Feidu Tech, known in Chinese as 飞渡科技, has used Beijing’s HICOOL entrepreneurship competition to position itself as a Chinese contender in spatial intelligence and Physical AI. QbitAI reports that the company won first prize at the HICOOL 2026 Global Entrepreneurship Competition for its Zhengrong Model project and received public support from Silicon Valley investor Bill Reichert.
Reichert, a partner at Pegasus Tech Ventures and a competition judge, invited Feidu Tech to make a presentation at a Pegasus global investor summit in Silicon Valley planned for 2027, according to QbitAI. The invitation is not a financing announcement, and the article does not say Pegasus has invested in Feidu Tech. It is nevertheless an unusual endorsement for a Chinese company pursuing models that are meant to understand and act in physical environments rather than simply process language.
Feidu Tech calls Zhengrong an industrial-grade spatial-intelligence model. The company says it has passed national filing in China and uses a five-layer architecture covering appearance, geometry, semantics, physics, and simulation. Those descriptions are Feidu Tech and HICOOL claims. They offer a view of the system’s design goals, but they do not independently establish performance against competing world models.
Feidu Tech Targets the Physical World Beyond Chat Interfaces
The company’s pitch reflects a growing change in China’s AI market. Large language models are built to interpret and produce language. Spatial or world models attempt to represent objects, locations, movement, and interactions in an environment. The intended uses can range from industrial simulation and infrastructure operations to robotics, design, mapping, and digital twins.
QbitAI says Feidu Tech’s technology has been deployed in more than 1,000 projects. One example cited in the report is a Beijing reservoir-scheduling system where Feidu Tech says its model shortened preparation for disaster simulations from months to days. That is a company-reported case study, not an independent assessment of the system’s accuracy or its effect on flood-management decisions. It does show the kind of industrial application Feidu Tech is prioritizing.
Reichert described Physical AI, also termed Entity AI in the report, as a post-LLM investment direction. The phrase is broad, but the commercial logic is straightforward. A model that can work with geometry, physics, and changing environments could support tasks that a text-only assistant cannot perform. The difficulty is that physical systems must handle incomplete data, unusual conditions, sensor errors, and safety limits that are less forgiving than an incorrect chat response.
China’s spatial-AI ecosystem includes several technical approaches. EastFrontier reported on September 5 that Chinese world models are responding to World Labs’ Atlas, with InSpatio pairing benchmark work with an open data initiative. Feidu Tech’s strategy is different in emphasis. It is presenting a layer-based industrial model and project deployments rather than a single benchmark result or generative 3D tool.
The distinction is important for customers. A factory, reservoir operator, or infrastructure manager is likely to care about scenario accuracy, integration with existing systems, and the cost of preparing simulations. A developer building a visual application may prioritize fast generation from images or video. The Physical AI label encompasses both ambitions, but each sector needs evidence on its own operating terms.
Competition Results Give Feidu Tech New Visibility
HICOOL is a Beijing entrepreneurship program that promotes domestic and international startup projects. Winning its top prize gives Feidu Tech a public platform, but the competition result should not be confused with validation from a commercial customer or a peer-reviewed technical evaluation. The company’s reported work in more than 1,000 projects is a larger indicator of commercial reach, although QbitAI does not provide a project-by-project breakdown.
The report also says Feidu Tech won 2026 CVPR competitions in multimodal understanding and reasoning and an ECCV competition related to embodied-AI interaction. Competition results can demonstrate technical skill under defined rules, but they do not reveal whether a model remains reliable after it is connected to industrial data, sensors, and business processes.
That gap between laboratory success and deployment is where spatial-AI companies will be tested. An industrial model must represent a site well enough to help with planning or simulation. It also has to be maintained when equipment changes, new sensor data arrive, or physical conditions differ from the training environment. Feidu Tech’s reservoir example is therefore more informative than a broad claim of Physical AI leadership, even though the available source is still the company’s own account.
China has been building other pieces of this stack. In August, EastFrontier covered Manycore’s Lux3D release for China’s spatial-AI push. Feidu Tech’s five-layer description suggests it is pursuing a more explicit connection between visual appearance, three-dimensional geometry, semantic labels, physical behavior, and simulation. The report does not disclose the model size, training data, pricing, or independent error rates, so those claims cannot yet be compared directly.
The Next Proof Point Is Repeatable Industrial Value
Reichert’s invitation matters because it links Feidu Tech’s Beijing competition win to a global venture audience. It does not guarantee overseas funding or an international commercial rollout. Investors will still need answers about the company’s revenue model, customer retention, deployment cost, and ability to operate across different environments.
The company’s industrial orientation could be an advantage. A reservoir simulation, a manufacturing site, or a logistics network offers defined assets and repeatable operating scenarios. Those features can make it easier to measure whether a model improves a process. They can also create high expectations, because an incorrect model recommendation can affect physical equipment, public safety, or project schedules.
Feidu Tech’s current public case is strongest as a signal of direction. It is trying to move China’s AI conversation from general-purpose chat to systems that represent real places and real processes. The HICOOL prize and Pegasus invitation give that effort greater visibility. Its reported reservoir work suggests a concrete application. But the available evidence remains largely company and competition material, so the company’s long-term standing will depend on independently observable deployments and technical results.
The opportunity is substantial if Feidu Tech can prove that its spatial model saves time without reducing the quality of planning. China’s push into world models has produced new benchmarks, 3D tools, and data alliances. Feidu Tech is adding another model: using spatial intelligence as an industrial operating layer. The company now has a prominent supporter and a high-profile competition win. The next question is whether customers can turn that promise into durable physical-AI contracts.