Top-tier capital firms rarely join forces to bet on optical interconnection. Optolink Semiconductor is accelerating the underlying revolution in AI computing power.

Amid the collective sprint of the optical module industry in preparation for the dawn, a hidden investment and financing deal is expected to reshape the underlying landscape of AI computing power.

Public information shows that Optolink Semiconductor, a technology company specializing in Optical Input/Output (OIO) for inter - chip optical interconnection, has completed multiple rounds of financing within less than two years of its establishment. Recently, it has once again received joint investment from two top - tier capital firms. This is one of the relatively large - scale early - stage financings in the domestic optical interconnection chip track. Previously, Zhenzhi Venture Capital, as the founding investor, participated in the in - depth incubation stage of Optolink Semiconductor from scratch.

Whether it is the rapid completion of multiple rounds of financing in a short period or the rare joint effort of two top - tier institutions, both send a clear signal: the mass production and commercialization process of Optolink Semiconductor's OIO technology is accelerating comprehensively.

A beam of light is rewriting the rules of the computing power world.

In 2025, the optical module industry reached the peak of the spotlight. In the secondary market, the so - called "Yizhongtian" (Infinera, New Fiberhome Telecommunication Technologies, and Tianfu Communication) soared all the way, and their stock price trends became the focus of the market. Just as the industry was busy with the production capacity of 800G optical modules, a deeper - level technological change was brewing: OIO, the inter - chip optical interconnection technology, is standing at the starting point of a new round of computing revolution.

Meanwhile, in large - scale computing clusters with tens of thousands or even hundreds of thousands of graphics processing units (GPUs) for large - model training, over 90% of the energy consumption is used for data transfer rather than actual computing. Electrical interconnection has approached its efficiency limit. Inter - chip optical interconnection (OIO) is considered a crucial leap from electrical to optical connection: it switches the short - distance interconnection between chips from copper wires to optical paths, achieving orders - of - magnitude breakthroughs in multiple dimensions such as transmission energy consumption, bandwidth density, latency, and distance.

In the financing events in this field, it is more accurate to say that capital is investing in the opportunity of the transformation of China's underlying computing power structure rather than just a company. "It's like you've assembled a top - level F1 racing team but made them race on a country dirt road instead of a top - level F1 track," said Chen Chao, the CEO of Optolink Semiconductor, at an industry forum. "Electricity is better at computing, while light is better at connection. What we are doing is building a 'light - speed highway' between AI chips."

Chen Chao, CEO of Optolink Semiconductor

From Chip Stacking to Efficiency Improvement: Say Goodbye to Performance Anxiety in the Era of Large Models

In the era of large models, the role of data centers has transcended the traditional infrastructure of "computer rooms + power supply" and has become an energy system centered around computing power output.

However, this system is facing deep - seated efficiency bottlenecks. The problem lies not only in the limitations of chip performance but also in the poor connection efficiency between chips. Fundamentally, there are three main reasons: hardware limitations on bandwidth, imbalanced transmission energy consumption, and the physical limitations of copper interconnection.

Firstly, hardware conditions impose natural limitations. The obvious shortage of data transmission bandwidth severely restricts the utilization rate of computing power. In multi - level communication scenarios, such as GPU - to - GPU, cabinet - to - cabinet, and even inter - data - center communication, the existing bandwidth cannot meet the requirements of large - scale parallel computing.

Secondly, although the data volume is growing exponentially, the transmission energy consumption is seriously imbalanced. Take the current large models with training parameters in the hundreds of billions or even trillions as an example. The energy consumed by data movement between chips accounts for over 90% of the total energy consumption of the entire system.

In other words, less than 10% of the energy is actually used for computing, and the rest is wasted on "transporting" data. This inefficient energy consumption not only increases the operating cost but also reduces the actual available computing power, imposing an intangible "hidden tax" on AI infrastructure.

Thirdly, traditional copper interconnection has insurmountable physical limitations. Affected by the skin effect, copper wires can only conduct current on the surface, and the interior of the wire does not fully participate in the transmission, resulting in a large amount of energy being converted into heat and significantly increasing energy consumption.

