American companies are increasingly turning to Chinese artificial intelligence (AI) models, driven by a widening U.S. performance gap and skyrocketing domestic tech costs.
Recent releases from Chinese developers like DeepSeek and Z.ai have emerged as highly competitive alternatives to flagship systems from Anthropic and OpenAI, sparking a significant shift in corporate AI adoption.
Data from OpenRouter, a platform allowing developers to access diverse AI systems, reveals that U.S. corporate use of Chinese AI models has consistently stayed above 30% weekly since February, occasionally peaking at 46%. This marks a dramatic surge from an 11% average over the previous year, and 4.5% in early 2025.
The shift comes as American AI labs hike token prices for their frontier models, leaving enterprises grappling with unexpectedly high operational costs. Industry experts note that open-source and open-weight Chinese models can be 60% to 90% cheaper than proprietary U.S. counterparts.
“Where previously U.S. companies were prioritizing AI adoption regardless of model, now they’re getting more cost-conscious,” Kyle Chan, a fellow at the Brookings Institution, told CNBC.
Financial reality is altering how Silicon Valley startups manage their infrastructure. In June, AI startup Lindy migrated 100% of its traffic from Anthropic’s Claude models to DeepSeek.
Lindy CEO Flo Crivello stated the transition caused their cost curve to “crash to the ground,” anticipating millions of dollars in savings within months. Similarly, Z.ai’s GLM 5.2 model experienced explosive growth following its June debut. According to developer platform Vercel, GLM 5.2 recorded a 27-fold increase in daily token volume and an 80-fold surge in customers within its first week.
While top-tier U.S. models still lead in raw capabilities, Chinese variants are rapidly closing the distance.
Analysts estimate Chinese frontier models lag just six to nine months behind U.S. leaders, proving highly capable for all but the most complex tasks. Notably, Z.ai’s GLM 5.2 performed within a single percentage point of Anthropic’s Opus 4.8 on a prominent agentic benchmark, while costing only a fifth of the price.
However, this commercial pivot is unfolding amid tightening political scrutiny.
The U.S. administration is actively reviewing ways to regulate powerful AI systems and curb the adoption of foreign alternatives. Just last month, OpenAI restricted a new model rollout at the government’s request, following a tense standoff between the Trump administration and Anthropic over export controls on its Mythos and Fable models.
As U.S. regulatory boundaries tighten and domestic pricing fluctuates, technology advocates warn of a growing market divide. Yacine Jernite, head of machine learning at Hugging Face, noted that American companies risk facing a stark ultimatum: pay premium prices for restrictive U.S. proprietary models, or rely on Chinese open-source infrastructure as the only economically viable path to maintaining control over their own AI development.
