36Kr Exclusive | Backed by Huawei, GL Ventures and other investors, the physical space intelligent infrastructure company has completed a financing of over 100 million yuan
This article is approximately 4,000 words, with a recommended reading time of 10 minutes
Author | QIAO Yujie
Editor | YUAN Silai
Hard Krypton learned that Ether Tech, a spatial intelligent infrastructure company based on the physical space engine, has recently successively completed its seed round and angel round of financing, with a total financing scale exceeding 100 million yuan. Industrial investors include Huawei Hubble, JLC, Independent Variable, Zhiyuan, CAS Fifth Era, and iSoftStone. Financial investors include GL Ventures, Shenzhen High-tech Investment, Shanghai Semiconductor, Zhongping Capital, Sineng Venture Capital, Jiaoyuan Asset, Neuron VC, Deshi, Xichen and other institutions.
Ether Tech was founded in December 2025, committed to building spatial intelligent infrastructure with the physical space engine as the core. The company's founder and CEO YANG Yeqing once served as Chief of Staff of HTC VIVE, and head of the Metaverse Division, Software and Services Division in China region, leading the development of the first domestic metaverse system platform from 0 to 1. The company's core team comes from well-known enterprises and institutions such as HTC VIVE, Microsoft, JD, Meituan, Peking University, The Hong Kong University of Science and Technology, and Beihang University.
YANG Yeqing introduced to Hard Krypton that the "physical space engine" that Ether Tech focuses on is not a simple superposition of 3D reconstruction, digital twin or spatial data, but aims to solve a more fundamental problem: how space can be continuously perceived, uniformly represented and reliably operated.
Language models enable AI to master the expression ability of the symbolic world, which can write articles, generate codes and conduct reasoning. But if we want AI to truly enter the physical world, we need to solve a more underlying problem: how the real world and the virtual world share a set of computable, operable and translatable spatial languages.
In the past, software mainly processed two-dimensional data, such as text, pictures, tables and web pages, but the real space is not a static map. The production environment of the factory is constantly changing, the warehouse changes with the flow of goods and personnel, and the construction site is also continuously reconstructed during the construction process. Objects in space have position, pose, geometry and material, as well as their own physical properties and motion laws, which are affected by human operations and environmental changes.
Therefore, what the industry needs is not just simple 3D scanning, but an underlying capability that can continuously understand the spatial state and interactive relationship. Traditional information recording focuses on describing "what it is", but it is difficult to fully express "where it is", "how it moves", "how to interact" and how the whole space evolves continuously. "This is exactly the problem that spatial intelligence needs to solve, and it is also the starting point for Ether Tech to build a physical space engine." YANG Yeqing said.
Based on this idea, the physical space engine under construction by Ether Tech is a set of underlying cognitive system that connects the real world and the virtual world relying on spatial cognitive data. Taking spatial data as input and high-precision true value as the benchmark, it continuously compiles the real space into a computable, operable, migratable and deployable spatial representation, and then continuously optimizes the data with the feedback of real scenarios, providing underlying support for embodied intelligence, world models and various industrial applications.
From the perspective of technical architecture, the engine includes five layers from bottom to top: input layer, core capability layer, spatial asset layer, platform layer and application layer. The input layer is responsible for collecting multi-modal data in the real and virtual environments, including sensor information such as vision, spatial positioning, force and tactile, inertia, etc., providing original spatial data for the engine, and solving the problem of "what original information exists in the space". The core capability layer fuses and temporally-spatially aligns these data, identifies spatial structures, object states and interactive relationships, and combines physical constraints to form a representation of spatial states, solving the problem of "what state the space is in and how it operates".
The processed information will be precipitated into loadable, editable and reusable spatial assets at the spatial asset layer. Different from simply recording the geometric appearance of space, these assets also include collision relationships, physical constraints and interactive attributes between objects, providing a basis for the invocation and reuse of different scenarios. At the same time, Ether Tech is also promoting the establishment of a unified spatial asset standard, which is not only used to evaluate the accuracy and consistency of spatial representation, but also supports different industries to directly call spatial capabilities under the same framework, reducing the threshold of cross-scenario reuse.
