approximately 0.3 percentage points of the increase in the 10-year U.S. Treasury yield this year are attributable to AI-driven borrowing demand. Remember the 5.3% yield on 10-year Treasuries from the previous discussion? Bonds issued by AI companies are competing for the same pool of money as the U.S. government.

This is where the business opportunity in the financialization of computing power lies: whoever can provide pricing, hedging, financing, and settlement for this industry will capture a share of this multi-trillion-dollar market.

It took nearly a century for oil to move from spot markets to futures. The computing power market has achieved this in roughly one year. Silicon Data has launched an hourly GPU rental index on the Bloomberg terminal, and Ornn has released forward curves for H100, H200, and B200. Before the advent of forward curves, future computing power prices were merely a cell in a NeoCloud spreadsheet; now, lenders can use them for credit assessment.

Architect Exchange (AX), which holds a license from the Bermuda Monetary Authority, has listed perpetual contracts for H100 rentals, priced in U.S. dollars per GPU hour. On-chain, Hyperliquid’s trade[XYZ] has launched perpetual contracts for H100 hourly rentals, settled in USDC. FalconX has completed its first over-the-counter computing power forward swap, Polymarket has executed institutional-grade on-chain block trades, and Kalshi has also introduced AI computing power forward curves.

The CME and Silicon Data have announced the launch of H100 and B200 rental index futures, tracking Silicon Data’s hourly GPU rental index. Each contract represents 730 GPU hours, equivalent to renting one card continuously for a month. The contracts are listed on NYMEX, a subsidiary of CME specializing in energy trading, and the high-level executive quoted in the press release is the CME’s Global Head of Energy and Environmental Products, indicating how the CME views computing power. Furthermore, the regulatory stance is clear: CFTC Chairman Michael Selig has called computing power the most important commodity of our time and issued a request for comments on computing power derivatives on August 19th, opening a 60-day public comment period.

However, it is important to note that the CME’s futures, originally slated for early October, have been postponed due to the CFTC extending its review period by 45 days to November 9th. This delay is attributed to concerns about the fragmented nature of the GPU rental market and its susceptibility to manipulation. Intercontinental Exchange (ICE) has also announced plans with Ornn to launch GPU computing power futures, which are also awaiting regulatory approval.

But here’s a crucial detail: the earliest derivatives markets for computing power used “U.S. dollar stablecoins” for margin and settlement. AX’s computing power contracts can use U.S. dollars or dollar stablecoins as margin, while Hyperliquid’s H100 contracts are directly settled in USDC. From day one, computing power and stablecoins have been intertwined.

Financial commentator Wen Hongjun has proposed a significant framework: the dollar’s hegemony has three store-of-value anchors. The Bretton Woods Agreement of 1944 pegged the dollar to gold. After the Nixon Shock of 1971, the dollar shifted to being pegged to oil and U.S. debt. The third anchor, expected in 2026, is computing power.

This is not just a personal projection. On September 29th, former President Trump met with Dario Amodei, Sundar Pichai, Jensen Huang, Greg Brockman, Mark Zuckerberg, and Elon Musk at the White House to sign a one-page agreement committing to establish internal safety controls, undergo external audits, and report to an independent committee under their boards. David Sacks, one of the architects of the agreement, referred to this gathering as a “Bretton Woods Conference for Superintelligence,” as data center operators, chip manufacturers, hyperscale cloud providers, and frontier model companies were all in the same room for the first time.

While this agreement focuses on AI safety governance and not monetary systems, the political framing of “Bretton Woods” reveals America’s strong ambitions. The reason the 1944 Bretton Woods system was dollar-centric was that the U.S. held over 60% of the world’s gold reserves, and any agreement had to reflect Washington’s preferences. In 2026, the U.S. holds frontier models, NVIDIA chips, and hyperscale clouds. The “gold reserves” of this era are computing power.

Capital has also followed suit. In August, NVIDIA CEO Jensen Huang, in collaboration with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, announced the formation of a $500 billion AI computing power investment platform. Huang stated that this marks the first time technology chips have been positioned as an investable asset.

Putting these pieces together, the closed loop of a “computing power dollar” becomes clear. This is why a computing power-backed stablecoin is one of the natural solutions for the U.S. to address its $40 trillion debt pressure: leveraging AI productivity growth to extend the life of dollar credit.

According to Reuters, the Shanghai Futures Exchange is in preliminary discussions to design AI Token futures, partly motivated by the AI competition between the U.S. and China. China’s daily Token usage has grown a thousandfold since the beginning of 2024, exceeding 140 trillion by the end of March this year. Shanghai has had a spot trading platform for computing power operating since 2023, and in June, a Shanghai municipal government document mentioned “computing power futures” for the first time.

