A quiet but consequential shift is reshaping the global artificial intelligence competition, and it has little to do with which country builds the most powerful model.

Jensen Huang did not mean to describe a geopolitical strategy. But when Nvidia’s chief executive declared, “Your workload is inference, your tokens are your commodity, and that compute is your revenue,” he was articulating, from the supply side, something China had concluded from the other direction.

To understand why, start with a basic concept. Tokens are the fundamental units AI models use to process and generate language: every word, response and automated task breaks down into them. Cloud providers charge by the token the way utilities charge by the kilowatt-hour. Whoever produces them most cheaply, at the greatest scale, holds an advantage in the AI economy like how cheap steel once decided industrial supremacy.

On OpenRouter, a widely used platform for developer AI access, Chinese models have surpassed American ones in token consumption for the first time: 5.16 trillion tokens against 2.7 trillion in a single week. By mid-March, Chinese models accounted for 36 per cent of OpenRouter’s global volume, with a weekly consumption of 7.36 trillion tokens.

Washington has spent three years building the most comprehensive technology export control regime since the Cold War, designed for a world where competitive advantage travels in hardware: chips that can be counted, shipments that can be blocked, supply chains that can be pressured. A different contest is now taking shape alongside it, one for which the existing toolkit has no architecture.

Industrial eras have been defined not only by their dominant technologies, but by the units used to measure them. Tonnes of steel told you who was industrialising. Barrels of oil told you who held leverage. Gross domestic product, constructed in the 1930s, defined what economic activity meant, and governments built policy architectures around maximising it.