Going further up, the platform layer encapsulates the capabilities of the spatial engine into services that can be called by different devices and systems through SDK, API, multi-terminal deployment and private deployment; the application layer faces industries such as robotics, industrial manufacturing, automotive, architecture, logistics, gaming, cultural tourism, education, etc., to explore the specific implementation methods of the spatial engine.
In the data input layer, Ether Tech divides spatial data into two dimensions: one is the physical structure of space, to understand the form and layout of space; the other is the interactive data of people in space, to understand what happens in the space, and how people move, operate and give feedback.
At this stage, Ether Tech gives priority to human hands and their interactive data. YANG Yeqing introduced that hands are the most direct interface for humans to act on space. Actions such as grasping, pushing, rotating, inserting, assembling and tool use contain a large amount of spatial relationships, operational experience and physical feedback. Truly valuable spatial data not only records the movement trajectory of the hand, but also needs to record the position where the action occurs, contact relationship, spatial judgment, object state change, and feedback during task execution.
These data can be used not only for embodied intelligence and world model training, but also for fields such as spatial process modeling and physical interaction generation. For example, in the industrial field, by analyzing the operation path of production line workers, the real contact relationship with equipment, tooling and materials, force feedback and state changes, we can extract migratable operation logic and physical constraints, and identify process bottlenecks.
In the gaming field, game developers do not need to build scenarios and actions from scratch, they can call real-anchored spatial assets in real time to generate levels, mechanisms and cross-terminal interactions; objects have real physical properties, players can interact naturally in UGC space, AI NPCs and digital humans can understand object positions, contacts and state changes and give feedback that conforms to physical logic; real operation data is fed back into callable spatial assets, so that the scenario evolves continuously with players' behaviors, forming operable, editable and migratable spatial content.
As one of the earliest input types that the physical space engine pays attention to, human interactive data records not only "where the hand moves", but also the environment people are in, the goal they pursue, the way they act on which object, and what result is finally produced. Ether Tech hopes to build on this basis to serve all industries that need to understand, reconstruct or operate space.
The Ether Tech team has nearly 10 years of XR technology R&D experience, has gone through the evolution from digital twin, metaverse to spatial intelligence, and has accumulated profound technical experience in spatial positioning, 3D reconstruction, human body and object tracking, physical solving, multi-sensor fusion and low-latency interaction.
These technologies are also the basis for the spatial engine to process spatial data. Taking XR spatial computing as an example, the system not only needs to determine the position and pose of the device itself, but also needs to identify the surrounding environment, track the movement of human bodies and objects, and unify the information collected by different sensors into the same spatio-temporal coordinate system. Only after completing these processing can we accurately restore the spatial structure and the interactive relationships in it, rather than just superimposing the real picture on the virtual scene. The team hopes to further abstract the capabilities previously applied to XR spatial computing, use them to process a wider range of spatial data, and transform them into spatial information that can be called by systems such as robots and world models.
According to the introduction, from the perspective of technical implementation, the core of spatial computing is to establish a set of state space synchronized with the real world in real time, continuously perceive and understand device movement, environmental geometric structure, object state changes and the interactive relationship between human and space, so that vision, action, space and physical constraints are consistent under the unified spatio-temporal benchmark, so as to support operable spatial representation and continuously build a state space that can be updated with real changes.
Specifically at the application level of the engine, YANG Yeqing introduced that Ether Tech has carried out applications in the fields of robotics and world models at present, and relevant cooperation is mainly carried out around spatial data. The team has previously served global leading B-end customers and massive C-end developers, covering cultural entertainment, gaming, industry, cultural relics, medical care and other fields, so the company is also exploring the direct application of spatial assets.
