The narrative was clear for a long time: American labs build the best models, the rest of the world plays catch-up. At the very top, that’s still true — but at the bottom of the invoice, things look different. According to a CNBC report, more and more US firms are switching to open Chinese models because frontier providers’ costs are going through the roof.
The sober math
The heart of the report isn’t ideology, it’s a spreadsheet. Open Chinese models can be 60 to 90 percent cheaper than the leading models from Anthropic and OpenAI. For companies deploying AI at scale, that’s not a rounding error — it decides entire budgets.
And the models have gotten good enough. Not every task needs the absolute frontier. For plenty of production workloads — classification, summarization, standard coding — a solid open model that costs a fraction of the price does the job.
Why this is happening now
The timing is no accident. Over the past few weeks I’ve covered several pieces of the same picture here: Tesla capping its AI spending, Microsoft building its own models to avoid paying Anthropic, and frontier prices rising rather than falling. As of this week, Fable 5 costs double what Opus 4.8 does.
When the most expensive models keep getting more expensive and the cheap ones keep getting better, the two curves eventually meet. For many companies, that crossover point has apparently arrived.
My take
This may be the most important shift in the 2026 AI market — and it’s largely flying under the radar of the big announcements. It’s not the next record-breaking benchmark model that changes the economics, it’s the question of what a prompt actually costs in production.
For Anthropic this is a serious challenge. Claude Code and enterprise revenue are booming, but the price pressure from below isn’t going away. The American bet has to be that their models are so much better on the truly hard, high-value tasks that the premium pays for itself. For everything else, price is winning right now.
Sources: