Artificial intelligence has rapidly transformed into one of the most significant, if not the fastest-growing, operational expenses for businesses across the United States. While U.S. firms like Anthropic, OpenAI, and Google currently lead the global race in AI development, the high cost of these premium services has created a financial burden that is forcing many companies to seek alternatives. A growing number of startups are now turning to Chinese AI models, which offer a more cost-effective solution for managing high-volume tasks.
Flo Crivello, the founder of the San Francisco-based startup Lindy.ai, which builds AI assistants for email and calendar management, found that his company’s reliance on Anthropic’s top-tier models had become his single largest expense—surpassing both payroll for his two dozen employees and office rent. Consequently, Crivello announced last month that Lindy had migrated 100% of its traffic to the Chinese AI model DeepSeek-V4. Crivello noted that the open-source landscape is currently dominated by Chinese developers, and he observed that nearly every founder he knows in the AI space is either already using Chinese models or actively considering the transition.
This financial strain is not limited to small startups. Uber CEO Dara Khosrowshahi recently discussed the issue on the Invest Like the Best podcast, revealing that the company exhausted its entire annual AI budget within a single quarter, forcing the firm to make significant operational adjustments. While Uber did not respond to inquiries regarding whether it utilizes Chinese technology, other major players have been more transparent. Bloomberg reported that Airbnb CEO Brian Chesky relied on Alibaba's Qwen model last year, describing it as fast, cheap, and effective. Furthermore, companies including Perplexity and Nvidia have also utilized the Qwen model.
Although industry experts suggest that Chinese models currently trail U.S. counterparts by six to 12 months in terms of raw capability, many companies find the performance gap increasingly irrelevant. Crivello compared the choice to driving a luxury Ferrari versus a reliable Honda, noting that for many businesses, a functional model at scale is sufficient. Data from OpenRouter, a platform that provides access to various AI models, reflects this shift; usage of China's DeepSeek has surged from approximately 9% in January to nearly 20%. Platforms have also seen increased adoption of models from Chinese companies such as MiniMax, Xiaomi, and Tencent.
To address security concerns, many U.S. companies avoid downloading Chinese models directly, instead accessing them through paid, U.S.-based AI-hosting services like Featherless and OpenRouter, which ensure that user data remains within the United States. Victor Su-Ortiz, who handles global product marketing for the Shanghai-based firm MiniMax, explained that the shift is driven by the cost per token. He noted that while cutting-edge models may be superior for complex research or deep reasoning, repetitive, high-volume tasks can be handled by models like the MiniMax M3 at roughly one-tenth the cost of leading American alternatives.
Not all companies are convinced by the cost-saving potential. Jon Gordner, CEO and co-founder of Comment.io, which is developing a tool for coders and AI agents, stated that for his firm, saving money is not worth the risk of spending weeks correcting AI-generated errors. He emphasized the need for high-quality software development over immediate cost reductions. However, Ara Kharazian, lead economist at the spending-tracking firm Ramp, believes the rise of Chinese models highlights a market demand that American companies are currently failing to meet. While Kharazian remains somewhat bearish on the long-term dominance of Chinese models, assuming that U.S. firms will eventually respond with more competitive pricing or superior open-source offerings, others are less optimistic.
Gordner suggested that the music may eventually stop for the current AI cost structure, particularly as major U.S. AI companies face mounting pressure to demonstrate profitability ahead of potential initial public offerings. With both Anthropic and OpenAI having filed confidential paperwork with the U.S. government to move toward IPOs, the industry may see a shift in pricing strategies that could further influence the adoption of international alternatives.
"Actually, a lot of open-source AI groups are perfectly fine being N-1, N being where the frontier is," he continued. "Because as the gap keeps shrinking, at some point the question is: Does it actually matter?"
"Then for us, it's going to make a lot more sense to start evaluating Chinese models and open-source models," he said.