For example, explore new content creation methods in the gaming field and reduce production costs; in the field of handicraft and industrial manufacturing, transform the operation process of veteran technicians into analyzable and reusable spatial assets for training or assisting automated repair.
At present, Ether Tech is building more general spatial computing capabilities, forming a complete spatial language system through the physical space engine: connecting the real environment and virtual environment downward, connecting different models, devices and industrial systems upward, so that space can be accurately perceived, structurally represented, continuously edited, physically simulated and deployed across terminals, so that all users can understand, represent and use space.
The following is an excerpt from the founder's Q&A:
Hard Krypton: What is the difference between the physical space engine and 3D reconstruction and digital twin? And how to understand the relationship between the world model of 3D generation route and the physical space engine proposed by Ether Tech.
YANG Yeqing: 3D reconstruction mainly answers what the space looks like at a certain point in time, it solves the problem of geometric restoration, and the output is static and for observation purposes, but it cannot fully describe how the spatial state changes continuously. For a space to generate real value, it is also necessary to understand the constantly occurring changes in it and the interactive relationship between people and space.
Digital twin goes a step further, it maps the state of some objects in the physical world to the virtual environment, describing "what state the real object is in", so that the devices, production lines or buildings in the real world have a corresponding state mirror in the virtual space. Both of them can become the input and component of the physical space engine, but they themselves do not solve the problem of "how space is continuously understood, structurally represented, edited, simulated and deployed".
The physical space engine not only needs to answer what is in the space, where the objects are, and what the relationship between them is, but also needs to answer which objects are operable, what people are doing, how actions change the spatial state, and whether virtual changes can be established in reality. It not only restores space, but also makes space computable, operable and callable.
We understand that the core of the world model is to generate the world, predict or generate the state of the world at the next moment; we do not do generation, but are responsible for representing the state of space. This is the essential difference between the two. Our positioning is the underlying infrastructure of the world model, and the world model can call our capabilities, such as high-quality spatial data, operable spatial assets and physical constraint environments. We transform the real space into computable and editable spatial assets, and the world model can call these assets and related spatial information.
Hard Krypton: What is the difference between the data collection of the physical space engine and the common embodied-oriented data collection now?
YANG Yeqing: The common data collection on the market is only a part of spatial input. The focus of embodied data collection is "recording what actions people have done"; while the focus of data collection of the physical space engine is "recording the relationship between people and space" — not only collecting hand trajectories, but also collecting information such as the spatial position where the action occurs, contact relationship, force, intention, object state change and task feedback. If it only serves for embodied intelligence or world model training, some existing data may be sufficient; but if we want to build a more complete and rich spatial representation, we need to introduce more dimensions of information to answer "what space people are in, for what goal, in what way to act on which object, what result is produced, and how this space can be reused and operated".
Hard Krypton: In terms of human interactive data collection, what advantages does Ether Tech have compared with other data collection companies?
YANG Yeqing: Our advantages are mainly reflected in three aspects: first, high precision, which can reach sub-millimeter level; second, rich data dimensions; third, deep technical accumulation in data processing.
These capabilities mainly come from the team's long-term accumulation in the field of XR spatial computing. XR essentially transforms various information in space into computable information. The technologies we have accumulated in the past, such as spatial positioning, 3D reconstruction and multi-sensor fusion, are very important in human interactive data collection.
At present, the quality of data in the market is uneven. The fundamental reason is that the quality of entry data is restricted by multiple factors such as hardware, software and technical architecture, which cannot be solved by optimizing a single link. Relying on the deep cultivation of spatial computing technology, the team has formed a complete capability closed loop from the source hardware and algorithm, engineering software implementation to the globally leading technical architecture, thus ensuring high-quality input at the data entry side. This advantage does not depend on a single technical path breakthrough, but stems from the systematic design and continuous evolution at the global architecture layer, which is also the core barrier accumulated by the company after more than ten years of deep cultivation in spatial intelligence.