2026033101549
Overview unavailable.
Annual Results Notice
- Hong Kong Exchanges and Clearing Limited and The Stock Exchange of Hong Kong Limited disclaim responsibility for the announcement’s contents.
- The notice states that no representation is made regarding the accuracy or completeness of the announcement.
- It identifies Knowledge Atlas Technology Joint Stock Company Limited, also named in Chinese as 北京智譜華章科技股份有限公司.
- The company is a PRC-incorporated joint stock company with limited liability, listed under stock code 2513.
- The document announces annual results for the year ended 31 December 2025.
– 1 –
Hong Kong Exchanges and Clearing Limited and The Stock Exchange of Hong Kong Limited
take no responsibility for the contents of this announcement, make no representation as to
its accuracy or completeness, and expressly disclaim any liability whatsoever for any loss
howsoever arising from or in reliance upon the whole or any part of the contents of this
announcement.
Knowledge Atlas Technology Joint Stock Company Limited
北京智譜華章科技股份有限公司
(A joint stock company established in the People’s Republic of China with limited liability)
(Stock Code: 2513)
ANNUAL RESULTS ANNOUNCEMENT
FOR THE YEAR ENDED 31 DECEMBER 2025
Knowledge Atlas 2025 Financial Results
- The Group reported a massive 131.9% year-on-year revenue increase to RMB724.3 million, driven by their 'foundational model + platform + ecosystem' strategy.
- Despite revenue growth, the company saw a significant net loss of RMB4.7 billion, largely due to a 44.9% increase in R&D expenses totaling over RMB3.1 billion.
- The company defines its commercial value through a specific formula: the product of the 'Upper Bound of Intelligence' and the 'Token Consumption Scale.'
- Global expansion has reached 218 countries, with the GLM series models achieving top rankings in both open-source and Chinese domestic evaluations.
- Management is pivoting toward 'industrial-grade Agentic Engineering' and addressing a 2026 computing power shortage through domestic chip optimization.
If the upper bound of intelligence determines the pricing power of technology, then the scale of token consumption determines the magnitude of commercial value.
The Board is pleased to present the audited consolidated results of the Group for the year
ended 31 December 2025, together with the comparative figures for the corresponding period
in 2024.
Certain amounts and percentage figures included in this announcement have been subject to
rounding adjustments, or have been rounded to one or two decimal places. Any discrepancies
between the total amounts listed in any tables, charts, or other data and the sum of the
individual amounts listed therein are due to rounding.
FINANCIAL SUMMARY
Year Ended 31 December
2025
2024
Year-on-
Year Changes
RMB in thousands
%
Revenue
724,334
312,414
131.9
Gross profit
296,656
175,889
68.7
R&D expenses
(3,180,443)
(2,195,436)
44.9
loss for the year
(4,718,167)
(2,958,007)
59.5
Net adjusted loss for the year
(3,181,972)
(2,465,569)
29.1
– 2 –
MANAGEMENT DISCUSSION AND ANALYSIS
Business Review
Business Overview
In 2025, the Company’s total revenue reached RMB724.3 million, a year-on-year increase of
131.9%. This is the best testament to the market for our long-standing strategic perseverance
in the “foundational model + platform + ecosystem” approach. The GLM series has not only
topped the global open-source rankings and ranked first in China, but we are also committed
to transforming leading cognitive capabilities into tangible productivity, driving the industry
towards industrial-grade Agentic Engineering. Currently, our programming, agent, and
enterprise-level large model services have transcended geographical boundaries, covering 218
countries and regions worldwide, and collaborating with over 4 million small and medium-
sized enterprises and developers to build an ecosystem. The journey towards AGI is long,
and we will continue to stay grounded, exploring the unknown boundaries with ultimate
foundational model capabilities.
Firm Strategic Positioning
Since our establishment in 2019, we have set “enabling machines to think like humans” as
our sole objective. On the journey towards AGI, we have always been convinced that “the
breakthrough at the upper bound of intelligence” is the only physical first principle of this era.
If the upper bound of intelligence determines the pricing power of technology, then the scale
of token consumption determines the magnitude of commercial value. We have internally
derived a concise formula:
AGI Commercial Value = Upper Bound of Intelligence × Token Consumption Scale
In 2025, Knowledge Atlas has demonstrated this formula through its actual position: on the
“upper bound” side: with the high-frequency iteration from GLM-4.5 to GLM-5, we have
consistently ranked first globally in open-source and first in China in international mainstream
evaluations, firmly securing our place in the global top tier. This sustained generational
leadership has granted Knowledge Atlas the core pricing power in cognitive intelligence.
On the “scale” side: The consumption of paid tokens has achieved exponential growth with
the deep penetration of GLM in meta-scenarios such as programming (coding) and agents.
We have significantly reduced unit costs while achieving steady improvement in gross profit
performance through extreme engineering optimization on the inference side.
– 3 –
Entering 2026, the computing paradigm is undergoing a radical transformation. The explosive
application of OpenClaw has prematurely ignited a token consumption frenzy. Faced with the
computing power shortage that has outstripped supply since February 2026, we will continue
to increase our investments, particularly in the “Day 0” adaptation of domestic chips and the
optimization of software-hardware integration. We push inference performance to its limits
The Leap to Agentic Engineering
- Knowledge Atlas transitioned from simple code generation to 'Agentic Engineering,' where AI acts as a digital engineer capable of autonomous planning and iteration.
- Technical optimizations like Muon Split and MLA-256 reduced KVCache usage and deployment costs by 50% while maintaining high performance.
- The introduction of the Slime framework enables asynchronous reinforcement learning, decoupling generation from training to maximize GPU utilization.
- The GLM-5-Turbo model emerged as the world's first OpenClaw foundational model, designed for secure, long-chain enterprise tasks.
- A shift in the developer market has occurred where Knowledge Atlas successfully raised prices by 30%, signaling a move from price wars to a focus on 'technological value.'
- Hardware-software co-design has allowed domestic chips to achieve inference efficiency comparable to international top-tier hardware.
AI is no longer just a simple code generator but a “digital engineer” capable of autonomous planning, testing, and iteration.
not for short-term profitability, but to support the continuously rising exponential curve of
high-quality token consumption. We firmly believe that the key to victory always lies in the
extreme execution of this formula in this marathon of computing power, data, and intelligence.
Innovative Core Technologies
Knowledge Atlas has consistently adhered to the GLM self-developed architecture. In 2025,
we took the lead in completing the leap from Vibe Coding to Agentic Engineering. AI is no
longer just a simple code generator but a “digital engineer” capable of autonomous planning,
testing, and iteration. At the foundational architecture level, we achieved stable model training
through the Muon Split optimization strategies and MLA-256 enhancements, significantly
reducing KVCache usage while delivering performance equivalent to GQA-8. Furthermore, by
employing a dynamic sparse attention mechanism, we overcame the computational challenge
of long-sequence reasoning, cutting deployment costs by 50% with zero performance
degradation. We have also introduced the Slime framework to achieve an efficiency revolution
in asynchronous reinforcement learning, addressing the idle-time pain point in long-
sequential tasks for agents and achieving complete decoupling of generation and training.
Slime maximizes GPU utilization and ensures robustness in large model training through the
Prefill-Decode (PD) separation and heartbeat fault tolerance mechanisms. Combined with our
proprietary direct two-sided importance sampling algorithm, we have overcome the action-
reward alignment challenge in asynchronous training, enabling the model to efficiently learn
from over 10,000 real software engineering environments. This system supported the birth of
GLM-5-Turbo as the world’s first OpenClaw foundational model, enabling secure execution
of long-chain enterprise-level tasks.
We recognize the importance of computing power autonomy, and the domestic adaptation of
GLM-5 has gone beyond simple operator transplantation, entering the Co-design (software-
hardware collaborative design) phase. At the underlying kernel level, we have maximally
hidden memory access and communication latency through customized fused kernels like
Lightning Indexer and FlashComm communication optimization. This deep tuning has enabled
GLM series models to achieve inference efficiency on domestic chips comparable to that
of international top-tier chips, realizing a perfect closed loop between the “upper bound of
intelligence” and the “computing foundation”.
– 4 –
Remarkable Business Progress
We firmly believe that AGI Commercial Value = Upper Bound of Intelligence × Token
Consumption Scale. Over the past year, leveraging the generational leadership of the GLM
series on the “upper bound” side and extreme cost optimization on the inference side,
Knowledge Atlas has achieved a comprehensive breakthrough, spanning from the developer
ecosystem to a global MaaS (Model-as-a-Service) platform.
Developer Ecosystem and Coding Plan: Returning from “Features” to “Engineering Value”
Developers are the most sensitive group in perceiving the upper bound of intelligence. The
GLM Coding Plan, launched in 2025, rapidly expanded globally by virtue of its native high-
quality engineering reasoning capabilities, reaching over 242,000 paying developers. Bolstered
by our technological leadership, we proactively raised prices by 30% and eliminated first-
purchase discounts in February 2026. This trend of “rising volume and price” demonstrates
that the market is shifting from a blind price war to genuine recognition of “technological
value”, namely, trading higher-order intelligence for more certain productivity.
MaaS Platform Expanding Horizontally: Building the Digital Infrastructure for the Intelligent
The Rise of Token Architecture
- The BigModel.cn platform has become a central hub for industrial AI, with 9 of China's top 10 internet companies integrating GLM-5 within 24 hours of release.
- Market demand for high-order intelligence has led to 'computing power panic,' allowing for significant price increases while demand still outstrips supply.
- The 'Claw Plan' and AutoGLM are setting global standards for Agentic AI, shifting the business model from software export to providing sovereign AI infrastructure.
- Future intelligence will evolve from simple coding to industrial-grade Agentic Engineering capable of autonomous planning and self-iteration.
- Core competitiveness is being redefined as Token Architecture Capability (TAC), measuring the efficiency of converting intelligence calls into economic value.
- The paradigm is shifting from traditional operating systems to LLM-OS, where models act as schedulers of intelligence rather than just hardware.
This confirms that high-order intelligence is a scarce resource today, and whoever controls the upper bound holds the pricing power.
