TECHNOMALIST
Chinese artificial intelligence models are making significant inroads in the United States, attracting a growing number of users with their combination of low cost and competitive performance. The shift is evident among tech leaders and companies seeking to cut expenses without sacrificing capabilities.
Mozilla Chief Technology Officer Raffi Krikorian recently switched to Moonshot’s Kimi K3 for many daily tasks, describing the model as “snappier” than more expensive U.S.-made alternatives. He had previously used Z.ai’s GLM-5.2 for routine work like managing calendars and emails. “The open frontier is becoming increasingly Chinese-built,” Krikorian said, noting that open-source Chinese models are now nearly as good as closed systems from leading U.S. firms.
Cryptocurrency exchange Coinbase is among the firms turning to Chinese AI to trim costs. Technology executive Curt Meinhold said he prefers DeepSeek for tasks such as finding business leads, citing the vast price difference. “If I can pay a handful of cents per million output tokens versus 30 bucks or 40 bucks or 50 bucks, then it’s good enough,” he said.
Chinese labs are releasing models at a rapid pace. In June, Zhipu (Z.ai) launched GLM-5.2, followed by Moonshot’s K3 in July. Alibaba previewed its Qwen3.8 Max, and DeepSeek rolled out previews of its V4 model, challenging OpenAI’s GPT and Google’s Gemini. These releases follow DeepSeek’s earlier breakthrough, which demonstrated parity with U.S. models at a fraction of the cost.
The cost advantage is amplified in “agentic” AI scenarios, where models autonomously perform multi-step tasks, compounding token usage. Goldman Sachs wrote in a July research note that Chinese models are reaching a “critical stage” for widespread adoption as demand for cost-effective AI surges.
Data underscores the trend. Over the past month, the five most popular models on OpenRouter, a tracking platform, were Chinese. Sensor Tower estimates showed Kimi was downloaded more than 930,000 times in the week after K3’s release, a 200% jump; in the U.S., downloads surged 387% to about 86,000. Moonshot temporarily halted new subscriptions due to overwhelming demand.
Most Chinese AI models are open-source, while U.S. leaders like Anthropic and OpenAI keep their frontier systems closed. Analysts expect Chinese vendors to leverage open-source software to boost global adoption. A group of U.S. companies, including Microsoft, Meta, and Nvidia, signed an open letter backing open models, even as Washington weighs further restrictions.
American-led controls already block China from accessing cutting-edge AI chips, and Treasury Secretary Scott Bessent has warned of more sanctions. The Trump administration accused Moonshot of using “covert” methods to build K3 off Anthropic’s Fable, which Beijing called “groundless.” Some experts say U.S. export policies can backfire: after controls on Anthropic’s Fable and Mythos models kept them offline for weeks, Chinese rival Z.ai quickly released GLM-5.2. “Restricting an American model can immediately create an opening for a Chinese competitor,” said Anastasios Angelopoulos of Arena, an AI evaluation platform.
Despite momentum, Chinese models still lag U.S. leaders in overall capabilities. American firms are responding with cheaper alternatives of their own. In China, fierce domestic competition is driving startups to raise funds and expand globally.
Sustainability questions remain. Z.ai reported a 132% revenue increase last year but its net loss widened to $694 million. For now, early adopters are enthusiastic. Krikorian advised anyone doing serious AI work to “at least evaluate” Chinese models.
See an error? Read our corrections policy or email [email protected].
