AI is a rapidly growing business expense. Some companies are cutting costs by switching to cheaper Chinese AI models.
Imen Ben Youssef/Hans Lucas/AFP via Getty Images
Hide caption
Toggle label
Imen Ben Youssef/Hans Lucas/AFP via Getty Images
Subscribe to NPR’s First up Morning newsletter with the stories you need to start your day.
SAN FRANCISCO — Flo Crivello’s San Francisco-based startup Lindy.ai is developing artificial intelligence “assistants” to manage your email and calendar. Initially, the company relied heavily on Anthropic’s world-class AI models.
But in meeting after meeting with his finance man, Crivello said one thing became clear: “By far our largest expense was Anthropic,” he said. “More than just payroll.”
More than just payroll – for over two dozen employees. More than rent. More than anything else. As a result, last month Crivello announced that Lindy had migrated 100% of its traffic to the Chinese AI model DeepSeek-V4.
“It was just 10 times cheaper,” he said, adding that it saved the company millions of dollars. “So it was a very, very simple business decision.”
Artificial intelligence has become one of – it is not The — Fastest-growing cost for U.S. businesses. But for many companies it is a double-edged sword: necessary but expensive. To survive, more and more companies are switching from American models to cheaper Chinese AI.
In the race for the best AI models, US companies such as Anthropic, OpenAI and Google are leading the world. Experts believe that Chinese models are six to twelve months behind in terms of performance.
But China has carved out a niche for itself in open-source models that are free to download and customize. “The open source scene is absolutely dominated by the Chinese right now. It’s not even close,” Crivello said.
He said that every founder he knows who works in the AI space is either thinking about switching to Chinese models or has already done so.
And the rising costs of AI are not just a startup problem. Uber CEO Dara Khosrowshahi spoke about this last month Invest like the best Podcast. “We basically used up our AI budget in one quarter, you know, for the entire year. And it’s forcing us to adapt,” he said.
(Uber did not respond to NPR’s request for information about whether it uses Chinese models.)
Bloomberg reported that Airbnb CEO Brian Chesky said the company last year relied on Alibaba’s Qwen model, which was “good,” “fast and cheap.” Perplexity and Nvidia have also taken advantage of Qwen.
Table of Contents
Like a Ferrari or a Honda
Many companies are wary of promoting the use of Chinese models due to political sensitivity, but the models are widely available on AI model hubs such as Hugging Face, on the code hosting platform GitHub, and through model aggregators and inference providers based outside of China.
This also includes the San Francisco-based company Featherless, which offers access to around 30,000 AI models. Founder and CEO Eugene Cheah said Chinese models are popular even if they are not “frontier” or best-in-class models.
“It’s like the difference between driving a Ferrari and a Honda. You can have the best luxury car, or you can just have a right-sized Honda that works,” he said.
“Actually, for many open source AI groups, it’s perfectly fine to be N-1, with N being the limit,” he continued. “Because as the gap becomes smaller and smaller, at some point the question arises: Does it even matter?”
For many, like Lindy, that doesn’t matter. AI’s Honda is perfectly good.
OpenRouter, another platform where startups can access a range of AI models, reported that usage of China’s DeepSeek has increased from around 9% to nearly 20% since January. The use of models from Chinese companies MiniMax, Xiaomi and Tencent has also increased.
Some users download and host Chinese open-source AI models themselves, but many use them through paid AI hosting companies such as Featherless and OpenRouter, so user data is stored in the United States.
Victor Su-Ortiz, global product marketing manager at Shanghai-based MiniMax, recently attended a conference for AI engineers in San Francisco. Companies pay to use AI models by paying for tokens or AI work units. Su-Ortiz said it all comes down to the cost per token.
“Many repetitive tasks can be completed with a model that is just as powerful but has a much lower cost per token compared to leading AI models,” he said. “And that’s essentially why these open weight models came to the United States.”
He said companies are moving from “tokenmaxxing” – deploying as much AI as possible – to cutting costs by limiting usage, moving to cheaper models or redirecting different types of AI work to other types of models.
For research or “deep thinking,” for example, the cutting-edge models could perform better, Su-Ortiz said. “But if you need to perform a high-volume, repetitive encoding task, then one of our models, particularly the MiniMax M3, will give you exceptional performance at just a tenth of the cost.”
Saving a few dollars isn’t worth it for everyone
For some companies, Chinese models are still not good enough. Jon Gordner is the CEO and co-founder of Comment.io, which launched just a few weeks ago and is building a product that he says is like Google Docs for programmers and AI agents.
“We need to develop as good software as possible as quickly as possible. And for us, it’s not worth saving a few dollars on a cheaper model if we have to spend two or three more weeks fixing the bugs,” he said.
Gordner said his company benefits from the Anthropic and OpenAI models in part because both companies subsidize users to attract customers. He said monthly subscriptions now offer tokens at a huge discount – but that likely won’t last forever.
“Then it will make much more sense for us to start evaluating Chinese models and open source models,” he said.
Ara Kharazian is a senior economist at Ramp, a company that helps companies track, control and automate spending. It has insights into AI spending, and Kharazian believes U.S. companies will continue to adapt — in other words, they could keep prices under control or adopt high-quality open-source models to outdo Chinese competitors.
“The rise of these Chinese models shows that companies want something that the American model companies don’t offer today,” he said. “The only reason I’m pessimistic about the Chinese models is because I expect the American model companies to respond competitively.”
Comment.io’s Gordner is less sure. He believes major U.S. AI companies may need to start charging fees more for AI, as pressure to demonstrate profitability increases, perhaps as the IPO gets closer. Both Anthropic and OpenAI have filed confidential documents with the US government to get the ball rolling on potential IPOs.
“At some point,” Gordner said, “the music will stop.”
Anthropic is a financial supporter of NPR.
https://www.npr.org/2026/07/15/nx-s1-5886476/startups-cheap-chinese-ai-models