Era Relying on BigModel.cn, our MaaS platform has become the hub connecting foundational
models with industrial applications. Within 24 hours of its release, GLM-5 was officially
integrated by leading companies such as ByteDance, Alibaba, and Tencent, and 9 of the top
10 Internet companies in China have now deeply integrated GLM. The platform has surpassed
4 million registered users as of March 2026, and despite an 83% increase in API call pricing
compared to the end of last year, the market still exhibits “computing power panic” with
demand outstripping supply. This confirms that high-order intelligence is a scarce resource
today, and whoever controls the upper bound holds the pricing power.
Agent Matrix and Global Expansion: Defining a New Paradigm for “Going Global”
From AutoGLM, the world’s first mobile phone agent, to AutoClaw, China’s first one-
click installation agent, we are defining the implementation standards for Agentic AI. The
Claw Plan, launched in March 2026, surpassed 100,000 subscribers within just two days and
400,000 subscribers within 20 days of going live. This validates the immense commercial
potential of long-chain agent tasks. On the global front, we are no longer merely “exporting
software” but are outputting national-level large model infrastructure through a “sovereign
AI” model and exploring partnerships with overseas computing power platforms through
revenue sharing. Currently, Knowledge Atlas’s business footprint has spanned 218 countries
and regions, truly realizing the global monetization of technological capabilities through token
value.
– 5 –
Future Outlook
Knowledge Atlas is not a traditional software company. We are a native intelligent laboratory
driven by the faith in AGI. Our moat lies not in the accumulation of computing power but in
the foundational deconstruction of the essence of intelligence and the perseverance to translate
this understanding into social productivity. Looking ahead to 2026, the intelligence paradigm
will evolve from lightweight Vibe Coding to industrial-grade Agentic Engineering, then
evolve into digital engineers with autonomous planning, environmental perception and self-
iteration capabilities, and finally achieve the closed-loop execution of Long-horizon Task with
logical consistency spanning multiple iterations, which will further bring about breakthroughs
in the upper bound of intelligence and exponential growth in Token calls.
Entering the TAC Era: Everyone as a “Token Architect”
In the era of large models, when large models possess the closed-loop capability for Long-
Horizon Task execution, core competitiveness will be reshaped as TAC (Token Architecture
Capability). TAC = volume of intelligence calls × quality of intelligence × economic
transformation efficiency. In the future, the standard for measuring the value of an individual
or organization will no longer be how much information is mastered, but its role as a Token
Architect in constructing complex Agent systems under a given budget and driving large
models to complete the autonomous operation of complex Agent systems. Knowledge Atlas
aims to become the infrastructure for enhancing the TAC of the whole society, allowing every
drop of Token to be converted into deliverable economic increments.
From “Conversational Interfaces” to “Large Language Model-Operating Systems” (LLM-OS)
The traditional OS is a scheduler of hardware resources, whereas LLM-OS is a scheduler
of intelligence. Large models are consuming software; future computing platforms will no
longer be stacks of apps but collaborations between API stores and agent matrices. Under the
LLM-OS architecture, models directly understand ambiguous intentions, decompose long-
The Intelligence Output Revolution
- The company aims to integrate the GLM model into system kernels to transition from cloud APIs to device-level native intelligence.
- An 'Intelligence Output Revolution' is shifting the focus toward high-quality token exports, leveraging China's industrial chain to move from 'Made in China' to 'Intelligence in China.'
- Total revenue saw a massive increase from RMB312.4 million in 2024 to RMB724.3 million in 2025, driven by enhanced model intelligence.
- Cloud-based deployment revenue experienced a surge of 292.6%, while on-premises deployment grew by 102.3%, reflecting strong market demand for versatile AI.
- The company is transitioning its financial reporting to a new classification system based on business form and product lines to better reflect its diverse AI ecosystem.
Token exports are not about low-price competition but about “high quality at competitive prices” based on top-tier intelligence levels like GLM-5.
term tasks, and schedule full-stack resources. Whoever integrates their model into the system
kernel will control the definition of next-generation computing. We are committed to making
GLM the core engine of this autonomous system, achieving a leap from cloud-based APIs to
device-level native intelligence.
Intelligence Output Revolution: A “Global Factory” for High-Quality Tokens
As token consumption, driven by applications like OpenClaw, enters an exponential trajectory,
an intelligence output revolution is underway. First, inference re-centralization: leveraging
the economies of scale of ultra-large-scale clusters and ultimate inference optimization,
the efficiency of cloud-based large-parameter foundation models will be further improved.
Second, high-quality token exports: relying on China’s full industrial chain advantages in
energy, chip-algorithm (Co-Design) adaptation, and IDC operations, we are achieving a leap
– 6 –
from “Made in China” to “Intelligence in China”. Token exports are not about low-price
competition but about “high quality at competitive prices” based on top-tier intelligence levels
like GLM-5. What we aim to supply globally are production factors representing the upper
bound of cognitive intelligence with ultimate cost-effectiveness.
Financial Review
The following discussion is based on the financial data and their accompanying notes set out
in other sections of this year’s performance announcement and should be read in conjunction
therewith.
ANALYSIS OF KEY OPERATING PERFORMANCE ITEMS
Revenue
As we commercialize our technologies to seize the immense market opportunities presented
by advanced artificial intelligence, we provide intelligent services to enterprise customers,
developers, and end-users in the most suitable, reasonable, and scalable manner. We offer
flexible deployment options to meet the diverse needs of enterprises while ensuring efficiency,
scalability, and data security. We primarily provide two service models: cloud-based
deployment services and on-premises deployment services. During the Reporting Period, we
achieved substantial revenue growth. Our revenues were RMB312.4 million and RMB724.3
million respectively for the full years of 2024 and 2025.
By deployment method:
Cloud-based Deployment
Revenue from cloud-based deployment increased from RMB48.5 million in 2024 to
RMB190.4 million in 2025, representing a surge of 292.6%. This robust growth was primarily
attributable to the Group’s continuous iterations, which significantly elevated the upper bound
of model intelligence. The enhanced model intelligence performance further drove an increase
in model invocations.
On-premise Deployment
Our revenue from on-premises deployment rose from RMB263.9 million in 2024 to
RMB534.0 million in 2025, marking a significant increase of 102.3%. This notable growth
was mainly driven by the Group’s continuous iterations that significantly elevated the upper
bound of model intelligence, coupled with enhanced model versatility and sustained strong
market demand.
– 7 –
2025
2024
RMB’000
% of Total
RMB’000
% of Total
Cloud-based deployment
190,379
26.3
48,484
15.5
On-premises deployment
533,955
73.7
263,930
84.5
Total
724,334
100.0
312,414
100.0
To more accurately reflect the Company’s strategic layout and the substance of its business
development, and to provide investors with more insightful analytical perspectives on our
business, we have optimized the method of revenue classification disclosure starting this year.
We believe that the original classification, primarily based on deployment methods, can no
longer comprehensively cover the Company’s increasingly diverse product forms and business
scenarios. Therefore, we have introduced a new classification that breaks down revenues by
business form and core product lines.
Open Platform and API
The Open Platform and API represent the standardized and platform-based cloud online
Enterprise AI Revenue Growth
- The Open Platform and API segment experienced explosive growth of 292.6%, driven by increased model invocations and enhanced intelligence capabilities.
- Enterprise-level agents now simulate human expert thought patterns to autonomously execute complex business loops and task planning.
- Revenue from autonomous agents grew by 248.8% as market demand surged for automation and auxiliary decision-making tools.
- Large-scale foundation models, including language and multimodal versions, remain the core revenue driver through private on-premises deployments.
- The company has expanded its model matrix to include specialized contemplation and reflection models for vertical industry scenarios.
These products integrate memory storage, task planning, and execution feedback mechanisms, enabling them to simulate the thought patterns of human experts.
services provided by the Company to developers and enterprise customers based on general-
purpose large model capabilities. The product transforms the complex algorithmic capabilities
of underlying large models into convenient cloud services through Application Programming
Interface (API), Software Development Kit (SDK), and Model-as-a-Service (MaaS) platform
services. Its revenue increased from RMB48.5 million in 2024 to RMB190.4 million in
2025, representing a growth rate of 292.6%. This robust growth was primarily attributed
to the elevation of the Group’s model intelligence upper bound and the increase in model
invocations.
Enterprise-level Agents
Enterprise-level agents refer to autonomous intelligent systems built for complex enterprise
scenarios, with general-purpose large models as the core control unit, combined with
enterprise knowledge bases and tool invocation capabilities. These products integrate memory
storage, task planning, and execution feedback mechanisms, enabling them to simulate
the thought patterns of human experts. They autonomously understand instructions, break
down tasks, and invoke external tools (APIs) to complete business loops, aiming to address
the needs for automation execution and auxiliary decision-making in enterprise business
processes, thereby comprehensively improving enterprise operational efficiency and digital
management levels. Revenue from enterprise-level agents increased from RMB47.5 million in
2024 to RMB165.7 million in 2025, marking a growth rate of 248.8%. This robust growth was
primarily attributable to the elevation of the Group’s model intelligence upper bound and the
surge in market demand for enterprise-level intelligent agent products.
– 8 –
Enterprise-level General-purpose Large Models
Enterprise-level general-purpose large models refer to a matrix of pre-trained models
independently developed by the Company, featuring large-scale parameters and strong
generalization capabilities, serving as the foundation for the Company’s core technological
innovations and business development. These products are primarily delivered through private
on-premises deployment, where the models and related software environments are directly
deployed on the customer’s own local computing infrastructure or private cloud environments.
Based on self-developed pre-trained foundation models with large-scale parameter sizes and
deep learning capabilities, the Company has constructed a core model system encompassing
language large models, code large models, multimodal large models, and contemplation and
reflection models. Through pre-training and instruction fine-tuning with massive amounts of
multi-source data, these models meet the AI application needs in various vertical scenarios. Its
revenue increased from RMB214.5 million in 2024 to RMB365.7 million in 2025, representing
a growth rate of 70.5%. This growth was primarily attributable to the elevation of the Group’s
model intelligence upper bound, coupled with sustained strong market demand.
