A Chinese rival has apparently raised a huge sum of cash to advance in the race to construct the next great broad language model.
According to Chinese media sources, Moonshot AI, an artificial intelligence business created less than a year ago to construct LLMs that can handle extensive text and data inputs, received over $1 billion in Series B funding. Moonshot AI may have raised $2.5 billion, the highest single financing deal for Chinese LLM developers.
Like many AI startups, YueZhiAnMian in China has been developing massive language models. Its USP is its ability to comprehend long-form context and reaction, which has long eluded researchers.
It’s launching its initial measures to solve this immediately.
Last March, the firm started with a 100 billion-parameter LLM to celebrate Pink Floyd’s The Dark Side of the Moon, creator Yang Zhilin’s favorite album and the startup’s moniker.
Moonshot debuted its first Chinese chatbot, Kimi, in October, claiming to comprehend 200,000 Chinese characters in a single conversation, eight times longer than OpenAI’s GPT-4-32K.
We will update this page if Yang Zhilin, the AI researcher and scholar who developed Moonshot alongside Zhou Xinyu and Wu Yuxin, answers.
Several big-name investors, including possible strategic partners, are backing the project, according to reports. The South China Morning Post reports that Alibaba and HongShan, previously Sequoia China, are heading the round. According to Chinese tech site LatePost, Meituan and Xiaohongshu—China’s Instagram—were also in the round.
Moonshot secured $200 million from the HongShan and Zhen Fund for a $300 million valuation, according to PitchBook Data.
HongShan will not comment on the reports. Alibaba has not replied to inquiries. We contacted Moonshot individually.
An authentic Hongshan participation would be significant. Last year, Sequoia Capital stated it would divide its India and China businesses amid escalating geopolitical concerns. That procedure should finish on March 24.
However, the U.S. government is investigating the China operation for its AI transactions in the U.S. and the U.S. business for its China operations. Given this, the once-high-profile investor may be hiding.
Importantly, the remainder of the investment roster is made up of tech titans. Financial investors are retreating from potential Chinese businesses, particularly those from the West and using U.S. money.
However, it shows how, just as Microsoft, Google, and Amazon have invested billions in LLM startups like OpenAI and Anthropic in the U.S., big tech companies in China are scrambling for their next AI play. A financial stake in a few promising candidates may help them grow in-house or supplement their efforts.
OpenAI dominates in the U.S. and wherever it has grown, but in China there is no clear leader, so investment activity seems like spreading bets.
So, Alibaba invests in Baichuan, founded by search engine pioneer Xiaochuan Wang, which raised $350 million by the end of last year, passing a $1 billion valuation; Zhipu AI, another LLM upstart; and 01.AI, founded by Kai-Fu Lee.
Tencent, Alibaba’s opponent, supports Baichuan, Zhipu, MiniMax, and Light Years Beyond. Internet companies have supplanted western-money VCs in financing LLM candidates in China.
However, while $1 billion seems like a lot of money to give a firm less than a year old, prominent names may be ready to take significant risks because of their pedigree.
Before launching Moonshot, Yang Zhilin, a Pink Floyd aficionado and AI pioneer, had a lengthy record of accomplishments.
Ruslan Salakhutdinov, who previously headed up AI research at Apple, advised him on his computer science PhD at Carnegie Mellon University. Apple quietly acquired Perceptual Machines, a startup he founded, which is noted in the professor’s LinkedIn profile and professional timeline.
He studied at Tsinghua University under Jie Tang. He works at Meta AI and Google Brain.
Yang is also working on Recurrent, an AI company. AI looks to concentrate on technology to assist salespeople in doing their jobs better (with capabilities similar to Gong.AI). Recurrent reportedly raised $60 million in 2021, according to PitchBook. Since then, capital activity has decreased, but the firm seems to be running.
Yang was also a key author of Transformer-XL, an LLM architecture development that enabled natural language understanding beyond a fixed-length context, which helped Moonshot develop its platform and mission.
Moonshot can target text-based use cases like legal documents, fiction writing, and deeper financial analysis that LLMs and generative AI applications haven’t fully addressed by focusing on longer input and output and producing more accurate results. The chatbot claims Kimi Chat is trained using January 2024 data.
There are other Chinese companies trying to eliminate lengthy contexts. In October, Baichuan introduced their Baichuan2-192K model, which processes 350,000 Chinese characters in one context window.
The global fundraising market is limited, but this round shows that individuals with substantial wallets will leap in when the proper opportunity arises. Despite the global AI frenzy—Goldman Sachs predicts $200 billion in investment by 2025—China’s financial infrastructure is unexpectedly weak.
CBInsight reported 232 AI investments in China in 2023, down 38% year-over-year. China’s AI businesses raised $2 billion, 70% less than last year.
China’s Moonshot AI has reached a value of $2.5 billion, securing $1 billion in funding for an LLM program that focuses on lengthy context.
