Despite recording a net loss of approximately 4.718 billion yuan, the company known as the "first global large-scale AI model stock" has maintained a market valuation exceeding 400 billion yuan. KNOWLEDGE ATLAS (02513.HK), which secured this title by listing ahead of MINIMAX-WP (00100.HK), released its first annual report after going public on March 31.
The financial results for the fiscal year ending December 31, 2025, highlight both rapid growth and the characteristic challenges of the AI large model industry: substantial investment and difficulties in achieving short-term profitability.
Revenue doubled, with API income surging significantly. In 2025, KNOWLEDGE ATLAS generated revenue of approximately 724 million yuan, a 131.9% year-on-year increase, making it the highest-revenue independent large model provider in China. Revenue from its development platform and API (cloud deployment) reached about 190 million yuan, surging 292.6% year-on-year. Enterprise-grade intelligent agent revenue was approximately 166 million yuan, up 248.8% year-on-year.
Concurrently, enterprise-grade general large model revenue was about 366 million yuan, increasing 70.5% year-on-year. Technical services and other revenues remained relatively small, at merely 2.544 million yuan. Notably, this report marked the first time KNOWLEDGE ATLAS segmented its on-premises deployment business into three categories: enterprise-grade intelligent agents, enterprise-grade general large models, and technical services and other. The company stated that its previous classification based primarily on deployment method could no longer fully encompass its increasingly diverse product forms and commercial scenarios, leading to the new segmentation by business form and core product lines. This reclassification provided greater clarity to the company's revenue structure.
In 2025, the proportion of revenue from KNOWLEDGE ATLAS's cloud deployment business increased to 26.3%, up from 15.5% in 2024. The gross margin for this business reached 18.9%, an improvement of 15.6 percentage points from 2024. According to multiple media reports, the annual recurring revenue (ARR) for KNOWLEDGE ATLAS's MaaS API platform reached 1.7 billion yuan in 2025, a 60-fold increase year-on-year. While ARR is not directly comparable to revenue, it indicates the rapid scaling of its MaaS API platform business.
This ARR growth is underpinned by rapid product iteration. From July 2025 to February 2026, KNOWLEDGE ATLAS launched four generations of flagship models—GLM-4.5, GLM-4.6, GLM-4.7, and GLM-5—within an eight-month period. The company disclosed that its latest model, GLM-5, was integrated by major firms like ByteDance, Alibaba, and Tencent within 24 hours of its release, with nine of the top ten internet companies in China deeply integrating GLM. As of March 2026, registered users on the KNOWLEDGE ATLAS MaaS (Model-as-a-Service) platform surpassed 4 million. In the first quarter of 2026, despite an 83% increase in API prices compared to the end of 2025, market demand continued to outstrip supply, with API call volume growing 400%. KNOWLEDGE ATLAS is now one of the leading providers in China by paid token consumption.
Gross margin faced pressure, and losses widened. While revenue grew rapidly, profitability did not improve simultaneously. In 2025, KNOWLEDGE ATLAS achieved a gross profit of approximately 297 million yuan, a 68.7% year-on-year increase, but its gross margin declined from 56.3% the previous year to 41.0%. The reasons for this decline are straightforward. On one hand, the increasing share of the cloud deployment business, whose gross margin, though improved from 3.3% in 2024 to 18.9% in 2025, is still in a growth phase. On the other hand, the gross margin for the on-premises deployment business decreased from 66.0% to 48.8% due to increased investment in delivery resources.
Compared to the 15.3 percentage-point decline in gross margin, the loss situation is more pronounced. The company recorded a net loss of approximately 4.718 billion yuan in 2025, a 59.5% year-on-year expansion. Under the non-International Financial Reporting Standards (Non-IFRS) measure, the adjusted net loss was about 3.182 billion yuan, widening 29.1% from the previous year.
KNOWLEDGE ATLAS's net loss did not decrease despite significant revenue growth, primarily due to high research and development spending. R&D expenses reached 3.18 billion yuan in 2025, a 44.9% year-on-year increase, approximately four times the revenue for the same period. Although the absolute loss increased, the efficiency of capital utilization improved compared to 2024, when R&D expenses were seven times revenue. Additionally, sales and marketing expenses were approximately 391 million yuan, largely flat year-on-year, while general and administrative expenses reached about 505 million yuan, a 278.3% increase, mainly attributable to increased administrative hiring and share-based compensation expenses.
