Recently, Light Wheel Intelligence (Beijing) Technology (光轮智能) completed a combined Series A++ and A+++ financing totaling RMB 1 billion. The round introduced multiple industry partners and financial institutions. Strategic investors include New Hope Group (新希望集团), Dingbang Investment (鼎邦投资) — the family office of the chairman of San’an Optoelectronics (三安光电) — as well as AUX Group (奥克斯) and Dingshi Asset Management (鼎石资管). Financial investors participating in the round include CCB Sci-Tech (建投华科), Guofang Innovation (国方创新), Daohe Long-term Investment (道禾长期投资), and Qingxin Capital (清新资本).
Following this financing, Light Wheel Intelligence has become the world’s first unicorn company in the embodied data sector. The newly raised capital will be primarily used to further develop its physical simulation engine, upgrade its large-scale model evaluation system, and expand global delivery and localized deployment capabilities, further strengthening the company’s leadership in physical AI data and simulation infrastructure.
The Physical AI Era: New Infrastructure Is Emerging
Over the past decade, computing power has defined the infrastructure of AI. GPU clusters powered the rise of deep learning, enabling large models to rapidly emerge within the digital world.
Today, however, AI is undergoing a deeper transformation—moving from the digital realm into the physical world.
As robots begin to operate in real environments, the true bottleneck is no longer hardware alone, but the ability to scale data. The era of embodied intelligence data has arrived, and the demand for such data is growing exponentially. At this stage, a new layer of infrastructure is taking shape: not just computing power, but data and simulation infrastructure. Light Wheel Intelligence (Beijing) Technology (光轮智能) is one of the builders of this new foundation.
A Three-Layer Architecture: A Scalable Engine for Embodied Data and Simulation
At the product level, Light Wheel Intelligence has established a scalable engine for embodied data and simulation. Built around a three-layer architecture—World, Behavior, and Eval—the system covers the entire pipeline from physically realistic simulation and large-scale data generation to model capability evaluation.
World Layer.
At the foundation, Light Wheel Intelligence has developed a fully self-built technology stack centered on the integration of solving, measurement, and generation. Its proprietary physics solver supports high-precision, real-time simulation across multiple physical domains, including rigid bodies, soft bodies, and fluid dynamics. Combined with the company’s original physical measurement factory, its sim-to-real benchmarking methodology, and its capability for large-scale non-rigid asset generation, the World layer forms a closed loop that connects physical realism with scalable data production.
Behavior Layer.
On top of this foundation, Light Wheel Intelligence has built the world’s largest non-embodied data engine, enabling large-scale production of embodied intelligence data through two primary pathways: simulation-generated synthetic data and human video data (EgoSuite). The company’s data solutions have already been deployed at scale globally, with production capacity continuing to lead the industry and providing a high-frequency, scalable data supply for embodied foundation models.
Eval Layer.
To complete the stack, Light Wheel Intelligence launched RoboFinals, the industry’s first high-difficulty, industrial-grade simulation evaluation platform designed to benchmark cutting-edge large models. The platform establishes a standardized evaluation framework for embodied intelligence and helps accelerate the development of a scalable industry-wide benchmarking ecosystem.
The three layers together form a self-reinforcing data flywheel. Physical measurements from the real world, together with EgoSuite human video data, continuously feed into the simulation environment to ensure physical fidelity. The physics-accurate simulation then drives RoboFinals to conduct large-scale evaluation, while insights from these evaluations inform the next round of data collection and generation.
With every cycle of the flywheel, simulations become more precise, datasets become more effective, and evaluation becomes more comprehensive. Through this process, Light Wheel Intelligence (Beijing) Technology gains control over the core supply of embodied data and establishes a complete data and simulation infrastructure for embodied intelligence, covering the entire pipeline from model training to evaluation.
Commercial Breakthrough: Three Global Delivery Milestones
On the commercialization front, Light Wheel Intelligence (Beijing) Technology has already achieved global leadership across three key domains:
Light Wheel Intelligence is currently the only company in the world capable of delivering all three capabilities at scale. In terms of revenue growth, the company recorded a 10× increase in total revenue in 2025, and its Q1 2026 revenue is expected to exceed the company’s entire revenue for 2025.
Its partners include leading organizations across the AI and robotics ecosystem, such as NVIDIA (英伟达), Google (谷歌), Figure AI, 1X Technologies, ByteDance (字节), Alibaba (阿里), Agibot (智元机器人), Galbot (银河通用机器人), Toyota (丰田), Bosch (博世), BYD (比亚迪), and Geely (吉利). All of the world’s top five world-model research teams have established collaborations with Light Wheel Intelligence. In addition, over 80% of the simulation assets and synthetic simulation data used by leading embodied AI teams globally originate from the company.
The industry partners introduced in this financing round will further extend Light Wheel Intelligence’s embodied data production and physical measurement capabilities into a wider range of real-world application scenarios, significantly expanding the sources of embodied data and simulation assets.
Beyond commercialization, the company is also actively contributing to the development of the broader industry infrastructure:
As a core contributor, Light Wheel Intelligence co-open-sourced the Isaac Lab-Arena benchmark framework with NVIDIA.
It is collaborating with World Labs to build a general pipeline spanning environment generation to large-scale evaluation.
Together with Tongyi Qianwen (通义千问), the company is building a reproducible and diagnosable industrial-grade evaluation loop based on RoboFinals-100, helping establish a standardized evaluation foundation for the embodied AI industry.
Meanwhile, LeIsaac, Light Wheel Intelligence’s proprietary simulation workflow, has been officially included in the documentation of Hugging Face, the world’s largest AI open-source community, becoming a standard simulation workflow used by millions of developers worldwide.
Building Data and Simulation Infrastructure for Physical AI
Every paradigm shift in AI has been driven by breakthroughs in infrastructure. In the era of deep learning, GPUs and CUDA defined the computing infrastructure that powered large-scale models. In the age of physical AI, a new layer of infrastructure is now emerging—data and simulation infrastructure for physical AI.
Dr. Xie Chen (谢晨), founder of Light Wheel Intelligence (Beijing) Technology, noted: “If GPUs and CUDA defined the computing infrastructure of the large-model era, then Light Wheel Intelligence is defining the data and simulation infrastructure for the physical AI era.”
The era of large-scale embodied data generation has arrived. Light Wheel Intelligence is building the long-term infrastructure that will support world construction, data production, and the evolution of intelligent systems.