Meituan Long-Z Investments believes the company can outmuscle domestic competition by delivering generative AI infrastructure and apps that combine technology and user awareness

China’s technology giants were seen as likely prime movers in the development of large language models (LLMs) that underpin generative artificial intelligence (AI). Their resources – across capital, talent, and computing power – represented a potentially decisive competitive advantage in training systems to understand and generate natural language from vast data.

But start-ups are still in the fight, with the likes of Moonshot AI, MiniMax, Baichuan AI and Zhipu AI each closing rounds of approximately USD 300m in the past year. Moonshot’s latest effort – a commitment of more than USD 1bn, led by Alibaba Group – values the year-old company at more than USD 2.5bn, according to sources familiar with the situation.

It is now worth the same as MiniMax but got there in half the time. Moonshot’s early traction is linked to the bona fides of its founder. Kimi Yang previously worked at Meta and Google and co-founded the domestic AI-driven corporate services provider Recurrent AI.

“Only through a sustained grasp of technological context and its value, can [investors] identify the talent relevant to this technology,” said Xinyu Wang, partner at Meituan Long-Z Investments. “Technological advancement isn’t a leap, while our attention to AI is a continuum.”

Long-Z – also known as Longzhu Capital and supported by Meituan – led the first of two Series A tranches raised last October. This came four months after an angel round featuring HongShan, Monolith Management, and ZhenFund, among others. Lanchi Ventures joined Long-Z as a new investor in the Series A. The most recent round, which includes new and existing investors, is ongoing, the sources added.

Wang first got exposure to AI through investments in autonomous driving specialists and chip designers made while working at GGV Capital and Kunlun Capital. Joining Long-Z in 2021, he established a team drawn from local AI unicorns and moved on to Moonshot as soon as Yang opened to external investors last May.

Yang noted Long-Z’s “clear judgement on the outlook of AGI [artificial general intelligence]” and its assessments of the strengths and weaknesses of his team. The firm put forward candidates for recruitment before investing, cementing its status as “a long-term partner worth working with.”

At the time, Moonshot had only 20 employees – headcount has since risen to 80 – and no LLMs. But Wang was impressed by Yang’s “exceptional grasp of AI, robust academic foundation, and genuine personality,” which suggested that progress could be rapid.

The company’s first LLM, Moonshot-v1, launched upon the completion of the Series A. It has trained 100bn parameters – comparable to offerings from ByteDance, Baichuan AI, SenseTime, MiniMax, and Zhipu AI. OpenAI’s GPT-4 reportedly has 1.76trn parameters, but Wang believes Moonshot could narrow the technology gap in different ways.

“While OpenAI may excel overall, Moonshot could shine in specific areas,” he said. “The focus is more on leveraging these advantages effectively.”

Over the past five months, Moonshot-v1’s text processing capability has risen from 200,000 to 2m Chinese characters. This means it can handle longer texts than rival local LLMs. Moonshot also claims to distinguish itself through “lossless long context” – maintaining a deep understanding of context without loss of information – and user experience.

The company has built Kimi Chat, a chatbot based on Moonshot-v1 that operates in English and Chinese. Available via web, mobile app, and WeChat mini-program, it recently crashed following a traffic surge. The start-up has also flagged growing interest in what it describes as Kimi concept stocks, or A-share listed companies that have ties to the AI value chain or Moonshot specifically.

Yang aspires to create a super app that can match Douyin or WeChat for scale. The functionality has yet to be defined, but it would be AI-driven and aimed at mainstream users. Talking to local media, he emphasised the importance of user trust and interaction, data engagement, and understanding customer demand, adding that companies that fall short “may ultimately fail to achieve AGI.”

According to Wang, this clarity of vision should silence concerns about Yang’s youth – he is 32 – and commercial rawness. Moreover, he believes a dual focus on technology and product can help LLM start-ups gain the edge over China’s incumbent tech giants.

“If you solely prioritise product development, applications built on weak technological foundations will be drowned out amidst the iterations of models,” Wang said. “On the other hand, if product development lags fundamental technology in the long term, potential support from both customers and investors will inevitably diminish.”