The differing approaches from both sides are fascinating. This is akin to the historical competition between WTI, Brent, and Shanghai crude oil. Whoever’s pricing unit becomes the global benchmark will hold the pricing power for the next century.

Computing power possesses a characteristic that oil lacks: it cannot be stored or preserved long-term. Just as Ken Griffin, CEO of Citadel, the world’s largest hedge fund, compared AI computing power (especially GPU capacity) to airline seats at the New York Global Markets Future Summit on October 20, 2026, and warned of an “almost certain AI bubble,” the inability to store computing power means the traditional cost-of-carry arbitrage relationship between spot and futures will break down. GPU hours not used today are gone forever tomorrow.

This implies that while futures can address “hedging,” they cannot solve three other issues: the fragmentation of ownership, capital mobility, and inter-machine payments. These three aspects are precisely where tokenization excels.

The path to the capitalization and tokenization of computing power, as outlined by Wen Hongjun, can be divided into three layers:

1. Measurement Layer: Making data on electricity and GPU usage verifiable and immediately on-chain. Without credible measurement, there can be no credible pricing, and thus no credible lending.
2. Asset Layer: Tokenizing future rental income from GPU clusters. NVIDIA is already comparing computing power to toll roads or commercial real estate for financing, and tokenization is the next step: allowing these income rights to be sliced, transferred, and collateralized. As Simon Taylor puts it, when the record itself is the token, collateral doesn’t need to be repackaged because the token is the package.
3. Settlement Layer: AI agents purchase intelligence by the token, pay per second, and operate 24/7, meaning they cannot wait for banks to open. This layer requires a “computing power dollar stablecoin” denominated in U.S. dollars, dedicated to computing power purchases and settlement, and backed by short-term U.S. Treasuries.

Therefore, a computing power dollar stablecoin is not an invention but a natural product of these three layers combined: the measurement layer generates price, the asset layer generates capital, and the settlement layer generates currency.

However, a question remains. Wall Street’s capital is flowing into the computing power industry through two entirely different paths:

1. The “traditional debt path,” which is already highly industrialized. We are already seeing credit default swaps (CDS) for CoreWeave and Oracle nearing their limits.
2. The nascent “tokenized prepayment path.”

The former appears mature but is stuck on an unanswerable question: how much will a GPU be worth in four years? The latter offers the potential to transform prepayments locked into long-term contracts into liquid capital. The difference between these two paths is the true determinant of success in the financialization of computing power, and a detailed explanation would be too complex for this space, reserved for future installments of the “Stablecoin New Finance Series.”

As a veteran of the traditional financial markets for two decades, I must also offer a dose of realism:

• Depreciation Risk: GPUs are rapidly depreciating assets. According to SemiAnalysis, the price of a one-year H100 rental contract rose from a low of $1.70 per hour in October 2025 to $2.35 within five months, an increase of nearly 40%. Assets that can rise this quickly can also fall just as fast.
• Index Quality: The CFTC’s extended review is precisely due to the market’s fragmentation and susceptibility to manipulation. If the price benchmark is unreliable, all financial products built upon it will falter.
• Demand Validation: The entire computing power dollar closed loop bets on AI productivity growth outpacing the cost of debt. If it fails to do so, computing power assets will be the first to be repriced.

Taiwan possesses the manufacturing capabilities for the vast majority of high-end chips and AI servers globally. However, historically, Taiwan has earned money from “shipping” rather than “pricing.” If computing power truly becomes the third anchor for the U.S. dollar, Taiwan should not remain merely a passive contract manufacturer. Instead, it should form an industry alliance to directly price and settle chip and computing power assets in U.S. dollar stablecoins, creating a “tech dollar” settlement layer accessible to Taiwan’s tech industry or the global industry. This also extends the viewpoint from “Stablecoin New Finance Series-8”: Taiwan’s greatest opportunity for stablecoins has never been in retail payments, but in functional stablecoins pegged to industry demand. The goal is not to issue another currency, but to make computing power a financial infrastructure that can be measured, tokenized, and settled in U.S. dollars.

From gold to oil, and now to computing power. The U.S. is pricing in GPU hours and U.S. dollar stablecoins, while China is pricing in tokens and the RMB system. CME and ICE are still knocking on the CFTC’s door, while perpetual contracts on-chain are already trading in USDC. Tokenization of computing power and computing power dollar stablecoins are not a matter of choice but an inevitable new financial species emerging from the U.S.-China tech competition. The only question remaining is: as a new Bretton Woods order takes shape, will Taiwan sit at the table, or will it only appear on the menu?

The above analysis represents the personal research and compilation of Wen Hongjun and does not constitute any investment advice, but is for industry development and interested parties’ reference only.