Chinese Starting Models for AI Startup Startups after success Deepseek

Chinese startup artificial intelligence renovate their business models when they fight for competitiveness after the broad adoption of Deepseek technology across the country.

According to two people, acquaintances with this issue, Zipu, who considered the largest linguistic model in China, caused his hopes for the original public proposal to maintain monetary growth.

Among China other leading generative ii Startups, 01.i stopped “previous training” of large language models to focus on selling individual business decisions using Deepseek models; Baychuan decided to focus on the health market; And moonshine has reduced its marketing budget on its Kimi Chatbot to focus on model training.

People close to these companies who have all refused or did not respond to a comment request, stated AI Industry.

Since the launch of its breakthrough R1 at the end of January, a startup founded in Hangzhou Adoption of lightning From her technology everywhere, from hospitals to local authorities.

This has left some of the starting AI countries, which in the last two years have received considerable support from domestic investors within the AI-re-re-evaluation strategies, seeking to repeat Deepseek’s success.

“The Chinese LLM market is quickly consolidated around a handful of leaders,” said Van Tuzhan, engineer AI Research Hub, which hugs the face. “Deepseek has pushed many companies to redirect resources to applications rather than the development of the fundamental model.”

On the basis of Beijing 01.Ah, founded by a venture capitalist and former head of Google China Kai-Fu Lee, completed his business in what he called “an era of depth”.

A group that launched a series of open source models, called Yi, stopped training in which developers use massive data sets for training-at the end of 2024 because of the cost increase, as its competitors trained increasing and more powerful models. According to Alibaba, its fundamental model was transferred to the Internet gigan, according to the people familiar with this issue.

Last week, 01. and announced that it would sell individual AI solutions for companies wishing to deploy Deepseek models. 01.

Instead of training one “thick model” on a wide database in which data from the Internet and other sources are cleaned, this approach combines numerous smaller models prepared according to the industry. The Moe structure allows companies that do not have chips, train large models at less computing, but for deployment can be more complex developers.

Deepseek who decided Focus on research Instead of seeking to maximize income by selling applications to companies, they left a space that will fill out the mediators as 01.AI. Baidu’s Internet Giant has also offered the same service in recent weeks.

Last year, Moonshot drew attention for the Viral AI Chatbot Kimi, but its popularity suffered after frequent shutdowns and rivals that launch competitive products.

In recent weeks, the startup has reduced Kimi marketing expenses as it increased its attention to the modeling training to repeat the success of the Deepseek breakthrough and improve the work of Chatbot, according to two people acquaintance with this issue.

But since Kimi is overtaken by other applications, Moonshot makes an uncertain future because it burns through cash, engaged in models without stable profits. The startup sought to make money by inviting users to send virtual gifts “Kimi”, AI Hero for Chatbat.

Last year, it raised more than $ 1.3 billion through two investment rounds, a mixture of computing Chinese technology giant Alibaba and CASH from Venture capital, people familiar with transactions reported.

At the beginning of 2024, Alibaba considered the moonshine a potential purpose of acquiring and ensured the first right to buy a startup in any future sale within its $ 800 million, people said. In recent months, Alibaba has been based on starting investments after Jack Ma founder managed Eddi Vo Woo to focus on AI’s internal efforts. The change makes less likely that Alibaba will seek to buy Kimi in the future, people added.

The Baichuan startup, based in Beijing, doubled his health care business after he previously worked on a consumer, which is facing AI Chatbots and enterprises for educational, financial and medical companies.

In February, Baichuan rejected its trading group, focused on selling its individual financial application to AI in banks and investment funds and graduated from the business lining, said two people familiar with this issue.

At the time, the company management announced to employees that it focused on the development of its technology for hospitals, which includes Doctor II, which provides assistance in diagnosis.

Unlike this, the cake based on Tan Ji, a well -known computer scientist at Tinhua University, still holding several business lines. He launched several consumer applications, as well as an enterprise that sells personalized applications by AI to local authorities and companies, the notorious competitive and low -report business in China.

The startup is burned at the expense of cash because it creates its business for the sale of businesses. In 2024, Ziphu made RMB300MN ($ 41 million) sales and RMB2BN for loss, according to three investors informed about the figures.

The cost of the balloons caused concern about some investors after Deepseek demonstrated the way to build advanced models for a smaller budget. Unlike a small Deepeseek labor force, which is about 160 employees, Ziphu works about 800 people, making it the largest LLM startup in number.

According to two people, acquaintances with this issue, Zipu hopes for raising cash after receiving one of Beijing’s cherished letters for IPO. The company requires approval from the regulators before it can transfer the list in the field of innovative technology councils.

The startup received a Beijing nod before Deepseek changed the competitive landscape of the II players in China. Earlier, the cinnamon told investors what he was trying to list by the end of the year, said two people with knowledge of this issue. But they added that Deepseek developments could affect this when they would go forward with IPO.

According to two people, acquaintances with this issue, investors in the company also expressed concern about the fact that government hugs Deepseek could threaten the ZIPHU business.

But Deepseek shook the AI ​​race in China, which led to some competitors to decide whether to challenge the group directly or accept their open source models to focus on the smaller potential market.

“By accepting higher level models, companies can eliminate the need to invest tens of millions of dollars annually into the training lower alternatives,” Vanga said.