As of the end of 2025, KNOWLEDGE ATLAS had 1,094 employees, with total annual employee compensation costs of approximately 1.363 billion yuan, translating to an average annual compensation of about 1.2462 million yuan per employee. The company also adjusted its computing power strategy, shifting in 2025 to primarily purchasing computing services supplemented by some equipment leasing, whereas it had relied more heavily on leasing computing equipment in 2024. As of December 31, 2025, cash and cash equivalents stood at approximately 2.259 billion yuan, a decrease of about 10.1 million yuan from the end of 2024. Total borrowing was about 689 million yuan, an increase, yet the debt-to-asset ratio actually decreased from 20.4% in 2024 to 15.2%. In simple terms, KNOWLEDGE ATLAS continues to use capital to buy time, but there are no signs of uncontrolled cash burn.
Stock price surged nearly 700%. The capital market's response was more direct than the income statement. On the morning of April 1, following the earnings release, KNOWLEDGE ATLAS's stock opened about 15% higher, reaching an intraday high of HK$938, a gain of 35.2%. By the close of Hong Kong trading that afternoon, the stock hit a new record high, closing at HK$915, up 31.94%, with a total market capitalization of HK$407.946 billion. Since its listing on January 8, 2026, the company's share price has accumulated a gain of 687.44%, solidifying its status as a top-performing stock.
KNOWLEDGE ATLAS conducted its Hong Kong IPO at an issue price of HK$116.2, raising approximately HK$5 billion (about 4.516 billion yuan). By listing on the Hong Kong Exchange just one day before MINIMAX-WP, despite both companies having completed regulatory filings on the same day, KNOWLEDGE ATLAS secured the "first stock" label.
A comparison between KNOWLEDGE ATLAS and MINIMAX-WP reveals interesting differences beyond the raw data. In 2025, MINIMAX-WP reported revenue of $79.038 million (approximately 543 million yuan), a 158.9% year-on-year increase; a gross profit of $20.079 million (approximately 138 million yuan), with a gross margin of 25.4%; and a net loss as high as $1.872 billion (approximately 12.868 billion yuan). However, on a Non-IFRS basis, MINIMAX-WP's adjusted net loss was about $251 million (approximately 1.725 billion yuan), which is actually lower than that of KNOWLEDGE ATLAS.
While both are large model companies, their situations differ. Firstly, their business paths diverge. KNOWLEDGE ATLAS is highly focused on the enterprise (B2B) market, selling model capabilities and AI solutions. MINIMAX-WP leans more towards the consumer (B2C) market, developing AI-native products around its self-developed multimodal model. These different business models directly impact their gross margins. B2B businesses typically enjoy higher and more stable gross margins, whereas B2C products require continuous investment in user acquisition and content ecosystem development, making gross margin improvement more challenging.
Secondly, their loss profiles differ. KNOWLEDGE ATLAS's losses are comparatively smaller. MINIMAX-WP recorded a fair value loss on financial liabilities of approximately $1.59 billion (approximately 10.929 billion yuan) in 2025, which was the primary driver of its massive accounting loss. Therefore, comparing only the adjusted net losses, the difference between KNOWLEDGE ATLAS and MINIMAX-WP is about 1.457 billion yuan, not as stark as the headline figures suggest.
Regarding human capital costs, at the end of 2025, MINIMAX-WP had 428 employees, with total employee compensation expenses of $84.3 million (approximately 579 million yuan) for the year, resulting in an average annual compensation of about 1.3536 million yuan per employee. This average is 107,400 yuan higher than that of KNOWLEDGE ATLAS, indicating that MINIMAX-WP operates with a leaner team but higher per-capita costs, more akin to an internet product company.
Despite these differences, it is clear that both companies are far from achieving profitability. In 2025, both doubled their revenues and saw gross profits grow, yet their losses continued to widen. This pattern of burning cash for growth is not unexpected during the expansion phase for AI large model companies.
In conclusion, the financial report indicates that KNOWLEDGE ATLAS is at the forefront of commercializing large models in China, leading in revenue scale with its MaaS API platform business accelerating. However, the challenges are equally clear: under规模化 expansion, the company faces high R&D investment, gross margin pressure, and persistent losses. The current KNOWLEDGE ATLAS is not yet a mature company but one still in its expansion phase, with growth potential yet to be fully realized. The capital market's willingness to pay a premium is predicated on a consensus view that large models represent not a short-term business but a long-term strategic positioning in AI infrastructure. At this stage, profitability is not the most critical metric; large model capability is.
Comments