Zhejiang University 00s founder's AI startup valued near 10 billion CNY

Investors have learned from insiders that Moxin Technology, led by Chen Tianrun, is about to complete a new funding round, with the amount potentially reaching the billion-yuan level, pushing the company’s valuation to nearly 10 billion yuan.

This comes less than a month after the company’s previous funding round, when its valuation was RMB 4 billion. Such a rapid pace of fundraising and valuation growth is astonishing.

Chen Tianrun, born in the 2000s, graduated from Zhejiang University under the supervision of Academician Pan Yunhe. Even during university, Chen embarked on his entrepreneurial journey, transitioning over five years from 3D printing and 4D world models to consumer-grade hardware and physical AI infrastructure.

The significance of this leap may extend beyond a single company’s valuation—it’s more like a signal that China’s younger generation of entrepreneurs is stepping onto the starting line to redefine AI.

It's uncommon to see valuation more than double in just one month.

According to insiders, Moxin Technology’s new round of funding continues the strong involvement of industrial capital, with increased investment from state-owned and market-based capital, covering multiple sectors including chips, computing power, smart terminals, digital content, and advanced manufacturing.

Unlike typical financial investments in the market, this comprehensive capital matrix represents a collective bet by core players across the industry’s upstream and downstream chains on the industrialization potential of the World Model, signaling that multi-party collaborative deployment in the World Model industrialization has entered a critical stage, while also affirming MagicChip Technology’s scarce value within this sector.

Not long ago, Moxin Technology completed a RMB 500 million funding round, reaching a valuation of RMB 4 billion, and has assembled an impressive lineup of investors.

First, Shenzhen Capital Group and Zhejiang Capital Group joined forces. As typical patient capital, state-owned institutions making significant investments in cutting-edge technology sectors at an early stage reflect, to some extent, policy recognition of the strategic value of world models.

Secondly, there has been a surge in industrial capital involvement. Among them, Huawei HuaBo is an early investor in MoXin Technology, and in recent years, both parties have collaborated on model training and inference optimization for the Ascend NPU. Notably, Huawei has ranked the world model as the number one trend among the “Top Ten Technological Trends for the Intelligent World of 2035,” which forms the fundamental rationale behind its investment in MoXin Technology.

It is well known that world models require processing long-duration, high-resolution visual data, placing higher demands on chip computing power, memory bandwidth, device communication, and energy consumption; chip companies also need to validate their product capabilities through new models and scenarios.

Therefore, the collective involvement of semiconductor industry investors—such as Zhongxin Juyuan, Xi Xin Investment, Fuhan Micro, and Jiehuajue—signifies that chip companies’ participation goes beyond mere capital investment; it breaks down the synergistic barriers between underlying chip hardware and upper-layer model algorithms, driving a positive feedback loop where advancements in chip computing power empower model upgrades, and model innovations uncover new hardware requirements.

Of course, it doesn’t stop there—Lenovo Holdings, which spans PC and smart device terminals; Ganbei Fund, backed by Bilibili and possessing vast content resources; and StarRui Capital, with its roots in Transsion Industries. These powerful investors connect smart devices with digital content ecosystems, establishing a complete application chain for world models, from “entry points” to “use cases.”

Meanwhile, advanced manufacturing and energy capital represented by JinkoSolar Holding are poised to fill the gap in real-world industrial applications for world models. This will drive the deep integration of world models from consumer entertainment scenarios into physical industries such as smart manufacturing, significantly expanding the commercial boundaries of MagicChip Technology.

Ultimately, unlike traditional financial investors who seek short-term returns, this round of industrial capital investment follows a “strategic positioning” logic—industrial players invest in MoXin Technology to secure their place in the next-generation spatial intelligence infrastructure and ensure their business scenarios can be among the first to access world model capabilities.

The significant investments by leading institutions such as P7 USD Fund,鼎晖百孚 & Longxin Venture Capital, Tunlan Investment, Shuyun Capital, and Junlang Venture Capital validate, from a market perspective, Moxin Technology’s technical capabilities and potential for global competition.

Upon careful review, from the end of 2025 through this past August, Moxin Technology successfully completed four rounds of financing, each bringing in new industrial investors while existing shareholders continued to increase their stakes. Combined with the upcoming new round of financing amounting to billions of yuan, this level of funding density is exceptionally rare among global model startup companies.

