The race to build the next big large language model is on, and now a contender out of China has reportedly banked a major round of funding to catapult it to the front of the pack.
Moonshot AI, an artificial intelligence startup founded less than a year ago building LLMs that can handle long inputs of text and data, has raised over $1 billion in a Series B round, according to multiple media reports out of China. If accurate, this latest injection of capital would value Moonshot AI at $2.5 billion — the largest single funding round for Chinese LLM developers on public record.
The startup — which goes by YueZhiAnMian in China — has been focused, like many in AI right now, on the development of large language models. Specifically, its unique selling point is that it’s working on being able to process long-form context and response, an area that has long outfoxed others in the field.
It’s been moving quickly to roll out its first efforts to address this.
In March of last year — to coincide with the 50th anniversary of Pink Floyd’s The Dark Side of the Moon, founder Yang Zhilin’s favorite album and the inspiration for the startup’s name — the startup launched with a 100 billion-parameter LLM.
Then in October, Moonshot launched its first chatbot in China, Kimi, which claims to be able to support the processing of 200,000 Chinese characters in a single conversation — allegedly eight times the length of what OpenAI’s GPT-4-32K can achieve.
We have contacted Yang Zhilin, the AI researcher and academic who founded Moonshot with Zhou Xinyu and Wu Yuxin, for comment and will update this post if and when he responds.
Meanwhile, the reports say the funding is coming from a list of big-name investors that include a number of potentially interesting strategic partners. Co-leading the round are e-commerce giant Alibaba and HongShan — the VC firm formerly known as Sequoia China, according to South China Morning Post. Others in the round included Chinese “super app” Meituan and Xiaohongshu (sometimes called China’s answer to Instagram), according to Chinese tech blog LatePost.
(Bloomberg reported a week after the funding news first emerged that Alibaba led the $1 billion round with Monolith Management participating.)
Moonshot previously raised $200 million from HongShan and Zhen Fund in a round that valued it at $300 million, according to PitchBook Data.
HongShan, contacted for comment, declined to comment on the reports. Alibaba has not responded to a request for comment. We also separately reached out to Moonshot.
If accurate, HongShan’s involvement here would be notable. Sequoia Capital formally announced last year that it would split off its Asian operations in India and China amid rising geopolitical tensions. That process is due finally to be completed by March 24.
But in the meantime, the China operation has come under scrutiny from the U.S. government for its AI deals in the U.S.; and the U.S. firm is equally being probed for ongoing activity in China. Given all this, it’s no surprise that the once high-profile investor may be laying low here.
Significantly, the rest of the investor list that’s been reported is a veritable who’s who of household tech names. This underscores the ongoing retreat — or at least pause — among financial investors into promising Chinese startups, especially those out of the West and doing deals in U.S. dollars.
But it also points to how — just as we have seen play out in the U.S., with companies like Microsoft, Google and Amazon ploughing billions of dollars into LLM startups like OpenAI and Anthropic — big tech companies in China are scrambling for what might be their big AI play in the months and years ahead. Having a financial foothold in a handful of promising hopefuls is one way to shortcut, or augment, whatever they may be trying to build in-house.
OpenAI rules the roost right now in the U.S. and arguably everywhere else that it has expanded, but in China there is no anointed leader, so a lot of the investment activity feels like it is about spreading bets.
Alibaba, to that end, is also an investor in Baichuan — founded by Xiaochuan Wang, a pioneer in the search engine space, which raised $350 million by the end of last year, passing a $1 billion valuation in the process — as well as Zhipu AI, another LLM upstart, and 01.AI, the LLM company founded by Kai-Fu Lee.
Alibaba’s archrival Tencent meanwhile has backed Baichuan, Zhipu, MiniMax and Light Years Beyond. Make no mistake: China’s internet giants have replaced those western-money VCs in backing the country’s LLM contenders.
All the same, if $1 billion sounds like an awful lot of money to hand over to a startup that is less than a year old, one of the reasons big names might be willing to take big bets is because of the pedigree.
Pink Floyd fan turned AI trailblazer
Yang Zhilin had a long list of achievements to his credit even before starting Moonshot.
He holds a computer science PhD from Carnegie Mellon University, where he was advised by Ruslan Salakhutdinov, who previously headed up AI research at Apple after the iPhone maker quietly acquired a startup he founded called Perceptual Machines — an acquisition that seems never to have been reported but is noted in the professor’s LinkedIn profile and professional timeline.
Prior to that, he studied at Tsinghua University, advised by Jie Tang. He’s also worked at Google Brain and Meta AI.
Yang also has another AI startup in the works, Recurrent.AI, which appears to be focused specifically on technologies built to help salespeople to their jobs better (with features that sound not dissimilar to say, Gong.AI). As of 2021, Recurrent had raised some $60 million according to PitchBook. And while there hasn’t been much capital activity since then, the business does appear to remain operational.
Importantly for Moonshot, Yang was also a key author of Transformer-XL, a key development in LLM architecture for enabling natural language understanding beyond a fixed-length context, something that has played a significant role in the development of Moonshot’s platform and, arguably, wider mission.
Moonshot’s focus on longer input and output and producing more accurate results for queries involving this lays the groundwork for the company to target text-based use cases that haven’t been widely tapped by existing LLMs and generative AI applications, such as legal documents, fiction writing and deeper financial analysis. Kimi Chat is trained on information up to January 2024, the chatbot says.
It’s not the only Chinese player working to remove the limits of long context. Baichuan back in October announced its Baichuan2-192K model, which is said to process around 350,000 Chinese characters in one context window.
The market for fundraising remains constrained globally, but this round speaks to the willingness of those with deep pockets jump in, when the right opportunities come up. Yet even with the wider, global AI frenzy — where some $200 billion is set to be invested by 2025, forecasts Goldman Sachs — the funding landscape in China is surprisingly tepid.
In 2023, China recorded around 232 investments in the AI space, a 38% decline year-over-year, according to research firm CBInsight. The total amount raised by China’s AI firms amounted to roughly $2 billion, 70% less than the year before.
Comment