As the transmission rate increases, the effective distance of copper interconnection decreases sharply, from tens of centimeters to just a few centimeters. This means that multi - GPU interconnection within a rack or across racks faces a physical bottleneck of being unable to connect. In large - scale clusters with tens of thousands of GPUs, this limitation directly restricts the expansion of overall computing power and becomes a key factor hindering the development of domestic computing power and the efficiency of large - model training and inference.

At this technological standstill, Optolink Semiconductor sees the starting point of innovation. It doesn't matter if the single - chip computing power has a certain gap with NVIDIA. If we can connect domestic chips with high - bandwidth and low - energy - consumption optical interconnection, it is entirely possible to surpass NVIDIA at the system level of "computing + interconnection," and the operating cost can also be reduced by orders of magnitude.

The Optolink Semiconductor team conducts OIO chip testing

Optical interconnection shows its unique value at this time - it is no longer a theoretical alternative but a feasible architecture to solve the existing computing power bottlenecks.

The advantages of optical interconnection lie not only in bandwidth improvement but also in system - level energy - efficiency optimization. Currently, the electricity cost accounts for a large proportion in large - scale clusters with tens of thousands of GPUs. Through Optolink Semiconductor's OIO solution based on optical interconnection technology, the bandwidth is expected to be increased by two orders of magnitude, and the energy consumption is expected to be reduced by two orders of magnitude. This means that the operating cost and energy consumption of data centers will be significantly reduced. With the application of this technology, the available value of China's computing power will be completely rewritten, and the traditional model of relying on chip stacking is being replaced by the new logic of "optical interconnection."

Overtaking on the Curve in the Process Technology Dilemma: The "Chinese Opportunity" of Optical Interconnection

Today, when China's semiconductor industry is still restricted by the "chokepoint" in advanced process technology, the logic of finding a breakthrough has shifted from single - chip performance to system - level optimization. Compared with the competition model of single - chip process upgrade, system optimization, optical interconnection, and full - link localization will become new fulcrums for China's computing power to overtake on the curve.

In this process, optical interconnection technology has definite application prospects in China.

Currently, three structural realities are in front of this market: First, the national demand for computing power has entered a phase of definite expansion. Reports show that the annual compound growth rate of the national computing power from 2020 to 2023 was about 30%, and leading enterprises have entered the investment cycle with hundreds of billions of capital expenditure (CAPEX). Second, plans such as the "East - West Computing Corridor" have made cross - regional computing power scheduling a rigid demand, and traditional interconnection solutions can no longer support this expansion. Third, China has a global supply - chain advantage in the optical module and packaging fields, and the cost of some products with specific specifications is significantly lower than that of overseas solutions.

Silicon - photonics wafer - level system testing

This means that optical interconnection is expected to be a home - field technology that China can scale up and implement first. It can support the expansion of hundreds of billions of computing power of leading enterprises by providing higher bandwidth and energy - efficiency ratio. With low latency and high - speed transmission, it can meet the rigid demand for computing power scheduling of super - nodes in intelligent computing centers. At the same time, relying on the supply - chain advantage of domestic optical modules and packaging, it can significantly reduce costs and achieve controllable and replicable large - scale deployment.

Optolink Semiconductor was founded precisely to seize this opportunity. Chen Chao described the potential of optical interconnection: "What we focus on is not just the amount of computing power but more importantly, the efficiency. A chip with high computing power may be less powerful at the system level than a well - coordinated domestic chip cluster if only 30% of its computing power can be utilized."

The core team members of the company include those from well - known domestic and foreign universities and research institutions such as the Massachusetts Institute of Technology, Tsinghua University, Zhejiang University, and the Chinese Academy of Sciences, as well as industry giants like Marvell. This all - star lineup supports its solid and remarkable technology system. This high - density technology accumulation has rapidly advanced its R & D process and engineering efficiency, enabling Optolink Semiconductor to turn theoretical solutions into mass - producible chips in a short period, significantly shortening the innovation cycle.

So far, Optolink Semiconductor's strategy is clear and bold: Use electricity for computing and light for interconnection, similar to how undersea optical cables support large - scale computing power transmission across racks and data centers.

