A post-2000s PhD candidate studying at Zhejiang University has reached a valuation of 10 billion yuan.

According to ChinaVenture, a new round of financing for Moxin Technology, a Hangzhou-based 4D world model enterprise, is about to be completed, with a scale likely reaching the order of 10 billion yuan, and its overall valuation has jumped to nearly 100 billion yuan. Only one month has passed since the previous round of 4 billion yuan valuation settlement, and the valuation has more than doubled.

Chen Tianrun, the founder and CEO leading the team, is a post-2000s doctoral candidate studying at Zhejiang University. He started his entrepreneurial path as early as his undergraduate stage, and his supervisor is Pan Yunhe, an academician of the Chinese Academy of Engineering. To put it in a more traffic-grabbing way, he and Liang Wenfeng of DeepSeek are both alumni of Zhejiang University, and he is one of the representative figures of the new-generation entrepreneurs from the Zhejiang University system in the public eye.

There are two noteworthy details in this financing deal of Moxin Technology.

The first is the speed. Moxin has quickly completed 5 rounds of financing within 9 months, with its valuation rising all the way from 840 million yuan in the Pre-A+ round in March this year to nearly 100 billion yuan. A more direct comparison is that the financing scale of this new single round equals the total financing amount of the previous 4 rounds in the past 8 months. Such a fast financing pace and sharp valuation increase are relatively rare among domestic AI companies.

The second is the composition of the capital entering the market. The investors of this round further cover multiple fields including chips, computing power, smart terminals, digital content and advanced manufacturing, with state-owned capital and market-oriented capital both increasing their investment intensity. In other words, chip companies, terminal companies, manufacturing enterprises and state-owned capital have all joined in.

Of course, apart from the financing news, the question that the market cares more about is: what makes Moxin stand out?

What is the capital paying for?

If you only interpret Moxin's new round of financing as a doubling of a company's valuation, you will underestimate the industrial significance of this sum of money.

First look at the investment lineup. After sorting out the information, the list of Moxin's shareholders across all rounds can be divided into four categories according to the industrial coordinates:

Domestic computing power and semiconductor capital: Including Huawei Hubble, SMIC PE, JHIC, Silicore Investment;

Smart terminal and content capital: Including Lenovo, Transsion, Cheers Fund;

Advanced manufacturing and energy capital: Represented by relevant platforms of JinkoSolar;

State-owned capital and market-oriented capital: Including Shenzhen Venture Capital, Zhejiang Venture Capital, Tunlan, Prosperity7.

In terms of equity structure, founder Chen Tianrun holds about 36.6% of the shares personally. Among them, Hubble, under Huawei, has continuously increased its holdings since the Pre-A+ round, and its latest shareholding is about 7.98%, making it the largest external shareholder. From the composition of this list, it is mostly industrial capital from the upstream and downstream of the industrial chain that has entered the market collectively.

It is easy to understand why the four types of capital enter the same game: domestic computing power and semiconductor capital join in, most likely because the world model is the next key starting point to absorb the computing power inventory; smart terminal and content capital join in, because the competition for the entrance of the next-generation operating system needs the support of the world model; advanced manufacturing and energy capital join in, because digital twin is a new tool to reduce the digitalization cost of factories and power stations; state-owned capital and market-oriented capital, which belong to long-term patient capital, are more inclined to bet on the future industry of physical AI infrastructure at the national level.

This kind of market entry happens once before every generation of infrastructure-based technologies matures.

It will be clearer if we extend the time scale. From 2009 to 2012, during the formation period of mobile Internet infrastructure, BAT respectively bound a group of upstream and downstream organizations in the industrial chain, with terminals, chips and content all in their proper positions; from 2016 to 2018, during the formation period of cloud infrastructure, AWS, Azure and Alibaba Cloud respectively attracted a number of enterprise customers bound by the ecosystem.

In every round of infrastructure formation period, capital pays collectively for the position occupation in the upstream and downstream of the industrial chain. The current world model track is repeating the same process. Moxin's speed of completing 5 rounds of financing in 8 months and its valuation jumping from nearly 40 billion yuan to nearly 100 billion yuan is essentially the product of this pricing mechanism.

At present, these capitals are paying for three types of capabilities at the same time: the capability of the model itself, the scarcity of the domestic computing power path, and the entrance occupation of industrial scenarios. The first two are technical narratives, and the third is a commercial narrative. Looking at the whole industry, there are only a handful of targets in China that pass the three lines at the same time, and Moxin is one of them with a relatively complete position.

A complete position means that Moxin is stuck in a middle-layer position: the upstream is connected to chips and computing power, the downstream is connected to terminals and industries, and the world model is used in the middle to turn real data into interactive 4D scenes. The four variables of chips, terminals, data and industrial applications are all taken over by one company at the same time.

Looking out from this position, the pricing power of physical AI infrastructure is shifting from technical competition in the laboratory to systematic competition involving computing power, models, data and industrial applications. This middle layer that Moxin occupies is exactly one of the most valuable entrances in this shift.

As for why this shift is happening now, not earlier or later, the answer is that the global competition of world models has entered the actual combat stage simultaneously in 2026.

In the past three years, the main battlefield of large models has been language. Since this year, the battlefield of the next stage has shifted to the physical world.

The global bets are clearly visible.

