Cheng Yuan, Associate Professor at Shanghai Jiao Tong University, Founds Zhizi Guangyuan, Securing 5 Consecutive Seed Rounds in Half a Year

Hardware Kr learned that optical computing chip company Zhizi Guangyuan has completed a seed round of funding in the hundreds of millions of yuan. Since its founding, the company has completed five consecutive seed rounds, with investors including Haitong International Innovation Center, Yunqi Capital, Guotai Haitong, Xianfeng Changqing, Atom Ventures, Tongxin Capital, and the Shanghai Jiao Tong University AI Future Fund. The funds will be mainly used for core chip R&D and tape-out iterations, recruitment of top talent and team building, as well as application scenario expansion and ecosystem cooperation.

Zhizi Guangyuan was founded in May 2026 and focuses on reconfigurable all-optical computing architecture and photonic processor R&D, using the parallel computing characteristics of light to directly participate in AI computation, bypassing the bottlenecks of traditional electronic chips in process scaling, power consumption, and heat dissipation, and driving optical computing from specialized acceleration toward a more general-purpose computing platform.

At present, Zhizi Guangyuan has built an R&D team of nearly 50 people, whose core members have long been deeply engaged in the fields of optical computing and chip design, have cumulatively completed the design, tape-out, and verification of dozens of optical chips, and have published nearly 20 papers in the Nature/Science series. Founder Cheng Yuan is an associate professor at the School of Artificial Intelligence of Shanghai Jiao Tong University, has long been engaged in photonic chip and edge intelligence research, and previously served as co-founder/CTO of Xinmier Technology, leading the construction of a full-stack product system.

In terms of product layout, Zhizi Guangyuan has planned two product lines: server-side optoelectronic accelerator boards and edge-side optical computing chips. The first versions of the relevant chips for both products have been sent for tape-out, and deployable products are planned to be launched successively within the next six months.

At present, competition in AI computing power is shifting from the simple pursuit of peak computing power to how much effective intelligence output can be supported per kilowatt-hour. As AI moves from single-turn Q&A to agents' continuous thinking, tool calling, and task execution, token consumption is growing rapidly, and cost, power supply, and heat dissipation are becoming real constraints on the expansion of computing power.

Traditional electronic chips are already approaching the upper limit of Moore's Law, and process scaling alone can hardly keep up with demand growth. The idea of optical computing is to map matrix operations onto the interference and diffraction processes of light, using light propagation itself to complete large numbers of parallel multiply-accumulate operations.

At present, the optical computing industry is in the early stage of engineering implementation, with multiple technical routes proceeding in parallel in China. There are few available products on the market, and most are small-scale specialized computing, which makes it difficult to support real large-scale tasks and to switch between tasks.

"General-purpose optical computing is the industry trend, but it requires coordinated progress in architecture, devices, processes, packaging and testing, and software systems," said Cheng Yuan.

Specifically, Zhizi Guangyuan's technical route is a reconfigurable all-optical computing architecture. Through spatiotemporal reconfigurability, it achieves flexible adaptation to different models and computing tasks; it integrates the technical advantages of interference computing and diffraction computing, uses Chiplet to modularize interference and diffraction modules into the same architecture, and cascades multiple optical computing cores in a single chip to expand computing scale while balancing computing precision and flexibility. In addition, the company is simultaneously advancing "optical transmission and optical computing" and all-optical nonlinear technologies to reduce optoelectronic conversion in key computing links, and through software-hardware collaborative calibration, improve computing precision and chip consistency.

Reconfigurable: interference x diffraction x chiplet (Source/Enterprise)

English text is being prepared.

All-optical: optical transmission and optical computing x nonlinearity (Source/Enterprise)

At the product level, Zhulong OPU-08 is aimed at server computing acceleration, adopts the form of a PCIe optoelectronic accelerator board, and focuses on large-scale matrix operations in AI inference. Its specifications are a 256×256 computing core and 150 TOPS of computing power. It can be connected to existing servers and provide "photonic tokens" for cloud deployment of large models. Xuanjian TOCU-07 is aimed at low-power intelligent devices at the edge. Its specifications are a 128×128 computing core and 100 TOPS of computing power. In addition to optoelectronic computing, it also supports collecting weak ambient natural light for direct computation, eliminating optoelectronic encoding on that input path.

Cheng Yuan said that Zhulong is consistent with the form of current mainstream server optoelectronic accelerator boards, and its highlight lies in more general-purpose matrix acceleration and model deployment capabilities; Xuanjian is aimed at low-power edge scenarios, where there are not yet many similar products on the market. The longer-term goal is a general-purpose photonic GPGPU integrating training and inference, with the first version expected to be launched at the end of 2027 or early 2028.

Server optoelectronic accelerator card Zhulong OPU-08 (Enterprise/Source)

In terms of engineering, optical computing faces consistency issues such as device variation, noise, and temperature drift. Through the cooperation of software and hardware compensation modules, Zhizi Guangyuan has added an automated nanosecond-level calibration module to its optical chips. The company says it can reduce matrix operation errors to the 10^-6 level to improve the precision consistency of mass-produced chips. The company is currently working with domestic foundries and packaging and testing plants to advance mass production, but product performance still needs to be tested and verified after the chips come back.