Technical Services and Others
Technical Services and Others refer to the specialised technical services provided by the
Company to developers and enterprise clients in the AI field, based on the underlying
technologies of general-purpose large models. This business focuses on the commercialisation
and application of the latest technological achievements in the large model domain. Through
in-depth technical enablement, the Company offers professional technical consultancy on
large models, as well as certain resource matching, technical exchange and collaborative
innovation support. Revenue increased from RMB1.9 million in 2024 to RMB2.5 million in
2025, representing a year-on-year increase of 31.6%, primarily driven by increased customer
demand.
2025
2024
RMB’000
% of Total
RMB’000
% of Total
Open platform and API
190,379
26.3
48,484
15.5
Enterprise-level agents
165,687
22.9
47,492
15.2
Financial Growth and Margin Shifts
- The Group achieved a 68.7% increase in gross profit, rising to RMB296.7 million in 2025, despite a significant drop in overall gross profit margin from 56.3% to 41.0%.
- Cloud-based deployment saw explosive growth with a 2,150% increase in gross profit, driven by improved inference efficiency and economies of scale in computing capacity.
- On-premise deployment remains the largest profit contributor at RMB260.7 million, though its margin fell to 48.8% due to increased delivery resource investments.
- Enterprise-level general-purpose large models experienced a margin compression from 69.6% to 47.0% as the company faced higher phased delivery costs to meet diverse customer demands.
- Cost of sales surged by 213.3% to RMB427.7 million, primarily fueled by rising computing service fees necessary to support rapid business expansion.
- Technical services maintained the highest profitability with a 92.0% gross profit margin, reflecting the stable high value of technical enablement services.
The gross profit of the cloud-based deployment business increased from RMB1.6 million in 2024 to RMB36.0 million in 2025, representing a growth rate of 2,150.0%.
Enterprise-level general-purpose
large models
365,724
50.4
214,502
68.7
Technical service and others
2,544
0.4
1,936
0.6
Total
724,334
100.0
312,414
100.0
– 9 –
Gross Profit and Gross Profit Margin
The Group’s gross profit increased from RMB175.9 million in 2024 to RMB296.7 million
in 2025, representing a growth rate of 68.7%. The Group’s gross profit margin decreased
from 56.3% in 2024 to 41.0% in 2025, primarily due to an increase in the proportion of
cloud deployment business and a temporary decline in the gross profit margin of on-premise
deployment business.
By deployment method:
Cloud-based Deployment
The gross profit of the cloud-based deployment business increased from RMB1.6 million in
2024 to RMB36.0 million in 2025, representing a growth rate of 2,150.0%, mainly due to
diversification and expansion of cloud-based deployment business. The gross profit margin
of the cloud-based deployment business increased from 3.3% in 2024 to 18.9% in 2025,
primarily due to improved model inference efficiency, economies of scale from expanded
computing capacity leading to declining marginal costs, and price increases.
On-premise Deployment
The gross profit of the on-premises deployment business increased from RMB174.3 million
in 2024 to RMB260.7 million in 2025, representing a growth rate of 49.6%, mainly driven by
revenue growth resulting from market demand expansion. The gross profit margin of the on-
premises deployment business decreased from 66.0% in 2024 to 48.8% in 2025, primarily due
to the investment of more delivery resources to meet customer needs.
By business form and core product lines:
Open Platform and API
Our gross profit of the open platform and API business increased from RMB1.6 million in
2024 to RMB36.0 million in 2025, representing a year-on-year growth of 2,150.0%. The gross
profit margin increased from 3.3% in 2024 to 18.9% in 2025. Such increases in both gross
profit and gross profit margin are primarily driven by expansion of cloud-based deployment
business, the launch of programming subscription packages, and enhanced inference
efficiency.
– 10 –
Enterprise-level Agents
Our gross profit of the enterprise-level agent business increased from RMB23.4 million in
2024 to RMB86.7 million in 2025, representing a year-on-year growth of 270.5%. The gross
profit margin slightly increased from 49.3% in 2024 to 52.3% in 2025, primarily due to the
increase in revenue from relevant businesses, which resulted in relatively stable gross profit
margins.
Enterprise-level General-purpose Large Models
Our gross profit of the enterprise-level general-purpose large model business increased from
RMB149.2 million in 2024 to RMB171.7 million in 2025, representing a growth rate of
15.1% year on year, primarily driven by the amplification of gross profit alongside significant
growth in overall revenue for the business. The gross profit margin decreased from 69.6% in
2024 to 47.0% in 2025, primarily due to a phased increase in delivery costs in line with the
diversification of customer demands.
Technical Services and Others
Gross profit from the technical Services and others increased from RMB1.7 million in 2024
to RMB2.3 million in 2025, representing a year-on-year increase of 35.3%, while gross profit
margin rose from 89.5% in 2024 to 92.0% in 2025, primarily attributable to the stable high
gross profit generated by technical enablement services.
Cost of Sales
The Group’s cost of sales increased from RMB136.5 million in 2024 to RMB427.7 million
in 2025, representing a growth rate of 213.3%. This growth is consistent with the Group’s
business expansion and revenue growth. Specifically, the increase in costs was primarily
attributable to the rise in computing service fees resulting from business expansion and
revenue growth.
Capital Expenditure
The Group’s capital expenditure in 2025 was approximately RMB74.7 million, representing
Financial Performance and Strategic Shifts
- Capital expenditure dropped by 83.8% as the company pivoted from leasing equipment to a service-based procurement model for computing power.
- Research and development expenses surged by 44.9% to RMB3,180.4 million, driven by R&D team expansion and investments in foundational model iteration.
- General and administrative expenses saw a massive 278.3% increase, primarily due to aggressive administrative hiring and share-based payment costs.
- Finance costs nearly doubled, rising from RMB38.3 million to RMB74.3 million, largely due to a shift from exchange gains to net exchange losses.
- The carrying amount of financial instruments issued to investors doubled to RMB937.4 million following new equity financing with redemption rights.
This significant change was mainly due to adjustments in the computing power leasing strategy: We primarily used leases to acquire computing power equipment in 2024, with related expenditures included in capital expenditure.
a decrease of approximately 83.8% compared to RMB462.3 million in 2024. This significant
change was mainly due to adjustments in the computing power leasing strategy: We primarily
used leases to acquire computing power equipment in 2024, with related expenditures
included in capital expenditure; with the expansion of business scale and rapid growth in
computing power demand, we flexibly adjusted our computing power procurement model in
2025, primarily relying on service procurement supplemented by partial equipment leasing to
meet demand, resulting in a corresponding decrease in capital expenditure.
– 11 –
Other Income
The Group’s other income decreased from RMB19.3 million in 2024 to RMB14.8 million in
2025.
Research and Development Expenses
The Group’s research and development expenses increased from RMB2,195.4 million in
2024 to RMB3,180.4 million in 2025, representing a growth rate of 44.9%. This growth was
primarily attributable to (i) an increase in staff costs, mainly due to the expansion of the
Group’s R&D team and an increase in share-based payment expenses; and (ii) an increase
in computing service fees paid to third-party computing power suppliers, mainly due to the
Group’s significant efforts in iterating foundational models and investing in more advanced
model training infrastructure.
Sales and Distribution Expenses
The Group’s selling and marketing expenses increased from RMB387.5 million in 2024 to
RMB390.9 million in 2025, representing a growth rate of 0.9%. This growth was primarily
attributable to (i) an increase in staff costs, mainly due to the expansion of the Group’s sales
and marketing team and an increase in share-based payment expenses; and (ii) a decrease in
advertising and marketing expenses as well as technical service and consulting fees, mainly
due to the Company’s adjustment of market promotion strategies and optimization of resource
allocation.
General and Administrative Expenses
Our general and administrative expenses increased from RMB133.6 million in 2024
to RMB505.4 million in 2025, representing a growth rate of 278.3%. This growth was
primarily attributable to an increase in staff costs, mainly due to the recruitment of additional
administrative personnel and an increase in share-based payment expenses.
Impairment Loss on Financial Assets
The Group’s impairment loss on financial assets increased from RMB17.0 million in 2024 to
RMB21.6 million in 2025, representing a growth rate of 27.1%. This increase was primarily
due to the increase in the original value of financial assets resulting from business expansion.
Finance Costs
The Group’s finance costs increased from RMB38.3 million in 2024 to RMB74.3 million in
2025, primarily due to a shift from net exchange gains in 2024 to net exchange losses in 2025.
– 12 –
Share of Profit or Loss of Associates
The Group’s share of profit or loss of associates increased from RMB21.3 million in 2024 to
RMB55.0 million in 2025, representing a growth rate of 158.2%. This increase was primarily
due to improvements in the operating and financial performance of the associates during the
Reporting Period.
Changes in Fair Value of Financial Instruments Measured at FVPL
The changes in fair value of financial assets measured at fair value through profit or loss
decreased from RMB66.3 million in 2024 to RMB25.3 million in 2025. This decrease was
primarily due to adjustments in our investment portfolio allocation in 2025, resulting in a
decrease in recognized fair value gains.
Changes in the Carrying Amount of Financial Instruments Issued to Investors
The carrying amount of financial instruments issued to investors increased from RMB468.9
million in 2024 to RMB937.4 million in 2025. This increase was primarily due to new equity
financing with redemption rights in 2025.
Net Loss
2025 Financial Performance Analysis
- The Group reported an increased net loss of RMB4,718.2 million in 2025, primarily driven by heavy investment in research and development.
- Adjusted net loss under non-IFRS measures rose to RMB3,181.97 million after accounting for share-based compensation and financial instrument adjustments.
- Total borrowings surged from RMB137.2 million to RMB689.5 million to cover operational working capital needs.
- Shareholders' equity dropped significantly to a negative RMB8,111.0 million, reflecting the impact of sustained strategic R&D spending.
- The company maintains a workforce of 1,094 permanent employees with total remuneration costs reaching RMB1,363.4 million.
- Despite operational losses, the Group maintained a stable cash position of approximately RMB2.26 billion with no significant capital commitments or contingent liabilities.
The Company regards talent as its core asset and is committed to building a market-leading, incentive-oriented, and comprehensive talent development system to support the Company’s continuous innovation and business growth in the field of artificial intelligence.
The Group recorded a net loss of RMB4,718.2 million in 2025, representing an increase from
RMB2,958.0 million in 2024, mainly affected by the continuous increase in research and
development investment.
