Guosen Securities has issued a research report expressing continued confidence in KNOWLEDGE ATLAS's (02513) strong technological foundation and leading model capabilities. During this phase of rapid large model development, the primary focus is on revenue growth trends. Considering the accelerated demand driven by the proliferation of Agent scenarios both domestically and internationally, and the ongoing upward revisions to revenue from overseas model companies, the firm forecasts KNOWLEDGE ATLAS's operating revenue for 2026-2028 to be RMB 3.1 billion, RMB 8.6 billion, and RMB 20.4 billion, respectively. The "Outperform" rating is maintained.

Core Strengths and Talent

The firm's research and development DNA is its greatest advantage, featuring industry-leading talent density. Its technological foundation stems from the Tsinghua KEG Lab and the AMiner platform, with a core management and research team boasting prominent backgrounds. Key members hail from top-tier institutions like Tsinghua University. Co-founder Tang Jie is a Fellow of IEEE, ACM, and AAAI, a Chair Professor at Tsinghua University's Department of Computer Science sponsored by WeBank, and Director of the Fundamental Model Research Center at the Institute for Artificial Intelligence, wielding significant influence in academia. Furthermore, the company's number of R&D personnel and investment scale surpass industry peers. It employs 657 R&D staff, constituting 74% of its total workforce, with R&D expenses exceeding RMB 3.0 billion in 2025, leading the industry.

Leading Programming Capabilities and Data Flywheel

KNOWLEDGE ATLAS's GLM series of large models is evolving along the path of "enhanced coding capability—Agentization—long-duration tasks—autonomous operating system." Starting with GLM-5, it has established a leading position in the programming domain, with a strong focus on improving model coding abilities. The company's talent advantage is evident across dimensions like data and model architecture. After establishing this lead, as more high-quality programmer user data accumulates, the company is gradually forming a robust data flywheel effect in programming, which is expected to sustain its leading edge.

Accelerating MaaS Revenue and Rising Token Prices

Prior to 2025, the company's business model primarily relied on private deployment, with cloud deployment accounting for a lower proportion and exhibiting unstable gross margins. Starting in 2026, the explosion in demand for coding and Agent scenarios is expected to make MaaS (Model-as-a-Service) the primary revenue source. Given the company's sharp focus on models and products, it primarily achieves its business model closure through API calls. This year, as the value generated per model token has increased, model prices have continued to rise. The brokerage anticipates that the gross margin for cloud deployment will also see continuous improvement.

Key Risk Factors

The report highlights several risk factors: 1) Overly optimistic profit forecast risk. 2) Policy and regulatory risk. 3) Technological iteration and competitive risk. 4) Computing power and supply chain risk. 5) Risk of AI commercialization falling short of expectations. 6) Macroeconomic volatility risk.