US firms cut AI costs with open-source Chinese models amid security concerns

During Pinterest's earnings call, CEO Bill Ready claimed that the company's AI tools, built on open-source models including Alibaba's Qwen, cost less than 8% of comparable proprietary solutions from providers such as Anthropic and OpenAI. DoorDash has echoed these savings, attributing part of its efficiency to Chinese AI models, and has become the latest firm to receive a subpoena-style letter from US House committees, after similar requests were sent to Airbnb and Cursor.

Developers worldwide are mixing various models, but the extent of Chinese model adoption by large US corporations remains unclear. Public sentiment is low, with a June survey indicating only 9% of US respondents trust Chinese AI versus 52% for American counterparts, raising concerns about hidden biases and censorship. While lawmakers and White House officials argue for tighter controls over Chinese open-source AI on grounds of security and IP theft, company leaders like Luma AI's Amit Jain say there is no evidence of a shift in corporate behavior without clearer guidelines. The debate highlights a growing gap between cost-driven business decisions and consumer demand for transparency about the origins of the chatbots they interact with.