West-Washing Chinese AI Dispersion on the eve of Trillion Dollar IPOs

The race

We've talked about the IPOs here, and the debate on whether A.I. shows up in jobs or productivity at all.² Trillion-dollar valuations may hang on a narrative: that the frontier of digital intelligence is American.¹ But what if, in the long run, the smartest digital intelligence is the intelligence you can actually use and afford?

Their chart, our way

When buyers pay per token, they vote with their wallets. OpenRouter, a marketplace that routes each prompt to whichever model a developer picks, is one of the cleanest public windows into that vote. Weekly volume has climbed from roughly 2 trillion to more than 30 trillion tokens in a year.³

The growth is the headline everyone reports. Chinese domination of the open marketplace is curiously missing from the conversation.

The open secret

On the open token market, the most-used models are open-weight releases from DeepSeek, Hy3, and Kimi, all Chinese labs, at a fraction of frontier prices.³⁴ The largest band is DeepSeek's.

Why hide the most interesting angle exposed by the source data?¹ A world standardizing on cheap, competent Chinese open-weight A.I. does not support a two-horse American race with trillions in valuations.

The fork

Re-cut the same tokens by where the model was made, and the trajectory is plain:

The humanist case

From a humanist perspective, the diffusion issue is paramount.⁵ We know what happens when cultures compete against others with substandard tools. Cost is the lever, and cost is where Chinese labs compete hardest. Google is making a good effort here with its Gemma models: they are good, and cost-effective. But it feels like this omission from a storied media house pinpoints the precarious balance of power between China and the rest.

Is it wise to put all our hopes, and so much revenue, behind these tools inside an outdated, capital- and wealth-concentrated social structure? Are the stakes in this race so high that even the slightest taint on the dominance narrative is too big a risk to entertain?