Why Enterprises Are Quietly Routing AI Workloads to Cheaper Chinese Models
Reporting this year found Chinese open-weight AI models capturing a growing share of enterprise usage, priced dramatically below US frontier models. The interesting part isn't the discount — it's what enterprises are (and aren't) routing to them.
Business reporting this year tracked Chinese-origin AI models hitting a weekly peak of 46% of enterprise token volume on at least one major model marketplace — up from roughly 11% averaged over the prior year — at a price roughly 60–90% below leading US frontier models.
What's notable isn't just the price gap, it's the pattern of adoption: enterprises aren't wholesale replacing their primary model provider, they're routing specific, well-understood, lower-stakes workloads to the cheaper option while keeping premium models for tasks that need the strongest reasoning or the tightest trust guarantees.
That's a healthier instinct than it might sound. Model selection is turning into a genuine architecture decision — which workloads tolerate a cheaper, less-audited model, and which ones (identity decisions, compliance determinations, anything touching regulated personal data) don't get to be a cost-optimization exercise at all.
It's the same principle we apply across our own product suite: the components making high-stakes identity and compliance calls are held to a different bar than the components doing routine text processing, regardless of which model or vendor is cheapest this quarter.
This post is Pandasoft's own commentary, grounded in real reporting rather than a copy of it. Source: CNBC, Jul 7, 2026.