Manycore Tech, a Hangzhou-based design firm, is seeking to become a global leader in spatial intelligence, a burgeoning field that blends artificial intelligence (AI) with real-world spatial data to build what is referred to as a “world model”.
“All users pursue beauty, so we’re here to offer a platform where their ideas can be realistically and accurately depicted digitally and eventually materialise in reality,” Zhu Hao, co-founder and chief technology officer at Manycore, said in a recent interview with the Post. “What you see is what you get.”
The company’s goal is to “realise imagination” through the help of automated systems, Manycore said in its March initial public offering (IPO) prospectus filed with the Hong Kong stock exchange.
As China’s answer to the US firm Autodesk, Manycore initially gained a foothold in the home design sector with its flagship products for designers: Kujiale, KuSpace, and an international version of Coohom. These tools enable designers to quickly create detailed interior layouts by dragging and dropping 3D elements to generate construction blueprints, lighting systems and electrical plans, which are rendered in real time by the company’s clusters of graphics processing units (GPUs).
The company touted itself as the “world’s largest spatial design platform by average monthly active users in 2023, and China’s largest software provider by revenue”, citing research by US consultancy Frost & Sullivan. Last year, it had 2.7 million monthly active users, serving customers in more than 200 markets globally, according to its filing.
Founded in 2011 by three Chinese graduates from the University of Illinois Urbana-Champaign, Manycore gained prominence this year as one of Hangzhou’s “Six Little Dragons”, a cohort of rising start-ups that includes AI developer DeepSeek and humanoid robot maker Unitree Robotics.
During a recent visit to Manycore’s headquarters in the southeastern city, a sign at its entrance read: “There are no paid tours; beware of scams” – a reaction to mounting public interest. A large display in an open workspace showed live metrics, tracking daily rendering tasks and computing power usage.
Zhu stressed that the company’s ultimate goal looks beyond design software, as it was established to advance high-performance computing technology. The company’s name comes from manycore processors, which are designed for parallel processing in high-performance computing – a technology leveraged by Nvidia for its GPUs. Manycore processors are different from the typical multicore chips that deploy fewer, more powerful cores suited for general computing.
Before founding Manycore, CEO Huang Xiaohuang spent two years at Nvidia as a software engineer, working on the company’s proprietary CUDA parallel computing platform.
Long before “spatial intelligence” was widely discussed – popularised by influential AI scholars such as Stanford University’s Li Feifei and accelerated by the robotics boom – Manycore was working towards a “physical world simulator”.
“Intelligent behaviours around language form language intelligence [models],” Zhu explained. “Similarly, intelligent behaviours around space form spatial intelligence. Concepts intuitively grasped by humans, such as identifying walls or grasping objects, are incredibly challenging for machines.”
Zhu expected spatial intelligence to experience a so-called “ChatGPT moment” – a groundbreaking product release that popularises spatial models the way OpenAI’s popular chatbot did for large language models (LLMs). However, Zhu said this moment is still five to 10 years away.
Manycore has been releasing research since 2018, when it introduced InteriorNet, a large-scale, multi-sensor data set for indoor scenes. Last year, it launched SpatialVerse, a virtual training platform for AI-generated content, which has been utilised by Chinese robotics start-up AgiBot for robot training. In March, Manycore open-sourced SpatialLM, a 3D AI model capable of turning point cloud data into structured spatial understanding – something that is useful in robotics and navigation. More open-source projects are planned for the second half of the year, according to Zhu.
Zhu drew parallels with the rise of LLMs, pointing to three essential factors driving spatial intelligence: data, algorithms and computing power.
“Spatial intelligence is still at an early stage,” he said. “There’s no industry consensus on algorithms yet, computing power demands aren’t very high, and data remains scarce.” Manycore’s extensive experience in interior decoration and engineering gives it a substantial data advantage, he added.
The company is preparing for a Hong Kong IPO after scrapping a planned US listing in 2021. The decision followed the troubled New York debut of Chinese ride-hailing giant Didi Global, which was immediately hit with a Chinese cybersecurity investigation that eventually led to a delisting.
Manycore has yet to turn a profit, reporting revenue of 552.9 million yuan (US$76 million) for the nine months ending September 30, up from 486 million yuan for the same period in 2023. Losses in the same period narrowed to 422.1 million yuan from 489.5 million yuan year on year.