On the North American side, World Labs founded by Li Feifei advocates explicit 3D structure, JEPA promoted by Yann LeCun follows the latent space prediction path, and NVIDIA uses the Cosmos world foundation model platform to bind the computing power ecosystem. On the domestic side, ByteDance Seed has set up a world model team, while Alibaba, Tencent and Huawei have respectively increased their efforts in the direction of digital twin and spatial intelligence. According to the research report released by IT Juzi database on June 18, 2026, the cumulative financing scale of 33 domestic startups with the world model as the core concept has exceeded 260 billion yuan, and 7 of them have joined the unicorn ranks.

That is to say, the world model track is replicating the pattern of the 2023 large language model boom: the technical routes are still diverging, and the three paths of explicit, latent space and implicit representation are advancing side by side, but the three forces of capital, industry and policy are already competing for the entrance.

What makes Moxin stand out?

To break down why Moxin can get ahead, the founding team is undoubtedly the most important hidden card of the company.

Founder Chen Tianrun is the core source of the academic background of this company. He has published more than 50 papers in top journals such as Science, Nature Photonics and IEEE TPAMI, as well as top conferences such as ICCV, CVPR, ICML and SIGGRAPH. Many of his research results have been selected as Conference Highlight, Oral and Spotlight, and he also won the Chinagraph Best Paper Award in 2024.

This academic accumulation is the basis for Moxin to push the world model to the engineering boundary of real-time interaction. Public information shows that the core R&D team of the company inherits the research context of Zhejiang University in the direction of spatial intelligence and 3D vision. Most of the members have continuous publication experience in international top conferences and journals, and their long-term accumulation is focused on the basic directions of 3D vision, neural rendering and geometric reconstruction.

The combination of Chen Tianrun and his team enables this company to have two capabilities at the same time: it can distinguish the real problems from the packaged gimmicks in the world model track; it can turn judgments into deployable engineering products. This combination of "academic judgment plus engineering capability" is a hidden card that other new players are difficult to make up for in the short term in a track where basic research and engineering implementation are highly coupled, such as the world model.

Looking back at the entrepreneurial history:

In 2021, Chen Tianrun developed the first consumer-grade 3D printer when he was a junior college student, founded the KOKONI brand, with the cumulative product sales exceeding 200,000 units, turning a niche category that originally belonged to the geek circle into a consumer product. This execution capability of breaking through from hardware consumer products to the mass market is regarded by the outside world as the background that allows Moxin to quickly spread world model products in industrial scenarios later.

At the end of 2024, the company announced a full shift to 4D world models. Chen Tianrun described this transformation as the most painful cognitive refresh, jumping out of a profitable and still growing business to embrace a track with a long cycle, high computing power consumption and a commercialization path that has not yet fully converged.

But the result is excellent. In terms of model capability, what Moxin provides is not just another Demo, but pushes real-time interaction, the most difficult engineering indicator of the world model, from the demonstration stage to the deployable range.

In July 2026, Moxin jointly released MoWorld 4D with the National Artificial Intelligence Application Pilot Test Base and Huawei Cloud, put forward the concept of Flash World Model, and took high frame rate which ensures the availability of world models in real scenarios as the core indicator of productization. Technically, the model adopts a 28B parameter MoE mixed expert architecture, relies on 6-degree-of-freedom camera movement as the physical signal, and can achieve a maximum real-time inference of about 50 FPS.

In horizontal comparison, traditional video generation models generally run below 30 FPS, and most mainstream world model Demos stay at single-digit interactive frame rates. Moxin's 50 FPS pushes the boundary from "viewing the world" to "moving around in the world" in engineering practice.

In terms of the domestic computing power path, it is the most critical watershed that distinguishes Moxin from other world model companies. The full stack of MoWorld completes training, compression and deployment based on Ascend NPU, and the comprehensive reasoning cost is only 30% of that of the import GPU solution of the same scale.

Behind this number are two industry-level significances: first, it proves that domestic computing power can fully support the large-scale deployment of world models with high frame rate, low cost and real-time interaction; second, the 30% cost means that the deployment threshold of the world model has for the first time fallen to a level acceptable to industrial scenarios.

A number of industry observers compare it to the DeepSeek moment of the world model. The core is not that its performance surpasses others, but that it turns a capability that previously only belonged to scientific research and Demo into a component of industrial infrastructure.

In terms of industrial applications, Moxin has applied the model to four tracks:

In the direction of industrial digital twin, images captured by ordinary mobile phones or cameras can be converted into 3D space, and enterprises can use it to continuously update the digital status of factories and production lines, and conduct pre-demonstration of plans before equipment installation and production line transformation; in the direction of embodied intelligence, the model provides a continuous and controllable virtual training environment for robots, where collisions and mistakes only cause computing power loss without real cost; in the direction of urban and rural planning, the model supports spatial reconstruction, scheme simulation and digital scenic spots; in the direction of cultural tourism, the model is connected to immersive content production.

The four puzzles of geometric model, interaction model, domestic computing power and industrial scenarios are verified at the same time, which brings Moxin a new product portfolio narrative. Therefore, the key for Moxin to obtain intensive financing and achieve valuation leap is not one or two technical indicators, but the simultaneous verification of the three lines of model, computing power and industry.

Of course, this story is far from the time when the answer is revealed. The technical route dispute of world models has not yet converged, the engineering challenge of domestic computing power has just unfolded, and the real willingness to pay in industrial scenarios still requires longer market verification. Whether Moxin's nearly 100 billion yuan valuation in this round can be supported by performance depends on whether it can generate real scale revenue in the two main lines of industrial digital twin and embodied intelligence in the next 12 months.

This article is from WeChat official account "ChinaVenture", author: Zan Zhu, published with authorization from 36Kr.