Four Startups Are Leading China’s Race to Compete with OpenAI’s ChatGPT

Four Chinese generative artificial intelligence startups have been valued at between $1.2 billion and $2.5 billion in the past three months, leading a group of more than 260 companies tracking U.S. success – Want to emulate rivals like OpenAI and Anthropic.

The newly formed unicorns – Zhipu AI, Moonshot AI, MiniMax and 01.ai – have received significant support from a predominantly domestic investor pool and are racing to hire the best talent to develop the most popular AI products.

“There is still no winner of foundation models in the Chinese market. These are some of the names leading the charge for this title,” said Charlie Dai, vice president and principal analyst at technology-focused consulting firm Forrester.

In terms of technological development and overall fundraising, US startups outperform Chinese generative AI startups. But with ChatGPT and other groundbreaking AI applications like Character.ai unavailable in China, 262 startups are vying to launch homegrown alternatives, according to a tally by data provider IT Juzi.

According to IT Juzi, China’s generative AI startups raised a total of RMB 14.3 billion ($2 billion) in donations in the first four months of the year, defying a slowdown in investment in other consumer technology areas. However, many startups have not disclosed all of their fundraising rounds, protecting their war chests from rivals rushing to hire talent and compete for limited computing resources.

Beijing has approved more than 40 major language models and related AI applications for public use, while creating a supportive regulatory environment that encourages the sector’s growth through tax breaks and subsidies.

Zhipu has become China’s largest AI start-up by number of employees. Emerging from Beijing’s prestigious Tsinghua University, a major breeding ground for AI talent, the company, which has over 800 employees, is worth RMB 18 billion (US$2.5 billion), based on its valuation, according to two of its investors last round of financing in March. They did not disclose how much they had raised.

Founded by Yang Zhilin, a former student of a Zhipu founder, Moonshot was valued at $2.5 billion in a $1 billion investment round announced in February. Previously, Yang interned at Google Brain AI and Meta AI and founded a startup called Recurrent AI that analyzed calls from salespeople.

Moonshot, Zhipu and 01.ai have developed chatbots aimed at office workers and students who use the digital assistants to process long texts and optimize search results. Neither Moonshot nor Zhipu responded to requests for comment on their funding.

Moonshot’s chatbot Kimi, which takes its nickname from Yang’s English name, has become the closest competitor to Chinese internet search giant Baidu’s Ernie Bot. According to data provider Aicpb.com, Kimi received 12.6 million visits in March compared to 14.9 million visits for its more established competitor, but Kimi is growing much faster.

“Kimi has done a good job with the user interface and contextualization of the content through RAG,” said an industry insider, pointing to the retrieval augmented generation technique that allows the models to retrieve data from external sources and provide up-to-date answers to the queries of the User.

But Kimi has become a victim of its popularity and is struggling to keep up with demand as users flock to the easy-to-use chatbot, which industry experts have praised for its aggregation tool and delivery of clear, highly contextualized answers. In March, Kimi Bot experienced a two-day outage, prompting the young company to apologize.

Facing the same limited computing resources, many AI startups have opted to launch avatar chatbots, which don’t need to be as good at thinking as productivity chatbots. These chatbots are trained on smaller amounts of data and therefore require fewer computing resources.

“ChatGPT is hard to copy. The model is the product. It’s easier to create an avatar chatbot; “You can do it on an open source model with limited computing resources,” said an AI researcher in China.

Zhipu and MiniMax both have avatar chatbots that target the world’s largest gaming market with anime-style characters for banter and flirting, while generating user feedback to iterate on their models. Shanghai-based MiniMax is valued at $2.5 billion, based on a $600 million funding round announced in March.

01.ai, founded by AI pioneer Kai-Fu Lee, has launched a series of open source models called Yi tailored to the Chinese market and based on Meta’s freely customizable Llama architecture. Hugging Face, which tracks open source models, has placed several iterations of the Yi models at the top of its healthy thinking, math, coding and reading rankings. 01.ai has also launched a productivity chatbot called Wanzhi.

The AI ​​startup recently raised a $1.2 billion funding round and is backed by Lee’s own venture capital fund Sinovation Ventures, Shunwei Capital, Xiaomi and Alibaba Cloud, according to a person with direct knowledge of the deal.

With no clear market leader or “killer app” emerging in China, it is difficult for customers to choose the right provider, said Jeffrey Ding, an expert on China’s AI ecosystem and assistant professor at George Washington University. “It is very difficult to distinguish between all these similar applications. “How do you know which company will be more effective for your particular situation,” he said.

All four leaders have received funding from Alibaba, which has become a major backer of AI startups as it looks to repeat the success of Microsoft’s big bet on OpenAI.

But the pool of investors funding this group of startups is smaller than before because global technology funds are not playing the role they did in the last generation of AI surveillance startups, including SenseTime and Megvii.

“Foreign investors are largely sitting this one out,” noted one AI investor.

In addition to sufficient funding, Chinese AI startups are finding that they have both the technical talent needed to bring competitive products to market and sufficient computing resources to train existing models – despite US restrictions on the Exporting advanced chips.

“Chinese cloud computing companies have sufficient stocks of Nvidia GPUs purchased before the ban for startups to participate in this model training round,” said an employee of an AI startup that covers the terms of cloud computing. Negotiated computing contracts.

However, two other AI startup employees said they are targeting products that require less computing power as the industry adapts to U.S. restrictions. They added that limited computing resources are a key reason why many Chinese AI startups rely on open source models like Metas Llama to build their own models and applications, rather than using valuable computing resources and talent to develop a proprietary one Having to scratch to insert the model.

“Chinese AI startups do a lot of research and development based on the open source architecture, then refine the model and contribute to the entire ecosystem,” Forrester’s Dai said. “These startups do not need to do model design and validate the architecture, which is very resource intensive,” he added.

Low salaries for AI engineers compared to the US and Europe also help keep costs down. A newly minted computer science graduate student from a Chinese university can typically expect to earn between $80,000 and $240,000 a year at a large startup, a range that is about four times higher in Silicon Valley, according to several people familiar with the matter familiar people.

“We are growing so fast,” said an employee at one such startup, noting that the rapidly growing product and sales team allowed the company to further commercialize its technology and attract new customers.

As one deep tech investor in China said, “Chinese companies are not good at basic technology itself.” But they are very good at watching industry trends, following whatever comes out, and capitalizing on their engineering dividend, to follow everything that was created in the USA.”

Additional reporting by Andy Lin in Hong Kong