
Deal focus: Qiming backs Chinese GPT pioneer

Frontis.ai is a Chinese start-up founded by local entrepreneur who helped lay the groundwork for ChatGPT in the US. Qiming Venture Partners was won over by its product conception capabilities
When Bowen Zhou, head of artificial intelligence (AI) platform and research at Chinese e-commerce giant JD.com, decided to go solo in late 2021, Qiming Venture Partners didn’t hesitate in supporting him. Many weekends were spent with Alex Zhou, a partner at the VC firm, discussing direction and strategy. Corporate structure and the composition of the founding team were also covered.
At the time, AI-generated content (AIGC) had yet to emerge as a mainstream concept, but Bowen Zhou is a heavyweight in the field. Formerly a director at IBM’s US-based Institute of Fundamental Research in Artificial Intelligence, he authored a paper in 2016 that captured key aspects of transformer architecture – the magic behind ChatGPT. Generative pre-trained transformer (GPT) originated from transformer.
Bowen Zhou named his start-up Frontis.ai and Qiming led an initial funding round with support from Matrix Partners China. It amounted to several hundreds of millions of renminbi. The round closed last year, but it was announced only recently, amid the AI frenzy created by ChatGPT.
"We try to get ahead of the curve in our technology investments by identifying trends and patterns before the market gets hot,” Alex Zhou said. “We’ve taken this approach in AI for the past decade. We were in discussions with experts at OpenAI and Google Brain in 2018, and we were very optimistic about the prospects for this technology. However, we couldn’t have anticipated the public enthusiasm.”
Frontis aims to leverage AI’s content generation capabilities to deliver customer-centric products. Its AI platform can directly generate market, consumer, and trend insights, as well as product ideas based on these inputs. The goal is to accelerate product-to-market timelines and increase the probability of creating “explosive” product models.
“A clear trend in the past few years has been the shift from ‘Made in China’ to ‘Created in China’ or ‘Designed in China.’ Local players are building brands by leveraging the supply chain ecosystem we’ve gathered over the years as OEMs [original equipment manufacturers] for global brands,” said Alex Zhou.
“These local brands aim to make top-notch products that can match the likes of Apple or Samsung. Design remains the most challenging part and Frontis is addressing this pain point.”
Recommending content to customers is at present the most widely used AI function, which is not dissimilar to the Frontis offering. The key difference is that Frontis can match demand with products that don’t yet exist. Individual products comprise an array of subtle elements – materials, processes, usage scenarios – that the company layers on top of dynamic consumer data to conceive solutions.
Frontis has already worked with the likes of JD.com and dairy brand Mengniu. The company plans to generate revenue through a subscription-based software-as-a-service (SaaS) model.
“What we like about generative AI is that it has a very powerful ability to generalise. Frontis can not only help design cosmetics products but also refrigerators or even cars. This is the biggest breakthrough. Generative AI is not just about expertise in a specific domain, it is a more general intelligence,” said Alex Zhou.
Asked to assess the ChatGPT opportunity, he divides model-relevant startups into three groups. First, closed source models such as ChatGPT – and rival offerings being developed by the likes of Baidu, Facebook, and Google – that deliver an application programming interface (API) clients plug into their platforms. Qiming portfolio companies in this category include Zhipu.ai, a Tsinghua University spinout.
Second, open-source models (referred to as Model Hub) such as US-based Hugging Face, Stability, and Alibaba Group’s ModelScope. Users can not only modify the model but also host it. Third, integrated players – like Frontis – that combine the infrastructure model and end-consumer applications.
Most AI start-ups operate in the application layer that sits on top of the models. In this area, Qiming has invested in Gemsouls, an AI platform that powers virtual characters.
Alex Zhou believes that Frontis is one of few start-ups that can build from model to application. He estimates 90% of players are app developers. Moreover, China might be a generation behind the US on large language model technology, but it could enjoy late-mover advantage. Google demonstrated this effect in March on releasing an updated offering that closed the gap with ChatGPT.
“There is no secret sauce in terms of technology. The gap is in engineering which doesn’t make for a real moat,” Alex Zhou added. “Latecomers can have advantages. In the 1990s, US internet technology was a generation or two ahead of China, but within a few years, China was eating into this lead.”
Latest News
Asian GPs slow implementation of ESG policies - survey
Asia-based private equity firms are assigning more dedicated resources to environment, social, and governance (ESG) programmes, but policy changes have slowed in the past 12 months, in part due to concerns raised internally and by LPs, according to a...
Singapore fintech start-up LXA gets $10m seed round
New Enterprise Associates (NEA) has led a USD 10m seed round for Singapore’s LXA, a financial technology start-up launched by a former Asia senior executive at The Blackstone Group.
India's InCred announces $60m round, claims unicorn status
Indian non-bank lender InCred Financial Services said it has received INR 5bn (USD 60m) at a valuation of at least USD 1bn from unnamed investors including “a global private equity fund.”
Insight leads $50m round for Australia's Roller
Insight Partners has led a USD 50m round for Australia’s Roller, a venue management software provider specializing in family fun parks.