Non-IFRS Measure
2025
2024
RMB’000
RMB’000
Loss for the Year
(4,718,167)
(2,958,007)
Add:
– Equity-settled share-based compensation expenses
558,265
23,579
– Changes in the carrying amount of financial
instruments issued to investors
937,362
468,859
– Listing expense
40,568
–
Adjusted net loss for the year (non-IFRS measure)
(3,181,972)
(2,465,569)
– 13 –
ANALYSIS OF KEY ITEMS IN FINANCIAL POSITION
Financial Position
Shareholders’ equity decreased from RMB-3,955.1 million as of 31 December 2024, to
RMB-8,111.0 million as of 31 December 2025, primarily due to the expansion of net losses
resulting from increased research and development investment.
Liquidity and Financial Resources
We closely monitor and maintain adequate liquidity levels to meet working capital
requirements and mitigate the impact of cash flow fluctuations. As of 31 December 2025, the
Group held cash and cash equivalents of RMB2,259.1 million, a decrease of RMB10.1 million
from RMB2,269.2 million as of 31 December 2024. This fluctuation was mainly attributable
to consumption in daily operations.
Debt/Borrowings
As of 31 December 2025, the Group’s total borrowings amounted to RMB689.5 million, an
increase of RMB552.2 million from RMB137.2 million as of 31 December 2024, primarily
due to the borrowing of working capital to meet operational needs. All relevant borrowings
are interest-bearing, denominated primarily in RMB, and unsecured.
Lease Liabilities
As of 31 December 2025, the total lease liabilities recognized by the Group (including both
current and non-current portions) amounted to RMB545.1 million, a decrease of RMB126.1
million from RMB671.3 million as of 31 December 2024. This fluctuation was influenced by
the payment of lease obligations during the year and changes in office and data center lease
arrangements.
Gearing Ratio
The Group’s gearing ratio (Gearing ratio is calculated by dividing total interest-bearing
bank and other borrowings and lease liabilities divided by total equity as of the end of the
period multiplied by 100%) was 15.2% as of 31 December 2025, compared to 20.4% as of
31 December 2024. This change was primarily attributable to the impact of the Group’s
continued strategic research and development investment on total equity.
Asset Pledges
The Group had no pledged assets for the year ended 31 December 2025.
– 14 –
Capital Commitments
The Group had no outstanding capital commitment for the year ended 31 December 2025.
Contingent Liabilities
The Group had no significant contingent liabilities for the year ended 31 December 2025.
Significant Investments and Major Acquisitions and Disposals of Subsidiaries, Associates,
and Joint Ventures
For the year ended 31 December 2025, the Group did not make any significant investments or
engage in any major acquisitions or disposals of subsidiaries, associates, or joint ventures.
Future Plans for Significant Investments or Capital Assets
The Group had not made any significant investments or have any other future plans for
significant investments or the purchase of capital assets for the year ended 31 December 2025.
Employees and Remuneration Policy
As of 31 December 2025, the Company had 1,094 permanent employees. The Company
incurred total remuneration costs (including share-based payments) of RMB1,363.4 million
during the Reporting Period. The Company regards talent as its core asset and is committed to
building a market-leading, incentive-oriented, and comprehensive talent development system
to support the Company’s continuous innovation and business growth in the field of artificial
intelligence.
We adhere to the principles of value orientation and market competition in constructing
Employee Incentives and Risk Factors
- The Group utilizes a multi-tiered remuneration structure including fixed salary, performance bonuses, and a shareholding platform to align employee interests with long-term growth.
- Comprehensive welfare programs include supplementary insurance, health check-ups, and interest clubs to foster a harmonious work environment and personal growth.
- Employees are encouraged to adopt an AI-native work style by using the AutoClaw tool to automate standardized tasks and focus on high-level innovation.
- Foreign exchange risks are primarily linked to RMB/USD fluctuations, though the company currently does not engage in formal hedging activities.
- The company identifies significant operational risks, particularly the necessity of constant technological innovation in the rapidly evolving AI industry.
- Substantial ongoing investments in research and development are required to maintain a competitive edge and ensure future commercial success.
We encourage employees to shift from seeking suggestions from large models to using them to complete tasks, such as having AutoClaw take over most standardized execution work.
a competitive remuneration structure. The remuneration package consists of fixed salary,
performance bonuses, and medium- to long-term incentives, ensuring that employee rewards
are closely linked to company performance and individual contributions. We adopted the
employee incentive scheme and established an employee shareholding platform to incentivize
eligible middle and senior management, core technical and business personnel and other
employees, consultants and other persons who are vital to the development of the Group, to
align the interests of the Group with those of the participants of the incentive scheme, so as
to promote the long-term growth of the Group. We regularly conduct market salary surveys to
ensure that our salary levels remain competitive within the industry, attracting and retaining
top technical and management talent and stimulating organizational vitality.
– 15 –
The Company is committed to enhancing employee well-being and establishing a
comprehensive welfare protection network. In addition to strictly complying with national
regulations in paying social insurance and housing provident funds, the Company provides
supplementary commercial insurance, annual health check-ups, work meal subsidies,
and various festival bonuses, among other diversified benefits. Meanwhile, the Company
supports the establishment of various interest clubs, regularly organizes cultural and sports
activities, pays attention to employees’ physical and mental health, creates a harmonious and
positive work atmosphere, and has established a comprehensive internal training system and
knowledge-sharing mechanism to encourage cross-departmental communication and technical
sharing and focus on employees’ personal growth.
As a leading LLM company, we actively promote AI-native work methods and encourage all
employees to deeply integrate the use of AutoClaw into their daily work and life scenarios.
We encourage employees to shift from seeking suggestions from large models to using them
to complete tasks, such as having AutoClaw take over most standardized execution work,
allowing individuals to focus their energy on defining key problems, exploring cutting-edge
technologies, and unlocking innovative thinking.
Exposure to Exchange Rate Fluctuation
The Group’s foreign exchange risk arises from future commercial transactions and recognized
assets and liabilities denominated in currencies other than the functional currency of the
relevant Group entities. Our business is primarily conducted in RMB. Most non-RMB assets
and liabilities are cash and cash equivalents denominated in US dollars.
We are primarily exposed to changes in RMB/USD exchange rates in our domestic subsidiaries
whose functional currency is RMB. We currently do not engage in hedging activities aimed
at or intended for managing foreign exchange rate risks. However, we continuously monitor
currency exchange rate movements and will take necessary measures to mitigate the impact of
exchange rates.
– 16 –
Principal Risks and Uncertainties
Our operations and Global Offering involve certain risks and uncertainties, including (i) risks
related to our research and development; (ii) risks related to our commercialization; (iii)
risks related to our operations; (iv) risks related to our intellectual property; (v) risks related
to our financial position and additional funding requirements; and (vi) risks related to the
jurisdictions in which we conduct business, including but not limited to:
•
The AI industry is characterized by constant changes. If we are not able to upgrade,
enhance or innovate our technologies and services, our business, results of operations,
financial condition and prospects could be adversely affected.
•
We have made and expect to continue to make substantial investments in R&D. If
we cannot continuously invest in our R&D activities while achieving technological
AI Risk Factors and Financials
- The company faces significant uncertainties regarding the future realization of Artificial General Intelligence (AGI) and the performance of its proprietary models.
- Operational risks include heavy reliance on third-party computing resources and the potential for increasingly stringent regulatory environments in China.
- Financial statements reveal a substantial increase in revenue from 312 million to 724 million RMB, yet losses have widened significantly.
- Research and development expenses represent the largest cost driver, exceeding 3.1 billion RMB in 2025 as the company scales its technology.
- The company reports a total comprehensive loss of approximately 4.7 billion RMB for 2025, reflecting a limited track record in commercialization.
- Intellectual property protection and intense competition from current and future AI players pose ongoing threats to market share.
The development of AGI is still at an early stage and there are substantial uncertainties in the future realization of AGI.
innovation, our business, results of operations, financial condition and prospects may be
materially and adversely affected.
•
The development of AGI is still at an early stage and there are substantial uncertainties
in the future realization of AGI.
•
We are exposed to risks relating to our R&D team and our senior management.
•
We rely on third parties to provide computing resources to us, and any disruption of their
services or fluctuation of prices could adversely affect our business, results of operations
and financial condition.
•
Our commercial success depends on the performance of our models. Any failure in
research and development efforts to offer high-quality models and solutions could harm
our business, results of operations, financial condition and prospects.
•
China’s AI industry is subject to an evolving and increasingly stringent regulatory
environment. Future laws and regulations may impose additional requirements and other
obligations that could materially and adversely affect our business, results of operations,
financial condition and prospects.
•
We may not be able to obtain or maintain adequate intellectual property rights protection
for our business, or the scope of such intellectual property rights protection may not be
sufficiently broad.
•
We may not be able to compete effectively against current or future competitors.
– 17 –
•
The size of our addressable market and the demand for our solutions may not increase
as rapidly as we anticipate due to a variety of factors, which would materially and
adversely affect our business, results of operations, financial condition and prospects.
•
Any failure of our MaaS platform to perform as required could harm our business,
results of operations, financial condition and prospects.
•
We have a limited track record in the commercialization of our business.
The above list is not exhaustive. For further details, please refer to the “Risk Factors” section
of our Prospectus.
