Cheap, capable Chinese AI models are starting to reshape how factories and engineering teams automate their operations, prompting manufacturers and industrial software groups to reassess which systems underpin their engineering and production workflows.
According to data from AI model routing platform OpenRouter, Chinese AI platforms including DeepSeek, Xiaomi, Minimax and Tencent, overtook US models for the first time in June 2026 in terms of weekly total token share (the main measure of how much AI models are used).
It also said that DeepSeek alone saw its share of weekly AI token flow double from 9% to 18% between January and June this year, with much of it now flowing through AI agents rather than one-off chat queries from humans.
Many Chinese developers are pursuing a different strategy for automation buyers from their US rivals: models that are good enough for repetitive, high-volume industrial tasks, at a fraction of the inference cost.
OpenRouter’s Justin Summerville said Chinese open-source models typically run 60–90% cheaper than leading offerings from Anthropic and OpenAI, with some priced as low as $0.14 per million input tokens versus $5 for OpenAI’s GPT-5.5 — a gap that matters enormously once a model is running continuously inside an automated pipeline rather than answering occasional prompts.
The latest example is GLM-5.2, released last month by Beijing-based Z.ai. AI analysis platform Artificial Analysis ranks it fourth overall on its Intelligence Index and first among open-weight models, at a fraction of the price of comparable frontier systems — making it an increasingly attractive engine for automating engineering and operations tasks at scale.
And the commercial pull is already reaching industrial automation software directly. In March, Siemens announced an expanded partnership with Alibaba Cloud to integrate Alibaba’s Qwen models into its product lifecycle management (PLM) software, with the explicit goal of automating engineering workflows and product development tasks.
“Bringing AI into the real world requires more than great models,” said Roland Busch, President and CEO of Siemens, at the company’s RXD Summit in Beijing. “It requires an industrial AI operating system: a technology stack that connects data, software and intelligent hardware.”
Busch said Siemens is building that automation layer through its Xcelerator platform, aiming to “make industrial AI accessible at scale” — language that points squarely at automating engineering and production workflows rather than isolated AI features.
The trend is worrying Western governments, who fear that routing automated industrial workflows through Chinese-developed models could create data exposure risks.
Airbnb CEO Brian Chesky pushed back on that concern after questions from US lawmakers about his company’s use of Chinese models in its automated customer service systems. “We are not providing data to any Chinese companies. They don’t have access to any data,” Chesky told Bloomberg. “We’re primarily using a variety of open-source models, including U.S. open-source models. An open-source model does not have access to data. It doesn’t work that way.”
Chesky has called Alibaba’s Qwen models “fast and cheap” for certain use cases; Airbnb says automating its support system with AI has cut customer service response times from nearly three hours to six seconds — a case study other automation-minded enterprises are watching closely.
Analysts see the pricing gap as intentional. Researchers at the Mercator Institute for China Studies (MERICS) describe a strategy of “wide diffusion and cheap tokens” — releasing open-weight models and pricing them aggressively to maximize adoption in automated systems worldwide, ahead of near-term profit.
“By making its models widely available and cost effective, Chinese companies have found an alternative way to stay in the global AI competition,” said Wendy Chang, Senior Analyst at MERICS. “By leveraging a strong presence in the open-source community and extremely low token prices, the strategy hopes to capture market share across the world, even if it doesn’t translate into profits right away.”
