DeepSeek proved China could lob an AI grenade into global markets.
Now three more startups are lining up behind it, each taking a different swing at the future.
Moonshot AI is challenging the world’s leading language models.
GigaAI and PsiBot are developing “world models” that could help robots, vehicles and other machines understand their physical surroundings.
Two are preparing for potential Hong Kong listings. The third has just joined the unicorn club.
They remain highly speculative, but investors should probably learn the names now rather than after the hype machine hits full throttle.
Moonshot shoots for $50 billion
Moonshot is preparing for another funding round ahead of a possible Hong Kong IPO, with the Chinese AI lab seeking a valuation as high as $50 billion.
The excitement surrounds Kimi K3, its new 2.8 trillion-parameter open-weight model. K3 has reportedly performed close to the latest models from OpenAI and Anthropic, while giving developers greater freedom to access and adapt the technology.
The commercial numbers are also starting to wake up.
Moonshot reached $300 million in annual recurring revenue in June, up from $200 million in April. Daily sales have surged at least sixfold since K3 launched.
Founder Yang Zhilin built the roughly 300-person company alongside researchers from Tsinghua University. Its backers include Tencent, Meituan, China Mobile and several major venture capital firms.
The risk is valuation.
At $50 billion, Moonshot would reach public markets priced more like an established heavyweight than a scrappy challenger.
GigaAI takes AI onto the factory floor
GigaAI, also known as Jijia Vision, is discussing a Hong Kong IPO as soon as this year. It is also closing a funding round that could value it at $3 billion.
Rather than building another chatbot, GigaAI specialises in world models.
These systems help machines understand how people, objects and environments move, making them useful for robots, autonomous vehicles and video games.
Its customers already include automaker China FAW, e-commerce giant JD.com and postal operator EMS.
China’s advantage here may not be smarter researchers or better chips. It may be the country’s enormous network of factories, warehouses and vehicle manufacturers.
The internet supplied the data that trained chatbots. China’s factories could supply the data that trains robots.
PsiBot builds the robot brain
PsiBot, also known as Lingchu Intelligence, is raising almost $100 million at a $1.48 billion valuation.
It has now raised about $300 million since launching in 2024, making it one of China’s few unicorns focused on embodied AI and world models. Unlike Moonshot and GigaAI, it has not disclosed plans for an IPO.
PsiBot collects physical world data using customised gloves and humanoid robots. It wants to gather one million hours of data this year while developing a generalised model capable of telling machines how to act.
It is already testing its technology with a large logistics business and a major fibre-optic cable manufacturer.
This could create a valuable moat.
Text data is everywhere, but high quality information showing robots how to handle physical objects is expensive and difficult to collect.
The catch is that volume alone means little. One million hours of repetitive junk remains repetitive junk.
PsiBot must prove its data can produce robots that reliably perform useful jobs.
China’s next AI shock may have arms and legs
The US still leads in advanced chips and many frontier AI models.
China, however, has a vast industrial base where robots and physical AI can be trained, tested and commercialised quickly.
Moonshot currently has the strongest commercial momentum.
GigaAI has the most advanced industrial customer base. PsiBot is younger and riskier, but its physical world dataset could eventually become the hardest asset to copy.
DeepSeek showed China could compete in AI software.
These three are betting it can do something even bigger: give AI a body and put it to work.
