The beginnings of Chinese artificial intelligence are adjusting their business models as they struggle to remain competitive after wide adoption of the Deepseek rival technology across the country.
Zipu, once considered the most prominent beginning of China’s big language model, has supported her hopes on an initial public offer to maintain its money -intensity growth, as she focuses on building its own enterprise sales business, according to two people familiar with the matter.
Among the other generating beginnings of China’s generation, 01. He has stopped “pre-training” large language models to focus on selling adapted business solutions using Deepseek models; Baichuan has chosen to focus on the health care market; And Moonshot has reduced his marketing budget for his chemistry chatbot to focus on model training.
People near these companies, all of which did not accept or respond to commentary requests, said the shifts show how Deepseek has drastically changed the form of the growing industry of China.
Since the beginning of his model of Progress R1 at the end of January, the Hangzhou -based beginning was quickly crowned the country’s champion from Beijing and has seen the lightning of his technology everywhere from hospitals in local governments.
It has left some of the main beginnings of the one that over the past two years have gained significant support from local investors as part of its boom-to reassess their existing strategies in an effort to repeat Deepseek’s success.
“The Chinese market LLM is rapidly consolidating about a handful of leaders,” said Wang Tiezhen, an engineer at him Research Hub Hub hugging his face. “Deepseek has encouraged many companies to redirect resources in apps than developing the basic model.”
01. He based in Beijing, founded by the Entrepreneurship Capitalist and former Google Chief China Kai-Lee, has shared her business in what he has called “Epoka Deepseek”.
The group, which has launched a series of open-sourced models called YI, stopped the pre-trainim-developers using massive data groups to train models-at the end of 2024 due to increasing costs as its trained and more powerful models rivals. In an agreement with Alibaba, her team of the underlying model was transferred to the Internet giant, according to people familiar with the issue.
Last week, 01. He announced that he will sell adapted solutions to the companies that want to set Deepseek models. 01. He is throwing her expertise on the so -called “mix of experts” as its competitive advantage, the method also used by Deepseek to train its models.
Instead of training a “dense model” on a wide database that has removed data from the Internet and other resources, the access combines many smaller models trained in specific industry data. The MOE structure allows poor companies with chips to train larger models in less computing power, but can be more challenging to establish third -party developers.
Deepseek, who has decided to focus on research rather than seeking to maximize income by selling applications for companies, has left a gap to be supplemented by brokers like 01.ai. The Baidu Internet giant has also led to providing the same service in recent weeks.
Moonshot judged his attention last year for that viral chemistry chatbot, but his popularity has suffered after frequent interruptions and rivals who begin competing products.
In recent weeks, the beginning has reduced marketing costs for chemistry as it increases its focus on training the model to repeat Deepseek’s broken success and improve its chatbot performance, according to two people familiar with the issue.
But while chemistry is overcome by other applications, Moonshot is designing an unsafe future as it burns through model training by making money without sustainable income. The beginning has asked to make money by inviting users to send virtual gifts to “Kimi”, the character of he after chatbot.
It raised more than $ 1.3BN in funding through two rounds of investment last year, with a mix of computing loans from Chinese technology giant Alibaba and cash by entrepreneurial capital firms, according to people familiar with agreements.
At the beginning of 2024, Alibaba considered Moonshot a possible purchase target and secured the first right to buy the start on any future sales as part of their $ 800 million investment, people said. In recent months, Alibaba has renewed the initial investments, after founder Jack Ma addressed the chief executive Eddie Wu to focus on this in his internal efforts. The change makes Alibaba less likely to seek to get chemistry in the future, people added.
Behachuan -based starting has doubled in his health care business after previously working on consumers facing herself and business companies for educational, financial and health care companies.
In February, Baichuan dismissed his sales team focused on the sale of his adapted financial app in bank and investment funds and completed the business line, two people familiar with the issue said.
At the time, the company’s leadership informed employees that it was focusing on developing its technology for hospitals, which includes a doctor who helps with the diagnosis.
Some content could not load. Check your internet connection or browser settings.
In contrast, Zipu, founded by Tang Jie, a prominent computer scientist from Tsingghua University, is still following numerous business lines. It has launched some customer applications, as well as a business business that sells personalized applications for local governments and companies, a highly competitive business with low margins in China.
The start has been burnt with money while building the business of sale of its enterprises. In 2024, Zipu made RMB300MN ($ 41 million) on sale and RMB2BN in losses, according to three investors informed about the figures.
Ballooning costs have sparked concern among some investors after Deepseek demonstrated a road to build front models with a smaller budget. Unlike Deepseek’s small labor force with about 160 employees, Zhipu employs about 800 people, making it the biggest beginning of the headCount.
Zipu is hoping for a cash growth after receiving one of the recommended coveted Beijing letters for an IPO, according to two people familiar with the matter. The company needs approval by regulators before it can follow a list on board the Innovation of Stars focused on technology.
Beginning took the Beijing node before Deepseek changed the competitive landscape of the players in China. Zipu previously told investors that he was aiming to list before the end of the year, two people with knowledge of the matter said. But they added that Deepseek’s developments could affect it if they pushed forward with an IPO.
Investors in the company have also expressed concern that embracing Deepseek’s government could threaten Zhipu’s business model to sell it to local governments, according to two people familiar with the issue.
But Deepseek has rocked the race in China, causing some rivals to decide whether to challenge the group directly or approve its open source models to focus on a smaller possible market.
“By adopting high -level models, companies can eliminate the need to invest tens of millions of dollars a year in training inferior alternatives at home,” Hugging Face’s Wang said.
