Chinese open models now run the majority of the world's AI tokens

This is the other half of the token-bill story we keep telling. Everyone's watching their US model spend climb — and while they watch, the rest of the world has quietly moved its actual workload onto cheap Chinese open weights. The point isn't 'switch to DeepSeek tomorrow.' It's that the answer to a rising bill was never a discount from the same provider — it's a second, cheaper supply chain that's now mature enough that most real token volume already runs on it. If you've only ever tested US models, you're making a pricing decision with half the market out of view. Take one non-critical, high-volume task this week and run it on an open Chinese model beside your current one; compare quality and cost honestly. You may keep the expensive model for the hard parts — but then you're choosing it, not defaulting to it. One caveat: open weights you host are yours to govern; a model you call over someone's API still sits under their terms and someone's jurisdiction.