Conceptual diagram of a future photonic server cluster

This "overtaking on the curve" approach allows domestic chips to potentially surpass in terms of system computing power even if the process technology lags behind by 2 - 3 generations or more. Optolink Semiconductor's technical route proves that optical interconnection is not a single - point technology but a lever for system - level optimization. By improving the signal transmission bandwidth and energy - efficiency between chips, the potential computing power of the entire data center can be unleashed.

From an industrial perspective, China has global competitiveness in the optical module industry and has a faster engineering iteration speed and cost - control ability. In Optolink Semiconductor's design, the entire process from chip design to advanced packaging can be completed domestically without relying on overseas foundries.

This not only speeds up the product launch process but also ensures the autonomy and controllability of computing power. This makes the company fit into the investment logic that capital is willing to support in the long term: optical interconnection is not only a technological innovation but also the infrastructure for the country's AI strategy.

Against this background, in addition to a solid technical foundation, the company is also building a scalable ecological framework. It is gradually collaborating with several leading domestic GPU enterprises to build an open "light - speed network" to provide technical support for the future layout of data centers across the country.

Optolink Semiconductor has chosen the "open - ecosystem" route similar to the Android system rather than the "closed - model" like NVIDIA's iOS. Any domestic GPU enterprise can connect to the optical interconnection network, thus forming a truly domestic intelligent computing system.

Capital's Bet: The Road to Reshape Domestic Computing Power

The attention of the capital market often most intuitively reflects industry trends and technological value.

To clarify the underlying logic of Optolink Semiconductor's establishment and development, 36Kr conducted in - depth value exploration and research on the company's shareholder background. Fundamentally, 36Kr found that Zhenzhi Venture Capital, the founding investor of Optolink Semiconductor, was the first to discover the value of this track and provided the best starting support for the startup team through its ECT in - depth incubation model.

Zhenzhi Venture Capital is positioned as a Venture Studio with an in - depth incubation model. Different from traditional venture capital (VC), it can act as a co - founder. Based on the trends of the technology industry and major social needs, it selects significant opportunities that are expected to change the future, makes bold investments, and accompanies entrepreneurs in refining technology and products, formulating strategies, and validating needs to complete the startup process of a company from scratch.

Optolink Semiconductor is a representative of Zhenzhi Venture Capital's in - depth incubation model. It was jointly initiated by Ren Xuyang, the founder and chairman of Zhenzhi Venture Capital, and Chen Chao, the partner of Zhenzhi Venture Capital, who also serves as the CEO of Optolink Semiconductor.

The essence of this in - depth incubation model is the belief that a startup company is more likely to succeed when it has three capabilities: "E" represents entrepreneurial spirit, which can lead the company to leap from scratch, from 1 to 100, and even to 1000, and withstand the ups and downs of an industry; "C" represents excellent financing ability, which determines the company's ability to drive itself into a flywheel effect; "T" means having top - notch scientists and engineers to lead the company to continuously break through in advanced technologies.

Zhenzhi Model - ECT Model

Under this model, the founding team dares to explore the technological boundaries and can flexibly layout, make quick decisions, and accurately take actions among the industrial chain, capital, and the market. Driven by this concept, Chen Chao and his team have advanced optical interconnection from laboratory innovation to an industrially replicable technical path.

The rapid entry of market - oriented funds further confirms the certainty of this track. The joint investment of top - tier investment institutions shows significant recognition of the infrastructure status of optical interconnection technology in the AI 2.0 era. At a more fundamental level, it also proves the high recognition of Zhenzhi Venture Capital's Venture Studio model by the mainstream funds in the industry.

A source close to this round of investment revealed: "This is not just an investment in a company but an investment in the key path for China's computing power to be self - controllable." It can be said that capital values not only the financial return but also the potential impact of this technology on the national - level strategic layout.

The breakthrough of Optolink Semiconductor lies in that it can not only provide high - bandwidth, low - latency optical interconnection solutions but also enable the scalable deployment of large - scale intelligent computing centers through systematic design, empowering the domestic computing power ecosystem. The combination of technology localization, system scalability, and ecological openness makes optical interconnection not only a technological innovation but also a breakthrough point for China's AI industry to over