The race to build the next big language model is on, and a Chinese candidate has reportedly raised a massive amount of money to jump to the top.
Moonshot AI, a year-old artificial intelligence startup that built an LLM that can process long-duration text and data inputs, raised more than $1 billion in a Series B round, according to multiple media reports from China. It is said that they have procured it. If accurate, this latest capital infusion would value Moonshot AI at his $2.5 billion. This is his single largest funding round for his LLM developer in China on public record.
The startup, called YueZhiAnMian in China, focuses on developing language models at scale, like many companies in the AI industry today. Specifically, its unique selling point is its commitment to handling long text contexts and responses. This sector has been outperforming others in this field for a long time.
The company is moving quickly to roll out initial efforts to address this.
In March last year, the startup announced the 50th anniversary of Pink Floyd's “Dark Side of the Moon,” founder Yang Zilin's favorite album and the inspiration for the startup's name. was launched using a 100 billion parameter LLM.
And in October, Moonshot announced China's first chatbot, Kimi. It claims to be able to support the processing of 200,000 kanji characters in a single conversation, which is said to be eight times the length that OpenAI's GPT-4-32K can achieve.
We have reached out to AI researcher and academic Yang Zhilin, who co-founded Moonshot with Zhou Xinyu and Wu Yuxin, for comment and will update this post if we hear back.
Meanwhile, the funding is reportedly coming from a list of big-name investors, including a number of potentially interesting strategic partners. The round is co-led by e-commerce giant Alibaba and venture capital firm Hongshan, formerly known as Sequoia China, according to the South China Morning Post. Also included in the round were Chinese “super apps” Meituan and Xiaohongshu (also known as China's answer to Instagram), according to Chinese technology blog LatePost.
Moonshot previously raised a $200 million round from HongShan and Zhen Fund at a valuation of $300 million, according to PitchBook Data.
Hongshan declined to comment on the report when asked for comment. Alibaba did not respond to requests for comment. We also reached out to Moonshot separately.
If accurate, Hongshan's involvement here would be noteworthy. Sequoia Capital formally announced last year that it would separate its Asian operations in India and China amid heightened geopolitical tensions. This process is expected to be finalized by March 24th.
But on the other hand, China's activities have come under scrutiny from the U.S. government regarding AI transactions in the United States. And American companies are similarly being investigated for their continued activities in China. With all this in mind, it's no wonder that once famous investors are holed up here.
Importantly, the rest of the reported investor list is a bona fide name for the home technology company. This confirms that financial investors are continuing to retreat from promising Chinese startups, especially those from Western countries trading in US dollars, or at least have paused.
But at the same time, as we've seen unfold in the US, where companies like Microsoft, Google, and Amazon are pumping billions into LLM startups like OpenAI and Anthropic, we're seeing how big Chinese tech companies are It also shows that they are fighting for their own profits. AI will play a huge role in the coming months and years. Gaining a financial foothold in a small number of promising companies is one way to shortcut or enhance what they are trying to build internally.
Currently, OpenAI rules the roost in the US and every other country where OpenAI is expanding, but since China lacks an anointed leader, much of the investment activity is spreading the bet. It feels like that's the case.
To this end, Alibaba is also an investor in Baichuan, which, like Zhipu, was founded by Wang Xiaochuan, a pioneer in the search engine field, and which had raised $350 million by the end of last year and whose It passed a $1 billion valuation in the process. He has one more LLM startup, AI, and 01.AI, an LLM company founded by Kai-Fu Lee.
Meanwhile, Alibaba's biggest rival, Tencent, backs Baichuan, Zippu, MiniMax, and Lightyears Beyond. Don't get me wrong. Chinese internet giants are backing LLM candidates in the country on behalf of Western venture capital firms.
Still, if $1 billion sounds like a lot for a startup that's less than a year old, pedigree may be one reason major companies are willing to take big bets.
Pink Floyd fans become AI pioneers
Yang Zhilin had a long list of accomplishments to his credit even before he started Moonshot.
He holds a PhD in computer science from Carnegie Mellon University, where he was advised by Ruslan Salakudinov. He previously led AI research at Apple after the iPhone maker quietly acquired a startup he founded called Perceptual Machines, which appears to have gone unreported. However, the professor is listed on his LinkedIn profile and professional timeline.
Before that, he studied at Tsinghua University under the advice of Jie Tang. He also worked with his Google Brain and Meta AI.
Yang is also developing another AI startup called Recurrent.AI. This seems to be focused specifically on technology built to improve the work of salespeople (with similar functionality to Gong.AI). As of 2021, Recurrent has raised approximately $60 million, according to PitchBook. Since then, business appears to have continued, although there has not been much capital activity.
Importantly for Moonshot, Yang was also the primary author of Transformer-XL. Transformer-XL is a significant development in the LLM architecture that enables natural language understanding beyond fixed-length contexts, and has played a key role in the development of Moonshot's platform and perhaps its broader mission. .
Moonshot is widely leveraged in existing LLM and generative AI applications, such as legal documents and fiction, by focusing on longer inputs and outputs and producing more accurate results for related queries. It lays the groundwork for the company to target text-based use cases where it doesn't. Writing and more detailed financial analysis. According to the chatbot, KimiChat is trained on information up to January 2024.
It's not the only Chinese player working to remove long context constraints. Baichuan He announced the Baichuan2-192K model in October. This model is said to process approximately 350,000 Kanji characters in one context window.
Global funding markets remain constrained, but this round demonstrates the willingness of those with deep pockets to jump in when the right opportunity presents itself. But despite a broader global AI frenzy that Goldman Sachs predicts will see around $200 billion invested by 2025, China's funding environment has been surprisingly lukewarm.
According to research firm CBInsight, China recorded about 232 investments in the AI field in 2023, a decrease of 38% from the previous year. The total amount raised by Chinese AI companies was about $2 billion, down 70% from the previous year.