– 18 –
CONSOLIDATED STATEMENT OF PROFIT OR LOSS AND OTHER COMPREHENSIVE
INCOME
For the year ended 31 December 2025
(Expressed in Renminbi (“RMB”))
2025
2024
Note
RMB’000
RMB’000
Revenue
5
724,334
312,414
Cost of revenue
(427,678)
(136,525)
Gross profit
296,656
175,889
Other income
14,839
19,281
Selling and marketing expenses
(390,869)
(387,475)
General and administration expenses
(505,357)
(133,603)
Research and development expenses
(3,180,443)
(2,195,436)
Impairment losses on financial assets
(21,576)
(17,008)
Loss from operations
(3,786,750)
(2,538,352)
Finance costs
(74,315)
(38,321)
Share of profits less losses of associates
54,982
21,254
Changes in fair value of financial instruments
measured at fair value through profit or loss
(“FVPL”)
25,278
66,271
Changes in the carrying amounts of financial
instruments issued to investors
(937,362)
(468,859)
Loss before taxation
(4,718,167)
(2,958,007)
Income tax
6
–
–
Loss for the year
(4,718,167)
(2,958,007)
– 19 –
2025
2024
Note
RMB’000
RMB’000
Other comprehensive income for the year
(after tax):
Items that may be reclassified to profit or loss:
– Exchange differences on translation of
financial statements into presentation
currency
278
(79)
Total comprehensive income for the year
(4,717,889)
(2,958,086)
Loss attributable to:
Equity shareholders of the Company
(4,698,203)
(2,956,491)
Non-controlling interests
(19,964)
(1,516)
(4,718,167)
(2,958,007)
Total comprehensive income attributable to:
Equity shareholders of the Company
(4,697,925)
(2,956,570)
Non-controlling interests
(19,964)
(1,516)
(4,717,889)
(2,958,086)
Loss per share
Basic and diluted (RMB)
7
(12.03)
(8.72)
– 20 –
CONSOLIDATED STATEMENT OF FINANCIAL POSITION
At 31 December 2025
(Expressed in RMB)
31 December
2025
31 December
2024
Note
RMB’000
RMB’000
Non-current assets
Property and equipment
655,825
866,363
Intangible assets
50,963
50,359
Knowledge Atlas Financial Position
- The financial statement reveals a significant net liability position, with total equity deficit increasing from approximately RMB 3.96 billion in 2024 to RMB 8.11 billion in 2025.
- A major driver of the company's liabilities is 'Financial instruments issued to investors,' which surged to over RMB 10 billion by the end of 2025.
- The company underwent a significant corporate restructuring, converting from a limited liability company to a joint stock company in March 2025 before listing in Hong Kong in early 2026.
- Despite the heavy debt load, the group maintains substantial cash reserves and bank balances exceeding RMB 2.2 billion.
- The group's primary business focus is the provision of large model-related services within the People's Republic of China.
The Company and its subsidiaries (together, the “Group”) are principally engaged in the provision of large model-related services in the PRC.
Goodwill
39,379
39,379
Interests in associates
338,081
201,198
Other non-current assets
198,202
97,260
Time deposits
–
105,343
1,282,450
1,359,902
Current assets
Short-term investments measured at FVPL
377,287
42,621
Inventories and contract costs
125,817
32,465
Trade and other receivables
8
698,942
666,841
Contract assets
1,625
4,718
Time deposits
108,593
–
Cash at bank and on hand
2,259,147
2,269,222
3,571,411
3,015,867
Current liabilities
Trade and other payables
9
1,328,631
603,488
Contract liabilities
148,644
75,059
Bank loans
604,836
137,246
Lease liabilities
251,316
213,161
Financial instruments issued to investors
10,072,827
6,676,943
Convertible bonds
–
132,158
12,406,254
7,838,055
Net current liabilities
(8,834,843)
(4,822,188)
– 21 –
31 December
2025
31 December
2024
Note
RMB’000
RMB’000
Total assets less current liabilities
(7,552,393)
(3,462,286)
Non-current liabilities
Bank loans
84,616
–
Lease liabilities
293,827
458,107
Deferred income
180,146
34,752
558,589
492,859
NET LIABILITIES
(8,110,982)
(3,955,145)
CAPITAL AND RESERVES
Share capital/paid-in capital
40,281
36,224
Reserves
(8,132,783)
(3,992,853)
Total equity – deficit attributable to equity
shareholders of the Company
(8,092,502)
(3,956,629)
Non-controlling interests
(18,480)
1,484
TOTAL EQUITY – DEFICIT
(8,110,982)
(3,955,145)
– 22 –
1.
CORPORATE INFORMATION
Knowledge Atlas Technology Joint Stock Company Limited (the “Company”, formerly known as Beijing
Zhipu Huazhang Technology Limited) was established in the People’s Republic of China (the “PRC”)
on 11 June 2019 as a limited liability company On 26 March 2025, the Company was converted from a
limited liability company into a joint stock limited liability company and changed its registered name to
Knowledge Atlas Technology Joint Stock Company Limited. The Company’s shares were listed on the
Main Board of The Stock Exchange of Hong Kong Limited (the “Stock Exchange”) on 8 January 2026.
The Company and its subsidiaries (together, the “Group”) are principally engaged in the provision of
large model-related services in the PRC.
2.
STATEMENT OF COMPLIANCE
These financial statements have been prepared in accordance with IFRS Accounting Standards as issued
by the International Accounting Standards Board (the “IASB”) and the applicable disclosure requirements
of the Hong Kong Companies Ordinance. These financial statements also comply with the applicable
disclosure provisions of the Rules Governing the Listing of Securities on the Stock Exchange.
The IASB has issued certain amendments to IFRS Accounting Standards that are first effective or
available for early adoption for the current accounting period of the Group. Note 4 provides information
on any changes in accounting policies resulting from initial application of these developments to the extent
that they are relevant to the Group for the current accounting period reflected in these financial statements.
3.
BASIS OF PREPARATION OF THE FINANCIAL STATEMENTS
2025 Financial Performance and Liquidity
- The Group reported a significant net loss of approximately RMB 4.7 billion for the year ended 31 December 2025.
- Despite recording net liabilities of over RMB 8.1 billion, the directors maintain a 'going concern' status due to a successful stock exchange listing in January 2026.
- Revenue more than doubled year-over-year, growing from RMB 312 million in 2024 to over RMB 724 million in 2025.
- The business model is shifting toward large model-related services, with on-premise deployment remaining the primary revenue driver over cloud-based solutions.
- The Group's customer base has become more diversified, with no single customer accounting for more than 10% of total revenue in 2025.
- Financial statements are prepared primarily on a historical cost basis, with specific exceptions for investments and convertible bonds measured at fair value.
The financial instruments issued to investors have been converted into equity upon the listing of the Company’s shares on the Stock Exchange on 8 January 2026.
The consolidated financial statements for the year ended 31 December 2025 comprise the Group and the
Group’s interests in associates.
The measurement basis used in the preparation of the financial statements is the historical cost basis,
except for investments and convertible bonds which are measured at their fair values respectively.
The preparation of the financial statements in conformity with IFRS Accounting Standards requires
management to make judgements, estimates and assumptions that affect the application of policies and
reported amounts of assets, liabilities, income and expenses. The estimates and associated assumptions
are based on historical experience and various other factors that are believed to be reasonable under the
circumstances, the results of which form the basis of making the judgements about carrying values of
assets and liabilities that are not readily apparent from other sources. Actual results may differ from these
estimates.
The estimates and underlying assumptions are reviewed on an ongoing basis. Revisions to accounting
estimates are recognised in the period in which the estimate is revised if the revision affects only that
period, or in the period of the revision and future periods if the revision affects both current and future
periods.
– 23 –
For the year ended 31 December 2025, the Group incurred net loss of RMB4,718,167,000 and as
at 31 December 2025, the Group recorded net liabilities of RMB8,110,982,000 and net current
liabilities of RMB8,834,843,000, which included financial instruments issued to investors amounted to
RMB10,072,827,000. The financial instruments issued to investors have been converted into equity upon
the listing of the Company’s shares on the Stock Exchange on 8 January 2026. Taking the above into
consideration, together with the gross proceeds from the listing of the Company’s shares on the Stock
Exchange of approximately Hong Kong dollars (“HK$”) 5,000,365,000 (equivalent to approximately
RMB4,516,430,000), the cashflow forecast for the year ending 31 December 2026 prepared by
management of the Group, and available unutilised banking facilities of RMB5,235,909,000 as at 31
December 2025 can be utilised by the Group to fulfil its liquidity requirements when necessary, the
directors of the Company consider that the Group will have sufficient financial resources to continue as a
going concern for the next twelve months. Therefore, the directors of the Company are satisfied that it is
appropriate to prepare the financial statements on a going concern basis.
4.
CHANGES IN ACCOUNTING POLICIES
The IASB has issued a number of new and revised IFRS Accounting Standards, none of these
developments have had a material effect on how the Group’s results and financial position for the
current or prior periods have been prepared or presented. The Group has not applied any new standard or
interpretation that is not yet effective for the current accounting period.
5
REVENUE AND SEGMENT REPORTING
(a)
Revenue
The principal activities of the Group are the provision of large model-related services in the PRC.
(i)
Disaggregation of revenue
Disaggregation of revenue from contracts with customers by major service types and timing
of revenue recognition are as follows:
2025
2024
RMB’000
RMB’000
Revenue from contracts with customers
within the scope of IFRS 15
Disaggregated by major service types:
– On-premise deployment
533,955
263,930
– Cloud-based deployment
190,379
48,484
724,334
312,414
– 24 –
2025
2024
RMB’000
RMB’000
Timing of revenue recognition
At a point in time
667,977
250,521
Over time
56,357
61,893
724,334
312,414
The Group’s customers base is diversified. There is no individual customer contributing over
10% of the total revenue of the Group for the year ended 31 December 2025 (one customer
in 2024 amounted to approximately RMB59,465,000).
(ii)
Revenue expected to be recognised in the future arising from contracts with customers in
Segment Reporting and Financial Performance
- The Group utilizes an IFRS 15 practical expedient to omit transaction price disclosures for remaining performance obligations due to short contract durations of one year or less.
- Business operations are divided into two primary reportable segments: On-premise deployment and Cloud-based deployment of large model-related services.
- Segment performance is evaluated based on gross profit, with On-premise deployment generating significantly higher revenue and profit than Cloud-based services.
- Operating expenses such as R&D, marketing, and administration are managed centrally and are not allocated to specific business segments.
- Despite a growth in total segment gross profit, the Group reported a substantial consolidated loss before taxation of over RMB 4.7 billion in 2025.
- The Group's operations and assets are almost entirely concentrated within the People's Republic of China, precluding the need for geographical segment analysis.
Assistance provided by one segment to another, including sharing of assets and technical know-how, is not measured.
existence at the reporting date
The Group applies the practical expedient in paragraph 121(a) of IFRS 15 of not disclosing
the transaction price allocated to the remaining performance obligation as the original
expected duration of the Group’s contracts are one year or less.
