AI Pricing Is in Freefall and Nobody Knows the Floor

Xiaomi dropped its MiMo-v2.5 series API pricing by up to 99% in a move that hit both HN and Reddit hard. This follows DeepSeek's earlier pricing shock and signals a structural pattern: Chinese AI companies are treating inference pricing as a land-grab strategy, subsidizing usage to acquire training data and market position. One HN commenter put it bluntly: 'China doesn't care about money. We want AI in people's hands.'

The pattern here is not just competition, it is a race to zero with a different set of incentives on one side. Western frontier labs like Anthropic and OpenAI are pricing for margin and R&D recovery. Chinese labs appear to be pricing for penetration. The result is a 30x or greater price gap between frontier closed-source models and what is now available via Chinese APIs, and the capability gap is narrowing fast.

On HN, an older thread made the same point from first principles: local and outsourced AI is becoming more economical than frontier labs for most production workloads. Uber's AI team blew through its entire quarterly budget in one quarter, partly because engineers were incentivized by a usage leaderboard to maximize token consumption regardless of business value. The era of cheap experimentation is ending for teams using premium APIs, even as the floor drops out from under commodity inference.