CITIC Securities: The value of EDA is expected to be reassessed, and it is recommended to focus on relevant domestic companies

The Zhitong Finance App learned that CITIC Securities released a research report saying that AI EDA is becoming the next important direction for AI agent implementation. The value of EDA is expected to be re-evaluated, and it is recommended to focus on domestic EDA companies. EDA is gradually evolving from a “tool” to an “smart body” to become an agent operating base for the entire chip design process. EDA leaders such as Synopsys are progressing intensively, and AI EDA has begun to enter a verification period. AI EDA drives up the value of EDA companies, and the triple upgrade of business models, barriers, and requirements. Domestic EDA launched AI efficiency improvement tools, and engineering data became the core moat.

CITIC Securities's main views are as follows:

EDA is gradually evolving from a “point tool” to an “smart body.”

The essence of AI EDA is to use AI agents to take over the most laborious “design-verification-modification” cycle of chip design: from early AI-assisted PPA optimization, to generative Copilot to help engineers write code, to today's Agentic EDA — multiple AI agents independently orchestrate the complete design process. According to Synopsys' official website, Synopsys divides AI EDA autonomy into L1 to L5 levels according to autonomous driving logic: L1 is knowledge assistance, L2 is an operational agent that can independently complete specific tasks, L3 is multi-agent collaborative planning, and L5 is fully autonomous; this year, L3-L4 multi-agent workflows, such as natural language generation RTL, automatic verification and convergence, etc., have been demonstrated several times. This means that AI is no longer an isolated point tool, but is becoming an agent operating base for the entire chip design process.

EDA leaders such as Synopsys are progressing intensively, and AI EDA has begun to enter a verification period. According to Synopsys' website:

On September 28, Synopsys released AgentEngineer's seven long-term agents and the unified Autopilot platform. AgentEngineer is a set of long-term agents for specific fields that can “reason, plan, and execute” to cover complete engineering processes such as verification, implementation, simulation and mixed-signal design, and manufacturing. The measured data showed that verification convergence was accelerated by up to 50 times, coverage increased by 20%, productivity increased by 30%, token efficiency increased by 2 times, and delays were significantly reduced.

On October 1, Synopsys and OpenAI signed a multi-year binding agreement to jointly develop GPT-Synopsys, a dedicated chip design model, and establish a revenue sharing mechanism. The model is trained as a “native expert user” of EDA tools, which can directly manipulate tools, interpret output, and iteratively optimize designs. Engineers can entrust design goals such as PPA optimization and timing convergence, and the agent can independently run the tool and deliver verified results.

On October 10, Synopsys also officially announced a partnership with Amazon. The two sides signed a multi-year agreement worth more than 1 billion US dollars, extending the cooperation from IP licensing to the field of cloud and AI-driven engineering, and launched the “license fee+royalty” model for the first time — with the mass production and shipment of self-developed chips such as Amazon Graviton, Trainium, and Nitro, Synopsys is expected to receive continuous royalty revenue.

Drive the value increase of EDA companies, and triple upgrade the business model, barriers, and demand.

1) Model transformation: From selling licenses to a three-tier monetization system of “agent subscription+computing power consumption+performance sharing”, taking GPT-Synopsys as an example, the OpenAI payment tool authorizes subscription fees for training models. After product launch, both parties share revenue for chip design improvements according to the model;

2) Increased barriers: Large models are generic capabilities, but the real moat is EDA's decades of accumulated engineering knowledge, process data, signing processes, and trusted verification engines. AI agents must rely on deterministic EDA engines for signature verification.

3) Demand expansion: A significant increase in design efficiency will stimulate more design projects. Global technology giants collectively develop their own chips, superposition Chiplet/3DIC complexity leaps, and EDA enters a new upward cycle. According to Synopsys' financial report, it has raised its long-term revenue growth target to the middle double digits, and the non-GAAP operating margin is expected to reach about 50% in FY2030.

Domestic EDA launched AI efficiency improvement tools, and engineering data became the core moat.

Nine days at Huada University: AI matrix penetration, Agentic platform card slot. According to the company's announcement, the company launched the AI-driven process diagnosis platform Vision, Aether Coder, etc., with a defect identification accuracy rate of 99%, which is 6 times more efficient than manual labor. Aether, a fully customized storage circuit platform, achieved a 20% + increase in simulation anti-bid rate, 5 times faster full layout, 6 times faster key line tracking, and 15-30 times faster performance of the advanced packaging platform Storm engine. According to Chairman Liu Weiping's speech at the 2026 IICIE conference, the company clearly defined 2026 as the year “Agentic AI becomes an industry consensus” and is building an intelligent platform for multi-agent collaboration.

Overview Electronics: AI drives DTCO and opens up full-link automation. With “AI Empowering DTCO” as its strategic direction, the company advocates a full-link collaborative and systematic shortening of the iteration cycle through process, device, model, PDK to circuit simulation. The next-generation SPICE model automated extraction platform MS-SDEPM integrates the entire process of model analysis, parameter extraction and quality verification to efficiently improve modeling accuracy and efficiency through targeted collaborative optimization. The new AutoCell standard unit layout automation platform automates the entire process from layout design to verification, and effectively addresses the lengthening design cycle bottlenecks brought about by advanced process miniaturization.

Guang Liwei: Using a “knowledge base+smart body” architecture, yield analysis was changed from “people looking for answers” to “answers looking for people”. The SemiMind semiconductor large model platform collaborates deeply with models such as DeepSEEK-R1 to achieve professional intelligence such as design rule inspection, automatic test plan generation, and independent YMS/DMS analysis. SemicLaw's enterprise-level digital employee platform transforms the EDA process from “people looking for information, people writing scripts” to “task-driven, intelligent self-orchestration and operation”. QuickRoot uses AI one-click root cause analysis to increase the efficiency of troubleshooting bad problems tens of times, reducing the complicated analysis process that would have taken weeks to a few hours.

Risk factors: the company's technological innovation, product upgrades, and performance growth fall short of expectations; AI application progress falls short of expectations; downstream demand falls short of expectations; industrial policy falls short of expectations; industry competition intensifies; and the risk of international geopolitics intensification.