GL Ventures leads the investment in AI agent enterprises

Recently, Nanjing Guanfuxin Software Co., Ltd. (hereinafter referred to as "Guanfuxin") has completed its seed round of financing, which is jointly led by GL Ventures and Lisuan Hexin, with Hengxu Capital and Jinyu Maowu participating in the follow-on investment. The financing amount of this round and the specific use of funds have not been disclosed yet.

Founded in 2026 and located in Jianye District, Nanjing, Guanfuxin mainly provides AI agent products for chip R&D enterprises. Its entry point is not chip manufacturing, but connecting semiconductor expertise with chip design tools to help engineers handle troubleshooting, design optimization, repeated verification and other work.

According to the company, Xu Zijing, founder and CEO, graduated from Nanjing University with more than 15 years of experience in chip R&D and large-scale project management. He once served as the head of the SerDes and UCIe product lines at CoreAhoy, promoting the commercialization of high-speed interface IP, and later served as the project manager of 3D AI high-computing power chips at Eastern Suanxin, participating in project R&D and mass production introduction. These experiences cover different links from chip function module development to whole-chip project management.

Chip R&D usually requires the cooperation of multiple teams, and repeated modification and verification between different tools. Guanfuxin hopes to undertake part of the repeatable engineering tasks: after the agent reads the project materials and identifies problems, it calls the corresponding tools to perform operations, and then adjusts the plan according to the results, rather than just providing text suggestions to engineers.

According to the company's disclosure, its products rely on the semiconductor knowledge base, self-developed and third-party electronic design automation (EDA) tools to assist in completing tasks such as project analysis, fault location and design optimization. Among them, performance, power consumption and area optimization are directly related to whether the chip can meet customer requirements, and are also the links that the engineering team repeatedly invests time in.

For mature IP subsystems, the company also proposes to drive related engineering delivery by agents based on IP that customers have already purchased. The IP here can be understood as a reusable chip function module, and subsystem delivery is to integrate these modules within a certain scope. Covering a more complete chip R&D process is the company's subsequent expansion direction.

From the perspective of industrial chain position, Guanfuxin mainly connects upstream design tools, IP and professional knowledge, as well as the R&D teams of downstream chip enterprises. According to its disclosed product path, customers do not need to replace the original tool chain as a whole, but introduce agents into the existing R&D process. The company also provides local deployment solutions for enterprises to adapt to the demand that chip design materials are inconvenient to be transmitted externally.

In terms of commercialization, Guanfuxin said that a number of agent products have been introduced by leading enterprises in the fields of network communication, cloud computing power, GPU and security monitoring, and have carried out industry-university-research cooperation with the academician team of Nanjing University. However, the existing public information has not disclosed specific customers, the number of paid deployments, contract amount and revenue scale, nor has it explained the respective charging methods for software authorization, deployment services and project delivery.

For such enterprises, entering the customer's R&D process is the starting point of commercialization, and whether they can continue to be used in subsequent projects determines whether the revenue can be sustained. What customers buy is not just an AI assistant that can call tools, but a shorter R&D cycle, less repetitive work, and delivery results that can pass the established verification process.

Capital's attention to this direction has seen multiple cases.

In July 2026, domestic chip design startup Xingxingyuan announced that it has obtained tens of millions of yuan of first-round investment from Wuyuan Capital. The company tries to use AI to undertake design planning, tool invocation and iteration tasks, and promote industrial cooperation through a one-time engineering service model. Compared with providing agent software to customers, this path is closer to delivering design results on a project basis, and the target customers also include enterprises that do not have a complete self-developed chip team.

In the same month, US-based ChipAgents announced an additional $60 million in Series A2 financing, with B Capital becoming the new investor. The company mainly provides agent platforms for chip design, verification and fault location, and discloses that it has been deployed in more than 120 semiconductor enterprises including MediaTek and Micron. The above deployment scale is disclosed by the company and cannot be directly equated with the number of paying customers.

At the same time, traditional EDA vendors are also embedding agents into their own products. Cadence acquired ChipStack in November 2025, and successively launched agent products for design and verification in 2026; Synopsys released the AgentEngineer product portfolio and Autopilot platform in September 2026, covering engineering processes such as verification and design implementation.

From these layouts, the industry is forming different commercial paths: existing EDA vendors add agent capabilities to their original tools, independent software companies connect customers' existing tool chains, and design service companies try to deliver R&D results directly. They are all competing for the budget of chip enterprises to improve R&D efficiency, but the delivery content and charging basis are not the same.

For startups like Guanfuxin, connecting different tools and adapting to customers' existing processes is a viable entry point; while vendors such as Synopsys and Cadence master the underlying design and verification tools, they can also deeply integrate agents with their existing products. Startups need to prove that their products can not only access tools, but also understand customers' project backgrounds, and stably generate incremental value in real tasks.

The next step that deserves more attention is whether Guanfuxin can convert customer onboarding into continuous payment, and whether the same set of products can be reused in different projects.

If each new customer requires a large number of engineers to re-adapt, the business will still rely heavily on project services; if it can reduce deployment and maintenance costs while ensuring delivery quality, it will have more opportunities to form a scalable and replicable software business.

This article is from the WeChat official account "Sci-Tech Innovation Board Daily", and is published with authorization from 36Kr.