Businesses are increasingly feeling the financial strain from artificial intelligence expenses. Some companies are finding relief by adopting more affordable Chinese AI models as a cost-cutting measure.

Flo Crivello, the founder of Lindy.ai, illustrated this trend with his startup, which creates AI assistants for managing emails and calendars. Initially, Lindy relied on Anthropic’s sophisticated AI models, but these turned out to be the startup’s most significant expense, even surpassing payroll for over twenty employees and other costs like rent. Eventually, Lindy switched entirely to DeepSeek-V4, a Chinese AI model, which significantly reduced expenses by tenfold, saving the company millions.

This economic pressure from AI expenses affects many businesses, not just startups. Uber CEO Dara Khosrowshahi mentioned that his company exhausted its AI budget in just one quarter, which forced adjustments. Similarly, Airbnb CEO Brian Chesky acknowledged utilizing Alibaba’s Qwen model for its affordability and efficiency.

“The open-source scene right now is absolutely dominated by the Chinese. It’s not even close,” stated Crivello.

While American companies like Anthropic, OpenAI, and Google are leading in AI capabilities, Chinese models have found their place within the open-source market. These models are available on platforms like Hugging Face and GitHub, as well as aggregators and inference providers outside of China. Eugene Cheah, CEO of Featherless, described these models as popular despite not being top-tier, likening them to the choice between a luxury car like Ferrari and a reliable Honda at scale.

This shift is not limited to free models. Companies are using paid hosting services like Featherless and OpenRouter for more accessible AI models while keeping user data within the U.S. Victor Su-Ortiz from Shanghai-based MiniMax emphasizes token costs, highlighting how repetitive tasks can utilize models with lower expense per token. Such financial incentives push more companies to consider Chinese models despite potential quality differences.

Nevertheless, some businesses remain skeptical about Chinese models. Jon Gordner, CEO and co-founder of Comment.io, prioritizes speed and accuracy over cost savings, opting for American models with subsidies. He suggests evaluating Chinese models if costs rise in the future. Ara Kharazian from Ramp anticipates U.S. companies will adapt, potentially offering high-quality open-source models to compete with Chinese alternatives.

While the flexibility and affordability of Chinese models are attractive to many, others remain cautious, speculating that U.S. AI firms might adjust pricing as they inch closer to going public.

Anthropic supports NPR financially.