(b)
Segment reporting
The Group manages its businesses by service types. In a manner consistent with the way in which
information is reported internally to the Group’s most senior executive management for the
purposes of resource allocation and performance assessment, the Group has presented the following
two reportable segments. No operating segments have been aggregated to form the following
reportable segments.
•
On-premise deployment:
this segment develops and provides customised large
model-related services according to the customers’ specific
instructions and needs at the customers’ infrastructure.
•
Cloud-based deployment:
this segment develops and provides cloud-based large model-
related services to customers through cloud infrastructure.
– 25 –
(i)
Segment results
For the purposes of assessing segment performance and allocating resources, the Group’s
most senior executive management monitors the results attributable to each reportable
segment on the following bases:
Revenue and expenses are allocated to the reportable segments with reference to revenue
generated by those segments and direct expenses incurred by those segments. The measure
used for reporting segment result is gross profit. No inter-segment sales have occurred during
the year. Assistance provided by one segment to another, including sharing of assets and
technical know-how, is not measured.
The Group’s other operating income and expenses, such as other income, selling and
marketing expenses, general and administrative expenses, research and development
expenses, impairment losses on financial assets, finance costs, and assets and liabilities are
not measured under individual segments. Accordingly, neither information on segment assets
and liabilities nor information concerning capital expenditure, interest income and interest
expenses is presented.
Information regarding the Group’s reportable segments as provided to the Group’s most
senior executive management for the purposes of resource allocation and assessment of
segment performance during the year is set out below.
2025
On-premise
deployment
Cloud-based
deployment
Total
RMB’000
RMB’000
RMB’000
Segment revenue derived from
external customers
533,955
190,379
724,334
Segment gross profit
260,656
36,000
296,656
2024
On-premise
deployment
Cloud-based
deployment
Total
RMB’000
RMB’000
RMB’000
Segment revenue derived from
external customers
263,930
48,484
312,414
Segment gross profit
174,256
1,633
175,889
– 26 –
(ii)
Reconciliation of reportable segment profit or loss
2025
2024
RMB’000
RMB’000
Reportable segment gross profit
296,656
175,889
Other income
14,839
19,281
Selling and marketing expenses
(390,869)
(387,475)
General and administration expenses
(505,357)
(133,603)
Research and development expenses
(3,180,443)
(2,195,436)
Impairment losses on financial assets
(21,576)
(17,008)
Finance costs
(74,315)
(38,321)
Share of profits less losses of associates
54,982
21,254
Changes in fair value of financial instruments
measured at FVPL
25,278
66,271
Changes in the carrying amounts of financial
instruments issued to investors
(937,362)
(468,859)
Consolidated loss before taxation
(4,718,167)
(2,958,007)
(iii)
Geographic information
The Group’s revenue is substantially derived from customers located in the PRC, and all
of the Group’s non-current assets are located or allocated to operations located in the PRC
Accordingly, no segment analysis based on geographical locations is provided.
– 27 –
6
INCOME TAX IN THE CONSOLIDATED STATEMENT OF PROFIT OR LOSS AND OTHER
COMPREHENSIVE INCOME
Tax Reconciliation and Loss Per Share
- The group reported a significant increase in loss before taxation, rising from approximately RMB 2.96 billion in 2024 to RMB 4.72 billion in 2025.
- Tax calculations reflect a standard 25% rate in Mainland China, though certain subsidiaries benefit from a preferential 15% rate as High and New Technology Enterprises.
- Research and development incentives allow for an additional 100% deduction of qualified expenses from taxable income, providing a notable tax benefit.
- Non-deductible expenses, primarily stemming from financial instruments issued to investors and share-based payments, increased significantly in 2025.
- The company underwent a structural conversion into a joint stock company in March 2025, necessitating a retrospective adjustment of share counts for loss per share calculations.
- Basic loss per share worsened from RMB 8.72 in 2024 to RMB 12.03 in 2025, driven by higher losses attributable to ordinary equity shareholders.
For the purpose of computing basic and diluted loss per share, the weighted average number of ordinary shares deemed to be in issue before the Company’s conversion into a joint stock company was determined assuming the conversion into joint stock company had occurred since 1 January 2024.
Reconciliation between income tax expense and accounting loss at applicable tax rates:
2025
2024
RMB’000
RMB’000
Loss before taxation
(4,718,167)
(2,958,007)
Tax on loss before taxation, calculated at the rates applicable to
profits in the jurisdictions concerned (i)
(1,179,542)
(739,502)
Tax rates differentials (ii)
445,934
290,253
Tax effect of additional deduction on research and development
expenses (iii)
(139,429)
(305,408)
Tax effect of non-deductible expenses (iv)
246,469
83,782
Tax effect of unrecognised unused tax losses and deductible
temporary differences
626,568
670,875
–
–
Notes:
(i)
Entities of the Group established in the Chinese Mainland were subject to the PRC Corporate
Income Tax rate of 25% during the year (2024: 25%).
Taxation for subsidiaries incorporated in other jurisdictions is calculated at the applicable income
tax rates in the relevant jurisdictions.
(ii)
Certain subsidiaries of the Group obtained the certificates of “High and New Technology
Enterprise” (“HNTE”) from the tax authorities and were subject to a preferential tax rate of 15%
during the years ended 31 December 2025 and 2024.
(iii)
An additional 100% of qualified research and development expenses incurred is allowed to be
deducted from taxable income under the PRC Corporate Income Tax laws and regulations during
the years ended 31 December 2025 and 2024.
(iv)
Tax effect of non-deductible expenses mainly represented the changes in the carrying amounts
of financial instruments issued to investors and share-based payments expenses, which are not
deductible in accordance with the relevant tax regulations in the PRC.
– 28 –
7
LOSS PER SHARE
Basic loss per share is calculated by dividing the loss attributable to ordinary equity shareholders of the
Company by the weighted average number of ordinary shares in issue or deemed to be in issue during the
year.
The Company was converted into a joint stock company with limited liability on 26 March 2025. The
Company’s paid-in capital of RMB36,224,375 was converted into 36,224,375 shares of RMB1.00 each
accordingly. For the purpose of computing basic and diluted loss per share, the weighted average number
of ordinary shares deemed to be in issue before the Company’s conversion into a joint stock company was
determined assuming the conversion into joint stock company had occurred since 1 January 2024, at the
exchange ratio established in the conversion in March 2025.
In addition, the weighted average number of shares throughout the periods presented has also been
adjusted retrospectively for the impact of a share subdivision that became effective immediately prior to
the completion of the listing of the Company’s shares on January 8, 2026.
2025
2024
Loss for the year attributable to ordinary equity shareholders
of the Company (RMB’000)
(1,918,476)
(1,394,042)
Weighted average number of ordinary shares deemed to be
in issue
159,522,220
159,875,230
Basic loss per share (RMB)
(12.03)
(8.72)
(a)
Loss for the year attributable to ordinary equity shareholders of the Company
2025
2024
RMB’000
RMB’000
Loss for the year attributable to all equity shareholders
of the Company
(4,698,203)
(2,956,491)
Allocation of loss for the year attributable to financial
instruments issued to investors
2,779,727
1,562,449
Loss for the year attributable to ordinary equity
shareholders of the Company
(1,918,476)
(1,394,042)
– 29 –
(b)
Weighted average number of ordinary shares in issue/deemed to be in issue
2025
2024
Ordinary shares deemed to be in issue at 1 January
36,224,375
28,477,938
Effect of ordinary shares deemed to be in issue
2,841,407
2,112,110
Effect of increase in paid-in capital through transfer from
capital reserve
–
1,982,905
Effect of the financial instruments issued to investors
(23,113,560)
(16,585,430)
Effect of share subdivision
143,569,998
143,887,707
Weighted average number of ordinary shares in
Financial Receivables and Post-Listing Events
- Diluted loss per share remains identical to basic loss per share because convertible bonds and investor instruments were excluded as anti-dilutive.
- Trade receivables saw a significant increase from RMB 91.1 million in 2024 to RMB 303.2 million in 2025, primarily concentrated in the 'within 3 months' aging category.
- Trade and other payables more than doubled to RMB 1.33 billion, driven largely by a sharp rise in payables for computing service fees.
- The Group divested unlisted equity investments to an associate, Beijing Xinglian, for a total consideration of RMB 202.5 million based on arm's length negotiations.
- A major post-reporting event occurred on 8 January 2026, with the Company's successful listing on the Hong Kong Stock Exchange, raising gross proceeds of approximately HK$5 billion.
The financial instruments issued to investors and convertible bonds were not included in the calculation of diluted loss per share as their inclusion would have been anti-dilutive.
issue/deemed to be in issue at 31 December
159,522,220
159,875,230
(c)
Diluted loss per share
The financial instruments issued to investors and convertible bonds were not included in the
calculation of diluted loss per share as their inclusion would have been anti-dilutive. Accordingly,
diluted loss per share for the years ended 31 December 2025 and 2024 are the same as basic loss per
share.
– 30 –
8
TRADE AND OTHER RECEIVABLES
31 December
2025
31 December
2024
RMB’000
RMB’000
Trade receivables
339,198
100,170
Less: loss allowance
(35,990)
(9,035)
303,208
91,135
Deposits
70,538
67,912
Receivables from disposal of investments in equity securities
measured at FVPL (i)
7,098
45,216
Other receivables
52,029
263,805
129,665
376,933
Less: loss allowance
(18,994)
(27,327)
110,671
349,606
Financial assets measured at amortised cost
413,879
440,741
Input VAT deductible
168,768
98,729
Prepayments for computing service fee and others
116,295
127,371
285,063
226,100
698,942
666,841
Note:
(i)
In November 2024, the Group entered into a series of equity transfer agreements with Beijing
Xinglian, an associate of the Group, pursuant to which, the Group divested certain unlisted equity
investments measured at FVPL to Beijing Xinglian at a total consideration of RMB202,528,000.
The consideration was determined based on arm’s length negotiation and was with reference to the
most recent transaction price of the equity interests in those unlisted entities.
All of the trade and other receivables are expected to be recovered or recognised as expenses within one
year.