Behind this financing pace lies a clear signal: the Matthew effect in the sector is intensifying, and players with a first-mover advantage are more worthy of heavy investment.

The investment logic for AI projects in the primary market has been completely restructured, moving away from the previous era of rough, concept-driven investments toward a systematic approach that emphasizes technology, industry application, and closed-loop execution. Investors no longer blindly chase general-purpose large models; only those projects with vertical technological barriers and clear use cases stand out. The overwhelming success of MoXin Technology’s fundraising is a telling example.

It is reported that Moxin Technology has recently completed its transformation into a joint-stock company, undoubtedly laying the institutional groundwork for its subsequent capital market journey. Perhaps soon, we will witness the birth of a super IPO.

What supports a valuation of nearly ten billion?

This story begins with founder Chen Tianrun. Born in 2000, he started learning programming in middle school and developed a strong interest in 3D modeling technology. Later, when he enrolled at Zhejiang University, he participated in extensive research on 3D modeling and 3D reconstruction.

After completing his undergraduate studies, Chen Tianrun was recommended for direct admission to the Ph.D. program in the School of Computer Science at Zhejiang University, where he studied under Academician Pan Yunhe, a member of the Chinese Academy of Engineering and former president of Zhejiang University. During this time, he published over 50 high-level papers in journals such as Science, Nature Photonics, and IEEE TPAMI, and at conferences including ICCV, CVPR, ICML, and SIGGRAPH.

This is an impressive resume, but Chen Tianrun’s choice holds far greater significance than the resume itself.

In 2021, while still a junior in college, Chen Tianrun conceived the idea of starting a business. He believed that introducing emerging technologies to a broader audience through commercialization was worth pursuing. Thus, Moxin Technology was founded, initially focusing on consumer-grade 3D printing hardware.

Years of experience in 3D technologies enabled Chen Tianrun to see greater opportunities. He believes that 3D printing is merely the starting point for "enabling AI to understand the three-dimensional world," and is determined to enter the higher-dimensional field of physical AI. To this end, Chen Tianrun has led his team to fully shift toward 4D world models and is deeply investing in the field of spatial intelligence.

MoXin Technology has taken a unique approach by adopting an implicit-only architecture, centered on predicting future states within a high-dimensional latent space, rather than reconstructing the world at the pixel or explicit geometry level. This technical choice has given it a competitive advantage—enabling the AI to retain long-term scene states instead of “forgetting” them immediately after viewing.

This advantage stems from MagicChip Technology’s prior accumulation. Since its founding, the team has employed designers and digital artists to create 3D content, gradually building up petabytes of 3D models and scene assets covering complex physical environments and dynamic natural settings—a capability that other players cannot replicate or catch up to in the short term.

Building on this, Moxin Technology released the interactive world model prototype KOKONI-World at the end of last year. In July this year, Moxin Technology, in collaboration with the National AI Application Pilot Base and Huawei Cloud, launched the general-purpose world model MoWorld-4D, which was trained, compressed, and deployed using domestically produced NPUs, achieving up to approximately 50 frames per second in real-time inference.

Specifically, users can generate immersive, high-definition 3D scenes with spatial structure through inputs such as images and text, creating a fully coherent environment that adheres to real-world physical laws. MoWorld-4D uses the six-degree-of-freedom motion of a camera as a physical signal, allowing users to walk, turn, and circle around within the generated world to observe the full scene. The architectural structures, road layouts, and object positions in the generated environment all accurately reflect real-world spatial relationships.

Meanwhile, MoXin Technology has also introduced the concept of the Flash World Model—high frame rates are essential to keep world models usable in real-world scenarios, marking AI’s transition from “content generation” to “interactive world simulation.”

Why is "high frame rate" so important? Previously, the mainstream frame rate for world models was below 30 FPS, which was widely regarded as the minimum threshold for real-time rendering and delivering a smooth interactive experience. A frame rate of 50 FPS means each frame must be generated in just 20 milliseconds, enabling rapid response to user commands. In scenarios where latency is critically sensitive—such as humanoid robot control and autonomous driving—this difference is decisive.

