Developers in China are creating solutions that cost less money and require less sophisticated hardware.



One of the challenges of AI adoption is money: Building large language models is expensive. That assumption is being challenged by the efforts of Chinese developers who are creating approaches that cost far less money and require less sophisticated hardware.

The notion of cheaper AI was enough to rattle the stock market Monday as investors digested the possibility that models built by the likes of OpenAI, Google and Anthropic may be undercut by new products coming out of China.

According to media reports, the Chinese AI company DeepSeek unveiled DeepSeek-R1, an open-source reasoning model that, it claims, compares well to OpenAI’s 01 in performance tests. Last month, DeepSeek launched an open source large language model that it said took only two month and $6 million to create. The model is also open-source, which is bound to attract developers and other technology specialists.
Upgrade to paid
Another Chinese company, TikTok-owner ByteDance, released an upgraded its Doubao-1.5-pro model, which it said performs better than OpenAI o1. Experts said such models could challenge the business of better-known AI companies in both performance and cost. One version of the product costs 2 yuan per million (or about 28 cents) of output, while a more powerful version costs 9 yuan (or $1.24). OpenAI charges 438 yuan (or $60.41). (Tokens represent units into which text data is broken down for use in AI.)

DeepSeek’s app is now the most-downloaded from Apple’s App Store. It’s competing against tools from the likes of Google, OpenAI and Meta, whose development costs were in the hundreds of millions of dollars. As The Wall Street Journal observed, “But DeepSeek didn’t have that financial firepower — and its models are still roughly on par with top U.S. rivals.”

Lower costs are sure to impact AI if the price of solutions falls to a level in reach of smaller companies and less-sophisticated development teams, especially when they don’t require much of a performance tradeoff. DeekSeek’s new model is “super impressive in terms of both how they have really effectively done an open-source model … and is super-compute efficient,” Microsoft CEO Satya Nadella told CNBC. “We should take the developments out of China very, very seriously.”