– 31 –
At the end of the reporting period, the ageing analysis of trade receivable (net of loss allowance), based on
the invoice date, is as follows:
31 December
2025
31 December
2024
RMB’000
RMB’000
Within 3 months
243,484
74,191
3 months to 6 months
19,831
13,804
6 months to 1 year
31,406
1,444
1 year to 2 years
8,415
1,302
2 years to 3 years
72
394
303,208
91,135
9
TRADE AND OTHER PAYABLES
31 December
2025
31 December
2024
RMB’000
RMB’000
Trade payables due to third parties
297,139
58,293
Payables for computing service fees
727,348
269,467
Payables for marketing and promotion services
42,812
89,052
Payables of staff costs
146,507
104,229
Other payables and accruals
54,437
43,767
Financial liabilities measured at amortised cost
1,268,243
564,808
Other taxes payables
39,774
22,304
Provisions for warranties
20,614
16,376
1,328,631
603,488
– 32 –
As at the end of the reporting period, the ageing analysis of trade payables, based on the invoice date, are as
follows:
31 December
2025
31 December
2024
RMB’000
RMB’000
Within 3 months
229,890
57,676
3 months to 6 months
46,448
57
6 months to 1 year
19,506
257
More than 1 year
1,295
303
297,139
58,293
All of the trade and other payables are expected to be settled within one year or are repayable on demand.
10
DIVIDENDS
The directors of the Company did not recommend the payment of a final dividend for the year ended 31
December 2025 (2024: nil).
11
NON-ADJUSTING EVENTS AFTER THE REPORTING PERIOD
On 8 January 2026, the Company completed its listing on the Stock Exchange and a total of 43,032,400
ordinary shares were issued (including initial offering and the exercise of over-allotment option) at the
price of HK$116.20. Total gross proceeds received were HK$5,000,365,000 (equivalent to approximately
RMB4,516,430,000).
– 33 –
OTHER INFORMATION
Issuance of H Shares and Listing on the Hong Kong Stock Exchange
On 8 January 2026, we successfully completed the Global Offering of 37,419,500 H shares
on the Hong Kong Stock Exchange (HKEX), with an offering price of HK$116.20 per H
share (excluding 1.0% brokerage commission, 0.0027% Hong Kong Securities and Futures
Commission (SFC) transaction levy, 0.00565% HKEX trading fee, and 0.00015% Hong Kong
Post-Listing Compliance and Governance
- The company successfully completed its Global Offering and exercised an over-allotment option for 5,612,900 H shares in February 2026.
- The Board of Directors has decided not to issue a final dividend for the fiscal year ending December 31, 2025.
- Since listing on the Hong Kong Stock Exchange, the company and its subsidiaries have not engaged in any purchase, sale, or redemption of listed securities.
- The company transitioned to strict compliance with the Corporate Governance Code and the Model Code immediately upon its listing date.
- Directors and relevant employees have confirmed full adherence to the Model Code regarding securities transactions and inside information.
- The company has maintained the required public float and fulfilled its continuous disclosure obligations under the Listing Rules.
The Board has always been committed to maintaining a high standard of corporate governance system, believing that a sound governance structure is the core support for safeguarding shareholders’ rights and interests.
Accounting and Financial Reporting Council (AFRC) transaction levy). The underwriters
of the Global Offering fully exercised the over-allotment option, and we newly issued and
allotted a total of 5,612,900 H shares on 4 February 2026.
Final Dividend
The Board does not recommend the payment of a final dividend for the year ended 31
December 2025 (2024: None).
Purchase, Sale, or Redemption of the Company’s Listed Securities or Sale of Treasury
Shares
H Shares of the Company was first listed on the Main Board of the Hong Kong Stock
Exchange on 8 January 2026. From the listing date until the date of this announcement,
neither the Company nor any of its subsidiaries has purchased, sold, or redeemed any of the
Company’s securities listed on the HKEX (including the sale of any treasury shares). The
Company does not hold any treasury shares as of the date of this announcement.
Post-Reporting Period Significant Events
Save as disclosed in this announcement, there have been no other material events affecting the
Company from the end of the Reporting Period until the date of this announcement.
Compliance with the Corporate Governance Code
The Board has always been committed to maintaining a high standard of corporate governance
system, believing that a sound governance structure is the core support for safeguarding
shareholders’ rights and interests, enhancing the long-term value of the enterprise, formulating
scientific business strategies and policies, and strengthening operational transparency and
accountability mechanisms.
– 34 –
As the H shares had not been listed on the Hong Kong Stock Exchange during the Reporting
Period, the Corporate Governance Code was not applicable to the Company during that
time. Based on the principles and provisions of the Corporate Governance Code as set out
in Appendix C1 to the Listing Rules, the Company recognizes that the Board should consist
of a balanced combination of executive Directors and independent non-executive Directors
to ensure that the Board has independent supervisory capabilities and can make objective
judgments in major decision-making processes.
From the Listing Date until the date of this announcement, the Company has strictly complied
with all applicable provisions of the Corporate Governance Code. In addition, the Company
will actively refer to the recommended best practices of the Corporate Governance Code,
including establishing a sound internal control system, enhancing the timeliness and accuracy
of information disclosure, safeguarding shareholders’ right to information and participation,
and continuously optimizing the governance structure.
Compliance with the Model Code
Since its listing, the Company has adopted the Model Code as its code of conduct governing
the Company’s securities transactions conducted by its Directors and employees who have
access to inside information concerning the Group or the Company’s securities.
The Company has made specific inquiries to all the Directors, all Directors have confirmed
that they have complied with the Model Code from the listing date until the date of this
announcement.
Employees of the Group who may possess inside information of the Group shall comply with
the Model Code. From the listing date until the date of this announcement, the Company
has not identified any incidents of non-compliance with the Model Code by the relevant
employees.
Continuous Disclosure Obligations under the Listing Rules
Save as disclosed in this announcement, the Company has no other disclosure obligations
under Rules 13.20, 13.21, and 13.22 of the Listing Rules.
Adequacy of Public Float
Based on the information available to the Company and to the best knowledge of the
Directors, from the Listing Date until the date of this announcement, the Company has
continuously maintained the public float as required by the Listing Rules.
– 35 –
Scope of Work of the Auditors
Financial Verification and Definitions
- KPMG has verified that the figures in the annual results announcement align with the audited consolidated financial statements for 2025.
- The Audit Committee, chaired by Dr. Xie Deren, has reviewed the Group's annual results and confirmed compliance with International Financial Reporting Standards.
- The company officially listed its H Shares on the Main Board of the Hong Kong Stock Exchange on January 8, 2026.
- The announcement provides a comprehensive glossary of terms, defining the 'Group' as Knowledge Atlas and its various subsidiaries.
- Future annual reports will be distributed to H Share shareholders and made available on the official websites of both the Stock Exchange and the Company.
The work performed by KPMG in this respect did not constitute an assurance engagement and consequently no opinion or assurance has been expressed by KPMG on this announcement.
The figures in respect of the Group’s consolidated statement of financial position,
consolidated statement of comprehensive loss and the related notes thereto for the year
ended 31 December 2025 as set out in this annual results announcement have been agreed by
the Company’s auditor, KPMG, to the amounts set out in the Group’s audited consolidated
financial statements for the year ended 31 December 2025. The work performed by KPMG
in this respect did not constitute an assurance engagement and consequently no opinion or
assurance has been expressed by KPMG on this announcement.
Audit Committee
The Audit Committee comprises Dr. Xie Deren, Dr. Yang Qiang, and Dr. Li Juanzi. Dr. Xie
Deren serves as the Chairman of the Audit Committee. The Audit Committee has reviewed
the Group’s annual results for the year ended 31 December 2025, as well as the audited
consolidated financial statements of the Group for the year ended 31 December 2025, prepared
in accordance with International Financial Reporting Standards.
Publication of Annual Results and Annual Report
This results announcement has been published on the websites of the Stock Exchange
(http://www.hkexnews.hk) and the Company (http://www.zhipuai.cn). The Company’s
annual report for the year ended 31 December 2025, will be published on the aforementioned
websites of the Stock Exchange and the Company in due course and will be sent to the
Company’s H Share shareholders in accordance with their chosen method of receiving
corporate communications.
DEFINITIONS AND GLOSSARY OF TECHNICAL TERMS
In this announcement, unless the context otherwise requires, the following terms shall have
the meanings set out below:
Definitions
“Articles of Association”
the articles of association of the Company adopted on 28
June 2025 and effective from the Listing Date, as amended,
modified or supplemented from time to time
“Audit Committee”
the Audit Committee of the Board
“CG Code”
the Corporate Governance Code as set out in Appendix C1
to the Listing Rules
– 36 –
“China” or “PRC”
the People’s Republic of China, and solely for the purposes
of this announcement, excluding Hong Kong, the Macau
Special Administrative Region, and Taiwan
“Director(s)”
the director(s) of the Company
“Global Offering”
the Hong Kong Public Offering and the International
Offering as defined in the Prospectus
“Group” or “We” or
“Knowledge Atlas”
the Company and its subsidiaries
“H Shares”
the overseas-listed shares of the Company with a par value
of RMB0.10 each, traded in Hong Kong dollars and listed
and traded on the Stock Exchange
“Hong Kong dollar(s)”
Hong Kong dollar(s), the lawful currency of Hong Kong
“Hong Kong”
the Hong Kong Special Administrative Region of the PRC
“Listing”
the listing of the H Shares on the Main Board of the Stock
Exchange on 8 January 2026
“Listing Date”
8 January 2026, the date on which the H Shares were listed
on the Main Board of the Stock Exchange
“Listing Rules”
“the Rules Governing the Listing of Securities on The Stock
Exchange of Hong Kong Limited”
“Model Code”
the Model Code for Securities Transactions by Directors
of Listed Issuers contained in Appendix C3 to the Listing
Rules, as amended, supplemented or otherwise modified
from time to time
“Prospectus”
the Prospectus of the Company dated 30 December 2025 in
relation to the Global Offering and the Listing
“Renminbi” or “RMB”
Renminbi, the lawful currency of the PRC
“Reporting Period”
the year ended 31 December 2025
– 37 –
“SFC”
the Securities and Futures Commission of Hong Kong
“Share(s)”
ordinary share(s) of RMB0.10 each in the share capital of
the Company, including unlisted shares and H shares
“Shareholder(s)”
holder(s) of the Share(s)
“Stock Exchange” or “Hong
Kong Stock Exchange”
The Stock Exchange of Hong Kong Limited
“subsidiary(ies)”
the meaning ascribed to it under the Listing Rules
“the board of Directors”
the Board of Directors of the Company
“Company”
Glossary of AI Terms
- The document defines Knowledge Atlas Technology Joint Stock Company Limited as a PRC-based entity listed on the Stock Exchange.
