US Companies Increasingly Turn To Cheaper Chinese AI

Despite the continuing claims that 'this time is different,' it turns out that AI is still subject to Economics 1: if something becomes too expensive, the market will seek a cheaper alternative.

This is especially true for corporate clients concerned about the returns they still cannot quantify, as well as smaller enterprises that just can't afford extraordinary AI costs.  Some larger companies want to retain relationships with Anthropic and OpenAI because of strategic partnership interests and the brand umbrella, but even Microsoft is reportedly testing Chinese models. The reality is that token-based revenue models have altered the affordability equation. The result will thus inevitably be that less expensive options will have to be embraced for AI use to continue growing. JL

Viola Zhou reports in Rest of World:

While an hourlong coding session costs $10 on Claude, the same work cost less than 50 cents on DeepSeek. U.S.-based developers and small companies are turning to Chinese AI to cut costs. Although Chinese models still lag behind the best US ones in performance, they can handle most tasks at a fraction of the price. DeepSeek’s share of token usage jumped from under 1% to 17% in May. The affordability is especially attractive to independents and startups as well as "medium-sized businesses starting to get into AI but are wary of the costs.” Large companies, especially those in highly regulated industries, are still wary about Chinese models due to concerns about data security, censorship, and geopolitical risk. “Chinese models are becoming part of the global AI infrastructure layer. But usage is only the first step. The next test is whether the Chinese can convert usage into revenue, enterprise trust, and durable distribution.” 
Stu Clott, an operations manager and part-time developer in San Diego, used to code with Claude. But he recently found a cheaper alternative: DeepSeek. 

While an hourlong coding session would cost about $10 on Claude, the same work cost less than 50 cents on DeepSeek, Clott said. Over the past few weeks, he has used the Chinese model for everything from coding to personal counseling and building software to manage his family’s bank accounts. 

“I laugh every time I go see [the costs],” Clott told Rest of World. “The output quality, to be honest, I can’t tell the difference.” 

U.S.-based developers and small companies are turning to Chinese models to cut costs. Although Chinese models still lag behind the best American ones in performance, they can handle most tasks at a fraction of the price. Microsoft is also exploring using DeepSeek, or another open-source model, as a lower-cost alternative for Copilot Cowork, which is currently powered by Anthropic and OpenAI systems. But under political scrutiny, Chinese companies still face the challenge of turning that popularity into significant revenue.

Chinese companies have kept prices low thanks to lower salaries and infrastructure expenses at home. Many have also released open models and offered subsidized plans to attract early adopters. Models from DeepSeek, Xiaomii MiMo, and Minimax are now among the most cost-efficient available, according to an Artificial Analysis index that measures models’ performance against cost. 

The affordability is especially attractive to independent developers and startups. Lindy, a San Francisco-based company that builds AI work assistants, recently made a switch from Anthropic models to DeepSeek, according to its founder Flo Crivello, who announced the move on X in June. Crivello said the switch saved the firm millions of dollars. “You don’t need God to write your email,” he said on tech news show MTS. “If you can get those lower tiers of intelligence for a tenth of the price, it would be foolish not to do it.” 

Ruben Garcia Jr., a Dallas-based developer who builds websites and mobile apps for businesses, told Rest of World he uses both U.S. and Chinese models to run the AI agents that fulfill client requests automatically. 

He pays $500 a month for Claude and ChatGPT for the most complex planning and reviewing tasks, and another $200 a month for Minimax, Moonshot’s Kimi, and Xiaomi MiMo — models that handle 90% of the tasks, such as coding and voice recognition. 

On OpenRouter, a platform that helps developers direct tasks to different AI models, those from DeepSeek, Tencent, Minimax, and Xiaomi are now the four most popular. Vercel, another AI service provider, said DeepSeek’s share of token usage jumped from under 1% to 17% in May, although its share of revenue stayed near 1%. Tokens are the units of text or data that AI models process. 

“[AI] adoption at large companies is fairly saturated,” Kyle Chan, a fellow at the Brookings Institution, told Rest of World. “The growth market for Chinese companies would be medium-sized businesses that are starting to get into AI but are wary of the costs.”

In the U.S. market, however, Chinese model developers still face an uphill battle turning their popularity into revenue.

Companies that use Chinese models have come under political scrutiny. Lawmakers launched investigations into Airbnb and Anysphere, owner of coding platform Cursor, after the companies disclosed they had used Chinese open models like Qwen and Kimi to build their AI infrastructure. Airbnb chief executive Brian Chesky later clarified that the company was not sending any data to the model developers. 

If the Chinese models come out and they are frontier and cheaper, I’m going that direction.”Ruben Garcia Jr., Dallas-based developer

Large companies, especially those in highly regulated industries, will be reluctant to use Chinese models due to concerns about data security, censorship, and geopolitical risks, Poe Zhao, Beijing-based founder of China tech newsletter Hello China Tech, told Rest of World. 

Businesses that do use Chinese models are trying to have their data processed in the U.S. Some run open-source models on their own servers, and others have accessed them through American cloud companies rather than paying Chinese developers directly. Lindy’s founder Crivello said on X that his company accessed DeepSeek through an American provider, after a commenter questioned if user data was being sent to China. 

These cloud providers may share revenue with Chinese AI companies, but this arrangement prevents the companies from building long-term relationships with American clients, Zhao said. “Chinese models are becoming part of the global AI infrastructure layer,” he said. “But usage is only the first step. The next test is whether Chinese model developers can convert that usage into revenue, enterprise trust, and durable distribution.” 

Chinese companies could also lose some competitive edge as U.S. rivals join the price war. OpenAI is now considering significant price cuts as it competes with Anthropic for enterprise users, The Wall Street Journal reported.

For now, developers say Chinese models still provide the best value for what they are paying. Clott and Garcia said they didn’t believe U.S. companies would do any better at protecting user privacy than Chinese companies. 

“I don’t mind Chinese [companies] looking at my data: Hey, have fun with it. Learn and get better,” Garcia said. “If the Chinese models come out and they are frontier and cheaper, I’m going that direction.”