Shanghai chip startup Xijian Technology has raised nearly 500 million yuan across three financing rounds in its first year, according to a Cailian Press report. The company develops brain-inspired vision chips designed to extract useful visual information with much lower power and data-transfer requirements than conventional image pipelines. The financing is an early but notable vote of confidence in a part of China’s AI hardware market that sits between sensors, edge computing, robotics, and machine perception.
Cailian Press said Xijian’s latest angel financing brought its cumulative funding to nearly 500 million yuan. The report named Xuhui Venture Capital, Matrix Partners China, Huaying Capital, Inno Angel Fund, and the Tsinghua Alumni Seed Fund among the investors. Xijian was founded in 2025 and is based in Shanghai’s Xuhui District. Its CEO, Yang Zheyu, is a Tsinghua University brain-inspired-computing PhD and a co-first author of the Tianmouc research paper, according to the report.
The company is not selling the familiar story of an ever-larger AI accelerator. Its thesis is that a machine should not have to move every frame from a camera through a conventional pipeline before deciding what matters. Neuromorphic visual hardware seeks to process changes, motion, and other features in a way that resembles aspects of biological vision. If it works as intended, the result could be useful where a robot, vehicle, drone, or industrial device needs to respond quickly without sending a constant flood of data to a distant server.
A Different Route to Machine Perception
The underlying Tianmouc work attracted attention because it proposed a complementary vision architecture that combines different types of information from the visual world. Xijian is attempting to turn that research direction into production hardware. Cailian Press said its first-generation chip has been powered on and that its second-generation product has completed tape-out, the design milestone at which a chip is sent for manufacturing.
The report attributed a series of performance characteristics to the company’s second-generation chip: a 5-megapixel resolution, sensing at 10,000 frames per second, a 130-decibel high dynamic range, 90% lower transmission bandwidth, and power consumption of roughly one-tenth of a conventional approach. Those are company-reported specifications, not independently benchmarked results, and they should be treated accordingly. The same applies to the startup’s claims about response latency and the quality of the visual tokens it can deliver to large models.
Even with that caution, the architecture is strategically interesting. Conventional image systems often generate more raw information than an edge device can economically transmit and process. A vision sensor that filters or represents information more efficiently could reduce demands on memory, bandwidth, and power. That is valuable for AI systems operating outside a data center, where battery life, heat, and network connectivity are practical constraints rather than abstract engineering problems.
China’s AI chip strategy increasingly includes these specialized components. The country is investing in large training clusters and general-purpose accelerators, but it also needs sensors, controllers, interconnects, and storage that determine whether an AI system can function in the physical world. EastFrontier’s coverage of Xiaomi’s three new processors showed how major consumer-technology groups are trying to extend silicon design into mobile AI and autonomous-driving workloads. Xijian is pursuing a narrower but potentially important sensing layer.
Why Robotics and Vehicles Are the First Target Markets
Cailian Press identified embodied robots, autonomous driving, industrial inspection, high-speed imaging, and drone inspection as intended use cases. All share a demand for rapid perception in changing conditions. A robot moving through a warehouse cannot wait for a cloud response to avoid a collision. A vehicle must recognize changing light and motion while managing limited energy. An industrial system may need to identify a brief anomaly before it disappears.
These are difficult environments for ordinary cameras. Bright sunlight, shadows, reflective surfaces, and fast motion can overwhelm static image-capture workflows. A high dynamic range can help retain detail between bright and dark areas, while high temporal resolution can capture rapid movement. The challenge is that the startup must demonstrate such capabilities in real products, not only in laboratory specifications.
The edge-AI opportunity is also tied to China’s desire for more self-reliant AI hardware. Domestic supply chains cannot be built around a single processor category. They require a broad set of components that can be integrated with local CPUs, accelerators, memory products, and operating environments. The same lesson appears in storage. EastFrontier recently reported on DapuStor’s planned Hong Kong listing, an example of how AI demand is pulling attention toward supporting hardware layers that are not always visible to end users.
For Xijian, the immediate question is whether its chip can deliver a compelling system-level advantage. A sensor may be technically novel, but customers will ask whether it reduces the cost of a robot, improves safety in a vehicle, or allows an industrial device to work in conditions where alternatives struggle. They will also ask about software tools, manufacturing yields, component reliability, and compatibility with their existing AI models.
Funding Gives the Startup Time to Prove Its Silicon
Nearly 500 million yuan of cumulative funding is substantial for a startup founded only in 2025. It provides room to move from research-based prototypes toward a usable product cycle. The next steps are likely to be far harder than attracting early capital. Tape-out is an important milestone, but a company still must validate manufactured silicon, build reference designs, secure customers, and demonstrate that performance and power targets hold in real deployments.
The report says Xijian’s team includes engineers with backgrounds in image sensors and chips, including experience from companies such as Sony and Nvidia. That kind of operational experience can matter as much as a scientific paper when a startup moves toward mass production. Designing an unconventional sensor is one task. Building a stable supply chain and helping customers integrate it is another.
Shanghai has made brain-inspired intelligence one of several frontier areas it aims to develop, and national policy has also elevated research into advanced smart-chip technologies. Policy support can provide a favorable environment, but it will not resolve commercial risk. Xijian still has to persuade manufacturers that its approach is more useful than incremental improvements to existing camera and processor platforms.
The company’s financing deserves attention because it reflects a broader change in China’s AI investment market. Capital is moving toward hardware that can make models useful beyond chat windows and cloud data centers. Xijian’s promise is to turn visual data into a more efficient input for those systems. Its challenge is to prove that a neuromorphic vision chip can become an indispensable part of the next generation of Chinese robots, vehicles, and industrial AI devices.