More importantly, MoWorld-4D achieves a breakthrough by unifying high frame rates, high visual quality, and low cost. It is the world’s first 4D world model trained, compressed, distilled, and inferenced in real time on Huawei Ascend NPU super nodes, fully eliminating dependence on overseas high-end GPUs.

Moreover, its total inference cost is only 30% of that of an equivalent imported GPU solution, breaking the industry's cost barrier and demonstrating the feasibility of using domestic computing power to support the large-scale deployment of world models, providing domestic enterprises with a viable, replicable pathway.

Across the entire field, most current world models still focus on static image generation and video rendering optimization for consumer entertainment scenarios, emphasizing presentation over practical implementation, and lacking industrial-grade real-time interactivity and cost advantages. In contrast, MoXin Technology has, from the outset, targeted the core needs of industrial application, building a general-purpose spatial intelligence infrastructure.

Therefore, the core value of MoWorld-4D extends beyond generating realistic scenes—it lies in its ability to precisely simulate dynamic causality, spatial interactions, and environmental evolution in the physical world, meeting the real-time decision-making needs of various physical AI endpoints. This is a robust, industry-focused technology and undoubtedly a rare asset in the primary market.

Currently, a global storm of world models is sweeping the world.

At the beginning of this year, World Labs, founded by Fei-Fei Li, announced the completion of a new $1 billion funding round, reaching an astonishing valuation of $5 billion in just one year; shortly after, AMI, led by Yann LeCun, closed a seed round of approximately $1.03 billion, setting a new record for the largest seed round in European history.

Turning our focus back to China, on one side, world model startups such as Jiyi Vision, LiberAI, Wujie Power, and VAST are rapidly securing funding; on the other, internet giants are all entering the space—for instance, Tencent and Alibaba simultaneously released their respective world model products, Hunyuan 3D World Model 2.0 and Happy Oyster, while ByteDance’s Seed team unveiled its next-generation 3D generation large model, Seed3D2.0.

The reason world models have become so popular is that they are seen as the core engine of physical AGI, enabling AI to fill the critical gap of "understanding and interacting with reality," thereby transitioning AI from merely "generating" to "understanding, reasoning, and acting."

However, the competitive landscape for world models is undergoing a fundamental shift—from a laboratory-based technological race to a systemic competition involving computing power, models, data, and industrial applications. The market’s focus is no longer on “whether scenarios can be generated,” but rather on “whether real-time interaction, low-cost scalability, and alignment with real industrial needs can be achieved.”

In other words, technology must ultimately deliver real industrial value.

This is also a key focus area for the MoXin Tech team. On one hand, the company meets the stringent real-time requirements of industries such as industrial automation, transportation, and robotics through a high frame rate of 50 FPS, transforming world models from “demonstration technologies” into “production-grade tools.” On the other hand, the team’s exploration of a low-cost, domestically developed pathway has broken down barriers to deployment.

Following the release of the MoWorld-4D model, MoCore Technology has accelerated exploration and practical applications in multiple scenarios, including industrial digital twins, embodied intelligence training, urban and rural planning, and cultural tourism consumption.

In advanced manufacturing, the model can convert images captured by smartphones or ordinary cameras into 3D space, enabling enterprises to continuously update the digital status of their factories and production lines, and perform scenario simulations and preliminary risk assessments before equipment installation or production line modifications. In the field of embodied intelligence, MoWorld-4D provides robots with a continuous, controllable virtual training environment, reducing the cost of data collection and trial-and-error in real-world scenarios. For urban and rural planning and cultural tourism, the model can be used for spatial reconstruction, scenario simulation, digital attraction creation, and immersive content production.

Any technology that survives the cycle must ultimately return to the real world, to industry, and to people's genuine needs.

AI has already demonstrated tremendous potential in information processing and knowledge work, but the greater opportunities lie ahead in how it integrates into more complex, real-world systems. Just as MoXin Technology has grown from a valuation of 4 billion to nearly 10 billion yuan, capital is revaluing the company’s potential to transition from model breakthroughs to becoming industrial infrastructure.

The primary market brings together the smartest capital in the market. Their sustained heavy positioning is not just a vote of confidence, but also reflects a firm belief that Chinese strength will inevitably hold a prominent place on the global spatial intelligence map.

This article is from the WeChat public account "Investment Daily" (ID: pedaily2012), authored by Liu Bo.