- It provides formal definitions for core AI concepts including Artificial General Intelligence (AGI), foundation models, and deep learning.
- The glossary details proprietary technical innovations such as the Direct two-sided importance sampling algorithm and FlashComm.
- Specific hardware and infrastructure terms like GPU, computing power, and API are defined within the context of model training and inference.
- The text outlines specialized mechanisms for model robustness, such as the Heartbeat fault tolerance mechanism for automatic recovery.
AGI: artificial general intelligence, a sophisticated level of artificial intelligence that matches and even surpasses human capabilities across all cognitive tasks.
Knowledge Atlas Technology Joint Stock Company Limited
(北京智譜華章科技股份有限公司), a limited liability
company established under the laws of the PRC on 11 June
2019 and converted into a joint stock company with limited
liability on 26 March 2025, with its H Shares listed on the
Stock Exchange (Stock Code: 2513)
“treasury shares”
the meaning ascribed to it under the Listing Rules
“Unlisted Share(s)”
ordinary share(s) issued by the Company, with a par value
of RMB0.10 each, which is/are not listed on any stock
exchange
“United States,” “USA” or
“U.S.”
the United States of America, its territories, its possessions
and all areas subject to its jurisdiction
“U.S. dollars,” “US$” or
“USD”
United States dollars, the lawful currency of the United
States
“%”
percentage
– 38 –
Glossary of Technical Terms
“AGI”
artificial general intelligence, a sophisticated level of
artificial intelligence that matches and even surpasses human
capabilities across all cognitive tasks
“AI”
artificial intelligence, an area of computer science that
focuses on machinery simulation of intelligence displayed by
humans and other animals
“AI agent”
a system or program that utilizes AI to perform tasks and
achieve goals autonomously on behalf of a user or another
system
“algorithm”
a procedure or formula for solving a problem, based on
conducting a sequence of specific actions, especially by a
computer
“API”
application programming interface, a set of predefined
rules, protocols and tools that allow users to integrate AI
capabilities into applications, websites or software
“computing power”
the ability of a computer system to process data and perform
tasks
“deep learning”
a machine learning technique that constructs artificial neural
networks with multiple layers to extract features from the
raw input
“Direct two-sided importance
sampling algorithm”
Our proprietary asynchronous training algorithm that
addresses action-reward misalignment and improves the
efficiency of learning from real-world environments
“Dynamic sparse attention
mechanism”
A technique for long-sequence reasoning optimization that
dynamically selects critical attention heads or positions to
reduce computation without compromising performance
“FlashComm”
A communication optimization technique that reduces
communication latency during model training or inference
and improves the efficiency of distributed systems
“foundation model”
a pre-trained LLM that serve as the foundation for the
development of a variety of specialized models
– 39 –
“general-purpose large
model”
an AI model trained on a vast and diverse dataset and
designed to perform a broad range of tasks rather than being
limited to a specific, narrowly defined application
“GPU”
graphics processing unit, a specialized processor originally
designed for rendering graphics, but now widely used to
accelerate parallel computations in fields like AI, especially
for training and inference of deep learning models
“Heartbeat fault tolerance
mechanism”
A robustness mechanism for large model training that detects
failures via periodic heartbeat signals and enables automatic
recovery to prevent training interruptions
“inference”
the process of applying trained models to new data to
generate predictions
“large language model
(LLM)”
a deep-learning AI model trained on vast amounts of text to
understand, generate and interact in human language
“large model”
a deep-learning AI model trained on vast amounts of data
and designed to perform a broad range of tasks
“Lightning Indexer”
Customized fused kernel technique that optimizes memory
access, reduces latency, and enhances inference efficiency of
models on domestic AI chips
“Long-horizon tasks”
Complex tasks requiring multi-step iteration and logical
consistency (e.g., enterprise-level long-chain execution),
which demand long-term planning and memory capabilities
of the model
“MaaS”
Technical Glossary and Corporate Governance
- The document defines specialized AI delivery models like Model-as-a-Service (MaaS) tailored for specific industry verticals.
- Technical optimizations such as MLA-256 and Muon Split are introduced to enhance long-sequence processing and reduce KVCache usage.
- The Slime framework is highlighted as an asynchronous reinforcement learning tool designed to improve GPU utilization by decoupling generation and training.
- Operational efficiency is addressed through Prefill-Decode decoupling, which streamlines the model inference process.
- The text concludes with formal corporate disclosures for Knowledge Atlas Technology Joint Stock Company Limited, dated March 2026.
- A legal disclaimer specifies that in cases of ambiguity between Chinese and English names of regulations or entities, the Chinese version prevails.
Our asynchronous reinforcement learning framework that decouples generation and training, resolves idle cycles in long-horizon tasks for intelligent agents, and improves GPU utilization.
Model-as-a-Service, a delivery model of AI model and agent
solutions catered for specific industry verticals or scenarios
“MLA-256 enhancement”
Architectural and algorithmic improvements based on Multi-
Level Attention with 256-dimension feature optimization to
enhance long-sequence processing capability
“Muon Split optimization
strategy”
Our proprietary optimization for model training that splits
attention computation to reduce KVCache usage and improve
training stability
– 40 –
“open source”
a source code that is made freely available for possible
modification and redistribution
“Prefill-Decode (PD)
decoupling”
A technique that decouples the Prefill and Decode stages
of model inference to streamline the inference process and
boost efficiency
“pre-train”
the process of training models on large-scale data to learn
general features prior to task-specific fine-tuning
“SDK”
software development kit, a set of software development
tools that allows the creation of applications for a certain
software package
“Slime framework”
Our asynchronous reinforcement learning framework that
decouples generation and training, resolves idle cycles in
long-horizon tasks for intelligent agents, and improves GPU
utilization
“Token”
the basic unit of data processed by AI models
For ease of reference, the names of Chinese laws and regulations, government authorities,
institutions, natural persons or other entities are listed in both Chinese and English in this
announcement. In the event of any ambiguity, the Chinese version shall prevail.
By Order of the Board
Knowledge Atlas Technology Joint Stock Company Limited
Dr. LIU Debing
Executive Director and Chairman of the Board
Hong Kong, 31 March 2026
As at the date of this announcement, the Board comprises: (i) Dr. Liu Debing, Dr. Zhang Peng
and Ms. Zhang Xiaohan as executive Directors; (ii) Dr. Li Juanzi, Mr. Li Jiaqing and Mr.
Wang Meng as non-executive Directors; and (iii) Dr. Yang Qiang, Dr. Xie Deren and Mr.
Tang Ying as independent non-executive Directors.
Knowledge Atlas 2025 Financial Results
- The Group reported a massive 131.9% year-on-year revenue increase to RMB724.3 million, driven by its 'foundational model + platform + ecosystem' strategy.
- Despite revenue growth, the company posted a net loss of RMB4.7 billion, largely due to a 44.9% increase in R&D expenses to over RMB3.1 billion.
If the upper bound of intelligence determines the pricing power of technology, then the scale of token consumption determines the magnitude of commercial value.
The Leap to Agentic Engineering
- Knowledge Atlas transitioned from simple code generation to 'Agentic Engineering,' where AI acts as a digital engineer capable of autonomous planning and iteration.
- Technical optimizations like Muon Split and MLA-256 reduced KVCache usage and deployment costs by 50% while maintaining high performance.
AI is no longer just a simple code generator but a “digital engineer” capable of autonomous planning, testing, and iteration.
The Rise of Token Architecture
- The BigModel.cn platform has become a central hub for industrial AI, with 9 of China's top 10 internet companies integrating GLM-5 within 24 hours of release.
- Core competitiveness is being redefined as Token Architecture Capability (TAC), measuring the efficiency of converting intelligence calls into economic value.
This confirms that high-order intelligence is a scarce resource today, and whoever controls the upper bound holds the pricing power.
The Intelligence Output Revolution
- An 'Intelligence Output Revolution' is shifting the focus toward high-quality token exports, leveraging China's industrial chain to move from 'Made in China' to 'Intelligence in China.'
- Total revenue rose from RMB312.4 million in 2024 to RMB724.3 million in 2025, driven by enhanced model intelligence.
Token exports are not about low-price competition but about “high quality at competitive prices” based on top-tier intelligence levels like GLM-5.
Financial Growth and Margin Shifts
- The Group achieved a 68.7% increase in gross profit, rising to RMB296.7 million in 2025, despite a significant drop in gross margin from 56.3% to 41.0%.
- Cloud-based deployment saw explosive growth with a 2,150% increase in gross profit, driven by improved inference efficiency and economies of scale in computing capacity.
The gross profit of the cloud-based deployment business increased from RMB1.6 million in 2024 to RMB36.0 million in 2025, representing a growth rate of 2,150.0%.
AI Risk Factors and Financials
- The company faces significant uncertainties regarding the future realization of Artificial General Intelligence (AGI) and the performance of its proprietary models.
- Research and development expenses represent the largest cost driver, exceeding RMB3.1 billion in 2025 as the company scales its technology.
The development of AGI is still at an early stage and there are substantial uncertainties in the future realization of AGI.
Financial Receivables and Post-Listing Events
- Trade receivables rose from RMB91.1 million in 2024 to RMB303.2 million in 2025, primarily concentrated in the 'within 3 months' aging category.
- A major post-reporting event occurred on 8 January 2026, with the Company's successful listing on the Hong Kong Stock Exchange, raising gross proceeds of approximately HK$5 billion.
The financial instruments issued to investors and convertible bonds were not included in the calculation of diluted loss per share as their inclusion would have been anti-dilutive.
Financial Verification and Definitions
- KPMG has verified that the figures in the annual results announcement align with the audited consolidated financial statements for 2025.
- The company officially listed its H Shares on the Main Board of the Hong Kong Stock Exchange on January 8, 2026.
The work performed by KPMG in this respect did not constitute an assurance engagement and consequently no opinion or assurance has been expressed by KPMG on this announcement.