The idea that artificial intelligence compute will function like oil in the global economy, anchoring American financial power through a "token dollar," sounds elegant but doesn't hold up to scrutiny. While Groq founder Jonathan Ross recently argued that AI inference pricing could replicate the petrodollar system that has defined American economic dominance since the 1970s, the comparison overlooks fundamental differences in how compute markets actually work compared to physical commodity markets .

What Made the Petrodollar System Actually Work?

The original petrodollar wasn't just about pricing oil in dollars. It was a specific institutional arrangement with clear enforcement mechanisms. In 1974, following the global oil crisis, the United States and Saudi Arabia established the Joint Commission on Economic Cooperation. Beyond the public agreement on technology transfer and infrastructure, there was a private understanding that Saudi Arabia would price oil exclusively in dollars and recycle its surplus revenues into U.S. Treasury securities and American investments. In exchange, the United States provided military protection and arms sales .

This created a closed loop with real structural power. Every country needing oil needed dollars. The dollars paid for oil flowed back into U.S. government debt, financing American deficits without the economic consequences that would have crippled other nations. By 2022, Saudi Arabia, the UAE, Kuwait, and Iraq held over $271 billion in U.S. Treasury securities, demonstrating the recycling mechanism's durability .

Why Can't AI Compute Replicate This Model?

The comparison between oil and AI compute breaks down in three critical ways. First, there is no sovereign counterparty agreement equivalent to the U.S.-Saudi arrangement. U.S. cloud providers price in dollars because they are American companies, not because of a geopolitical bargain. That is ordinary commerce, not a currency regime .

Second, compute is fundamentally substitutable in ways oil never was. Model weights can be copied, distilled, fine-tuned, and run on non-U.S. hardware. Open-source models have evolved from experimental projects to production-grade alternatives in under two years. DeepSeek and Qwen now hold roughly 15% of the global AI market, up from just 1% a year ago. Qwen alone has surpassed 700 million cumulative downloads on Hugging Face, overtaking Meta's Llama as the most-used base model for fine-tuning .

None of these alternatives require payment in dollars. Users can run Qwen on their own hardware purchased in yuan, rent capacity from non-U.S. cloud providers using local currency, or self-host a distilled DeepSeek variant on a consumer graphics processing unit (GPU). The inference layer, which is Groq's domain, is precisely the layer most vulnerable to commoditization and currency fragmentation .

Where the Real Dollar Story Actually Lives

The genuine dollar advantage in AI compute exists not at the inference layer but at the capital markets and financing layer. GPU procurement, data center construction, and the structured credit financing both are overwhelmingly denominated in U.S. dollars and intermediated through American capital markets .

CoreWeave, a leading infrastructure provider, has raised over $28 billion in total financing commitments, with syndication from major institutions including Morgan Stanley, Goldman Sachs, JPMorgan, Blackstone, and Magnetar. Its most recent facility, an $8.5 billion delayed-draw term loan rated A3 by Moody's, is secured by GPU clusters and contracted revenues. The entire infrastructure financing stack runs through American banks, American legal structures, and American capital markets .

This creates genuine dollar demand, but through a completely different mechanism than Ross describes. It is not "the world pays for AI inference in dollars." Rather, it is "the world finances AI infrastructure through dollar-denominated debt markets because that is where the liquidity, legal frameworks, and credit expertise exist." The distinction matters because the former implies a stable fifty-year currency regime, while the latter implies a structural advantage that erodes as capital markets develop elsewhere .

How to Understand the Real Financial Dynamics of AI Infrastructure

• Facility-Layer Assets: Land, power, and cooling infrastructure have long-dated, bankable characteristics that naturally denominate in U.S. dollars and resemble traditional infrastructure finance with geopolitical stickiness.
• Compute-Layer Assets: GPUs and networking hardware have fundamentally different financial characteristics, with high substitutability that erodes dollar lock-in as open-source alternatives proliferate globally.
• Forward Curve Development: The absence of a liquid forward curve for compute keeps financing costs elevated because lenders cannot hedge GPU residual value risk, charging massive uncertainty premiums that a mature derivatives market could reduce.
• Index Infrastructure: Startups are currently constructing the benchmarking standards, marketplace architecture, and index infrastructure needed to support spot pricing, forward curves, and eventually derivatives for compute assets.

The Gulf States' Shifting Role in Global AI Finance

There is one additional subtlety that undermines the token dollar thesis. The same Gulf states that anchored the original petrodollar system are now major buyers of compute infrastructure, not sellers of the underlying commodity. Gulf sovereign wealth funds deployed $126 billion in 2025, accounting for 43% of total global sovereign investment. They poured $66 billion into AI and digitalization alone .

Saudi Arabia's PIF-backed entity HUMAIN plans to build up to 6 gigawatts of data center capacity by 2034, partnering with Nvidia, AMD, and Qualcomm. The UAE's MGX, backed by Mubadala, was one of four equity partners in $500 billion of new U.S. data center investment announced by the Trump administration .

In the original petrodollar system, Saudi Arabia was the supplier of the scarce resource, with surpluses flowing from Riyadh to Washington. In the compute buildout, the Gulf states are customers and co-investors. Capital flows from sovereign wealth funds into U.S. capital markets, U.S. technology companies, and U.S.-designed hardware. This creates dollar demand, but it is the demand of a buyer, not a seller locked into a denomination agreement. The recycling loop, if one emerges, will look nothing like the 1974 arrangement .

The token dollar thesis captures something real about American financial advantages in AI infrastructure, but it misidentifies where that advantage actually lives. It is not in the pricing of inference compute. It is in the depth, liquidity, and institutional expertise of American capital markets that finance the entire buildout. That advantage is real but also more fragile and more subject to erosion than a fifty-year currency regime would be.