Knowledge Atlas Technology Joint Stock Company Limited
2513 · HKEX · China
zhipuai.cnFinancials as of FY2025
Builds general-purpose AI models in China and earns primarily by metering paid access to them, mixing usage-based cloud fees with packaged deployments sold mostly to enterprises.
- Depends onDownstream position: depends on 18 industries, supplies 6
- ScaleMarket cap is $59.28B, higher than 95% of all stocks globally
What this company is and how it runs — written from structure, not news.
It sits between outside providers of computing infrastructure and training data on one side, and enterprises, public-sector bodies, developers and individual users on the other. It turns raw computing power and data into trained models, then delivers that capability back out through programming interfaces, on-site software packages, and autonomous agents that carry out tasks on a customer's behalf.
Revenue comes mostly from selling and installing large models directly on customer premises for a packaged price, charged once or annually, with the remainder from cloud access billed by usage or subscription and from deploying task-executing agents built on those models.
Growth runs through parallel mechanics: cloud access priced by usage or subscription, which scales automatically with how much customers consume, and on-premises deployments sold as discrete packaged engagements to enterprise and public-sector buyers. By its own account, how far this can scale is gated less by customer demand than by how much advanced computing capacity it can secure, how much specialized research talent it can attract and keep, and how much it can keep investing in research.
By the company's own account, it depends on outside suppliers of computing hardware and cloud computing capacity, on third-party sources for the data used to train and evaluate its models, on a small pool of specialized research scientists and engineers, and on keeping its models compatible with computing hardware and platforms built by others. More broadly, it sits downstream of a wide range of supplying industries.
Its own account describes institutional buyers, among them private businesses, public-sector bodies, research organizations and government bodies, alongside individual end users and independent developers who access its models directly. It states that no single customer accounted for a large share of revenue in its most recent full year on file, and only one customer is named specifically in its disclosures, with the rest identified only by code. It also sits upstream of a narrow band of industries that draw on what it supplies.
This way of building paid access around large AI models is not unique in shape: CompanyGraph places a broad group of similarly structured companies in the same category, so structure alone does not show whether rivals can copy it. The company itself claims a set of strengths: being among the earliest Chinese developers of very large self-built models, a wide portfolio of models, roots in academic research, a single platform combining model access and tooling, and an ecosystem built around open-source releases and autonomous agents. These are the company's own claims about itself, not an independent comparison against competitors.
The company's own disclosures describe customer contracts as typically running about a year or less, with cloud access billed by usage or subscription rather than a longer commitment, and it has not disclosed contracted revenue extending beyond that period. Nothing in what it discloses names a specific switching cost, data migration barrier, or customer retention rate, so its own account does not point to strong friction holding customers in place.
By its own account, growth is limited less by customer demand than by its ability to secure enough advanced computing capacity, retain scarce specialized research talent, and keep funding a large, continuing research effort, together with the pace of regulatory change and its ability to keep its models compatible with hardware and platforms it does not control.
The company's own risk disclosures list continuous technological change and the need for heavy, sustained research spending first, ahead of its dependence on its research team and senior leadership and on outside providers of computing resources. It and a number of its subsidiaries have also been placed on a United States export control list that limits, without special authorization, their access to certain controlled technology. Separately, its financial statements on file show a year in which it recorded a net loss rather than a profit.
The company operates under Chinese regulatory oversight of internet, data and cybersecurity services, and separately, it and a number of its subsidiaries have been restricted by United States export-control authorities from accessing certain controlled technology without special authorization. It also carries exposure to foreign-currency movements that it has chosen not to hedge.
Read from the company's own filings and public materials (gathered August 2026) together with figures CompanyGraph recomputed from its statements. Written August 2026. A question with no evidence behind it is left out rather than answered.
Sign in to view price data.
Sign inThe reported statements, read against the company's own industry.
Shared structure with peers — never a ranking.
Structural observations derived from financial data, industry benchmarks, and supply chain position.
Supply Chain
Scale
Companies that share the same coordination system — how they create, deliver, or capture value.