The 'token dollar': AI compute could replace oil as the foundation of US economic dominance
Apr 2, 2026
Key Points
- Nvidia's Jonathan Ross argues AI compute is replacing oil as the economic foundation of US global dominance, with American companies controlling a dollar-denominated resource worth tens of billions in training and hundreds of billions in inference.
- Countries are competing not just for AI talent and chips but to keep AI a dollar-denominated economy, making compute infrastructure the new geopolitical battleground driving policy and capital allocation.
- Compute-backed credit lines are emerging as a financial product despite rapid chip depreciation, with the real infrastructure constraint being powered data center shells and logistics rather than chips alone.
Summary
Jonathan Ross, chief software architect at Nvidia, argues that AI compute is poised to replace oil as the economic foundation of US global dominance. He calls this shift the 'token dollar'.
The petrodollar has anchored American financial power since the 1970s because oil is priced in dollars, forcing every country to hold dollars to fuel their economies. Ross draws a parallel to AI. Training runs cost tens of billions, inference is scaling to hundreds of billions, and the companies selling that compute are American. The currency is dollars. The structural leverage is identical, only the resource has changed.
The geopolitical stakes are sharper than simple compute competition. Countries are fighting over whether AI remains a dollar-denominated economy, not just over talent and chip access. This contest drives infrastructure, policy, and capital allocation across the world.
Compute-backed credit lines are emerging as a new financial product. Debt is being issued against GPUs with expected future cash flows, creating a novel asset class. Compute depreciates quickly due to innovation and scale economies, which makes it weak collateral on traditional grounds. But current chip supply constraints keep older silicon valuable in secondary markets, and capital markets increasingly price on revenue generated from chips rather than resale value.
Chips alone do not tell the full story. Powered shells, racks, energy, interconnect, and cooling matter as much. According to CoreWeave, the real bottleneck is powered shells and the logistical complexity of building data centers at speed, not chips themselves. This points to a broader shift. America is learning to build complex infrastructure fast again, a capability the country has largely lost over decades.