Chinese models of artificial intelligence repair business models after Dibsic’s success

Chinese artificial intelligence companies fix their business models while fighting in order to stay competitive in the wake of the dummy adoption of competition technology in all parts of the country.

Zhipu, which was once considered one of the most prominent language model in China, has operated its hopes in a preliminary general offer to maintain its intense cash growth as it focuses on building institutions sales business, according to two people familiar with this issue.

Between China Other leading Amnesty International Startups, 01.AI stopped large language models “before training” to focus on selling artificial intelligence solutions designed using Deepseek models; Baichuan chose to focus on the healthcare market; Moonshot cut its marketing budget for Kimi Chatbot to focus on typical training.

People close to these companies, all of which were refused or did not respond to the requests for comment, said that the transformations show how Deepseek changed the prosperous form of China Amnesty International industry.

Since its R1 launch in late January, Hangzhou, stationed quickly, has crowned the country’s hero in the country by Beijing and witnessed Adopting lightning Its technology everywhere from hospitals to local governments.

Some of the best startups in the country-which has gained great support during the past two years from local investors as part of the artificial intelligence boom-to reassess their current strategies in an attempt to repeat the success of Deepseek.

“The LLM market is quickly integrating around a handful of leaders.” “Dibsic has prompted many companies to redirect resources to applications instead of developing constituent models.”

Beijing-based 01.AI, which was founded by the investment capitalist and former Google China Kai-Fu Lee, evaluated its business in what he called “The Deepseek Age”.

The group, which launched a series of open source models called Yi, stopped before training-where developers use huge data collections to train models-in late 2024 due to high costs as their competitors trained larger and more powerful models. In a deal with alibaba, its primary model team was transferred to the Internet giant, according to the people familiar with the matter.

Last week, 01.AI announced that it would sell artificial intelligence solutions designed for companies that want to spread Deepseek models. 01.AI shows her experience in the so -called “a mixture of experts” as a competitive advantage, which is the way that Deepseek uses to train their models.

Instead of training a “thick model” on a vast database that has ridiculed data from the Internet and other sources, the approach combines many smaller models trained in the industry data. The MEE structure allows poor chips to train larger models on less computing power, but can be more challenging for third -party publication.

Depsik, who decided that Focus on research Instead of seeking to increase revenues by selling requests to companies, she left a gap filled by brokers like 01.AI. BAIDU, the Internet giant, has also provided the same service in recent weeks.

Moonshot crashed attention last year in favor of Ai Chatbot Kimi, but its popularity suffered after the power and competitors to launch competitive products.

In recent weeks, the startup company has reduced the marketing spending on KIMI as it increases its focus on the typical training to repeat the success achieved by Deepseek and improve the performance of Chatbot, according to two people familiar with the matter.

But as Kimi crosses through other applications, Moonshot is planning an uncertain future because it burns by conducting models training without stable revenues. The startup sought to earn money by inviting users to send virtual gifts to “KIMI”, the character of artificial intelligence behind Chatbot.

It raised more than $ 1.3 billion in financing through two investment rounds last year, with a mixture of computer credits from the Chinese technology giant alibaba and Cash from investment capital companies, according to persons familiar with deals.

In early 2024, Alibaba Moonshot considered a potential goal of acquisition and got the first right to purchase the startup in any future sale as part of their $ 800 million investments. In recent months, Alibaba has been affected by emerging investments, after the founder Jack Ma, CEO EDDIE WU, has instead on the efforts of internal artificial intelligence. People added that this shift makes it unlikely to seek Ali Baba to obtain Kimi in the future.

Beijing -based Baichuan, based in Beijing, has doubled in its health care after it has previously worked on consumers who face AI Chatbots and institutions’ business stadiums for educational, financial and health care companies.

In February, Baichuan refused that the sales team that focuses on selling artificial intelligence requests designed for banks and investment funds and ended the labor line.

At the time, the company’s leadership announced that it was focusing on developing its technology for hospitals, including Amnesty International’s doctor, helping the diagnosis.

On the other hand, ZHIPU, founded by Tang Jie, a prominent computer world from Tsinghua University, is still following multiple working lines. It has launched many consumer applications in addition to the business of companies that sell artificial intelligence applications for governments and local companies, a competitive and low margin company in China.

The emerging company was burning through cash, as it built institution sales business. In 2024, Zhipu Rmb300MN ($ 41 million) achieved sales and RMB2BN in losses, according to three investors launched by the numbers.

The costs of amplification have paid anxiety among some investors after Deepseek showed a way to build advanced models on a smaller budget. Unlike the small workforce in Deepseek, which includes about 160 employees, ZHIPU employs about 800 people, making it the largest LLM company by appointments.

Zhipu hopes to obtain a cash boost after receiving one of the desirable recommendations for Beijing to obtain public subscription, according to two people familiar with this issue. The company needs approval from the organizers before it can follow up on the Innovation Council, which focuses on technology.

The start -up company has received Beijing’s gesture before Dibsic changed the competitive scene of artificial intelligence players in China. Zhipu previously told investors that he was aiming at the list before the end of the year. But they added that Deepseek’s developments could affect him if they pushed forward with public subscription.

The company’s investors also expressed concern that the government’s embrace of Deepseek may threaten the ZHIPU business model of selling artificial intelligence solutions designed for local governments, according to two people familiar with this issue.

But Deepseek rocked the artificial intelligence race in China, prompting some competitors to determine whether the group will be challenged directly or the open source models to focus on a smaller possible market.

“By adopting the higher models, companies can eliminate the need to invest tens of millions of dollars annually in training the lower internal alternatives,” said Wang’s Wang’s Wang.