According to the company's plan, the near-term focus is to complete the packaging and testing, board-level joint debugging, model adaptation, and customer verification of Zhulong and Xuanjian, ensuring that products are launched as planned; subsequently, it will gradually advance larger-scale optical computing boards and general-purpose photonic GPGPUs.

The company hopes to rely on the team's accumulation in cutting-edge technology R&D, chip engineering, and system-level product deployment to accelerate the implementation and application of optical chips by 2-3 years. The realization of this goal still requires coordinated support from the upstream and downstream industry chain.

Cheng Yuan said that the company will also focus on AI computing infrastructure, the optical communications and silicon photonics industry chain, as well as upstream and downstream segments such as manufacturing, packaging and testing. The maturity of the optical communications and silicon photonics industry chain provides the device, process, and packaging foundations for optical computing; industrial cooperation will revolve around process packaging and testing, server manufacturers, edge-side application partners, and model adaptation.

In the long term, Zhizi Guangyuan is committed to promoting the implementation of general-purpose optical computing, enabling the same set of hardware to adapt to different models and tasks within the computing scope where light excels, and ultimately building an integrated training-and-inference photonic GPGPU and supporting software system.

"Specialization is the first step for optical computing toward application, and generality is the next step for optical computing to become computing infrastructure," Cheng Yuan said.

The following is an excerpt from an interview between Hardcore and Cheng Yuan, founder of Zhizi Guangyuan (edited):

Hardcore: Why do few in the industry combine interference and diffraction on a single optical computing chip? Why did Zhizi Guangyuan choose this route?

Cheng Yuan: We implement the combination of interference and diffraction on a single optical computing chip. This is first an architecture-level innovation; the team has long advanced system-level model deployment from the level of reconfigurable architecture, and its research is at the international frontier.

Adopting this technology is intended to break through the limitations of traditional optical computing architectures in flexibility and computing scale, achieve reconfigurability inside the optical chip, and adopt an all-optical computing path based on interference and diffraction. Reconfigurability determines how many tasks optical computing can adapt to, while the all-optical path opens up greater energy-efficiency space and reduces the need for optical-electrical conversion to near zero. Only by achieving reconfigurability and all-optical operation can the so-called general-purpose optical computing be realized—adapting to different tasks and models on the same set of hardware. The difficulty lies in how to regulate the internal operations of optical computing, implement reconfigurable algorithms inside the optical chip, and at the same time ensure consistent performance across every chip, which requires coordination across the entire industry chain from architecture design and process to packaging and testing.

Hardcore: The product has now been taped out. How are customer and mass production progress?

Cheng Yuan: We have already accumulated many potential customers. A board card manufacturer plans to launch an optoelectronic accelerator board, and our optical chip will serve as a coprocessor acceleration module on its GPGPU. In terms of mass production, we have established cooperation with well-known domestic foundry and packaging and testing manufacturers and are advancing related volume production work. At the same time, together with software and hardware calibration and compensation modules, we ensure consistency after the optical chips come out.

Hardcore: What is the hardest thing to overcome for optical computing to move into practical application?

Cheng Yuan: The hardest is stable delivery and batch consistency. The experimental performance of a single chip must become the stable performance of batch products in customer environments. Device variations, noise, and temperature drift all affect computing results, so we add a software-hardware collaborative nanosecond-level calibration module at the underlying level and move packaging and testing and model adaptation earlier into the design stage, solving engineering challenges from the source and avoiding leaving problems until the delivery stage.

Investor perspectives:

Yunqi Capital stated: Yunqi has long been optimistic that optical computing is expected to bring new breakthroughs to AI computing infrastructure. The Zhizi Guangyuan team has deep accumulation in reconfigurable all-optical computing architecture and also possesses industrial experience from chip R&D to large-scale delivery to major customers. We are optimistic about Zhizi Guangyuan's long-term potential in the optical computing field and look forward to the team promoting the large-scale application of next-generation AI computing architecture in real scenarios.

Zhizi Guangyuan's original exploration in reconfigurable architecture and optical transmission and optical computing has shown us new technological possibilities. Teacher Cheng Yuan and the team have both years of scientific research accumulation and complete entrepreneurial experience from chip R&D to large-scale product implementation; they can both explore the boundaries of technology and know how to bring technology out of the laboratory. Such a founding team is extremely rare.

Guotai Haitong stated: The second half of the development of the AI computing industry is moving toward reducing token costs. Optical computing chips are expected to alleviate the dilemma of AI computing shortages. We are optimistic that Zhizi Guangyuan's technology route has the core advantages of generality, reconfigurability, and large-scale computing capability, as well as successful experience from frontier scientific research to semiconductor engineering implementation.

Haitong International Innovation Center stated: We are optimistic about the innovation potential and team technical advantages of the Zhizi Guangyuan team in frontier photonics directions, and will firmly accompany the team in deep technology research and industrial implementation. Relying on our own hard-tech incubation ecosystem resources, we will continue to connect Zhizi Guangyuan with industrial partners, research platforms, and industry channels to help release technological value. We look forward to Zhizi Guangyuan's continued breakthroughs, forming differentiated advantages in the optical computing track, and jointly helping build the domestic frontier computing industry ecosystem.