HAIZHI TECH GP (02706) faced its first major test 44 days after listing. Having debuted on the Hong Kong Stock Exchange on February 13, 2026, with a first-day surge of 246% and a public offering oversubscribed by 5,065 times, market enthusiasm was not directed at a mere graph database company but rather at the growth logic of AI agents driven by 'graph-model fusion'. On March 27, the company released its first annual report post-listing. Total revenue reached 621 million yuan, a year-on-year increase of 23.4%. Adjusted net profit was 24.15 million yuan, up 42.6% year-on-year. Revenue from the core growth engine, Atlas AI Agents, surged 68.4% year-on-year to 146 million yuan. While existing analysis highlights the '68.4% growth rate confirming a second growth curve,' this report aims to dissect more precise figures to answer three key questions: What is the quality of this curve? How far is it from fundamentally altering the company's revenue structure? Which metrics should investors monitor for ongoing validation?
Examining the two business lines together reveals HAIZHI TECH GP's revenue composition: Atlas Graph Solutions (the core foundation) and Atlas AI Agent Solutions (the second growth curve). The full 2025 data shows a clear divergence in growth rates, but several easily overlooked details are more telling. The average contract value for AI Agents is 3.6 million yuan, nearly 30% higher than for Graph Solutions. This indicates clients are not merely 'testing the waters' but making significant 'bets.' A 3.6 million yuan procurement decision requires multiple layers of approval in any large institution; achieving this price point signifies the product's integration into clients' core business processes. More importantly, such substantial purchases typically represent strategic, enterprise-level investments in intelligent transformation, positioned by client management, rather than isolated departmental needs. Compared to routine tool software procurement, strategic projects have more robust long-term budget planning and greater potential for expansive budgets supporting full-scenario implementation and iterative upgrades.
The gross margin for AI Agents is 53.2%, approximately 14 percentage points higher than for Graph Solutions. In the To-B software industry, gross margin is a strict measure of product standardization. Project-based companies often see margins of 20-30%; achieving over 50% signifies high standardization, minimal custom development, and strong technology reuse. The 53% gross margin for AI Agents directly boosted the company's overall gross margin from 36.3% to 43.3%, a 7 percentage point increase in one year—a rare feat in the To-B sector.
However, the most striking figure lies within the client structure: 50% of AI Agent clients were converted from existing Graph Solutions clients. The importance of this statistic cannot be overstated. It demonstrates HAIZHI TECH GP has validated a client upgrade path: 'sell infrastructure first, then sell upper-layer applications.' Clients already deployed with graph databases and knowledge graphs possess the inherent technical foundation and data assets for evolving towards agents, significantly reducing customer acquisition costs and improving delivery efficiency. The annual report also disclosed that five clients have made repeated purchases since the launch of the AI Agent product. In the enterprise market, the 'first order' might be opportunistic, but the 'second order' relies solely on product merit. Securing repeat business from clients constitutes true commercial validation.
Countdown to Reversal: When Will AI Agents Take the Lead? Having assessed the curve's quality, the next question arises: At this pace, when will AI Agents surpass Graph Solutions to become the company's primary business? The current ratio is 77:23, a gap exceeding threefold. Three scenarios were modeled: - Optimistic Scenario: AI Agents maintain 65% growth; Graph Solutions maintain 15% growth. Under this scenario, AI Agent revenue share approaches 50% by 2028, overtaking Graph Solutions in 2029. - Baseline Scenario: AI Agent growth gradually slows to 40%; Graph Solutions maintain 12% growth. Here, AI Agent share nears 45% by 2030, with a complete reversal occurring between 2030-2031. - Conservative Scenario: AI Agent growth drops to 30%; Graph Solutions growth is 10%. Even under the most conservative assumptions, AI Agent share breaks 40% by 2030.
All three scenarios point to the same conclusion: a reversal in revenue structure is directionally certain, with the only divergence being a 3-year or 5-year timeframe. Furthermore, these models do not yet factor in an accelerant: the conversion potential of the existing Graph Solutions client base. By the end of 2025, there were 172 Graph Solutions clients, with only approximately 20 converted to AI Agents—a conversion rate of about 12%. The remaining 150+ clients represent a substantial 'pool for potential conversion.' If the conversion rate increases to 30-40%, AI Agent growth would gain additional internal momentum, potentially accelerating the reversal timeline.
10 Billion Yuan in Cash: What's the Next Move? Growth requires resources. Currently, HAIZHI TECH GP is well-equipped in this regard. As of February 28, 2026, the company held cash and financial assets exceeding 10 billion yuan. For a company with annual revenue of 620 million yuan, this represents reserves equivalent to approximately 1.6 years of revenue—a solid financial foundation amid the普遍 'cash-strapped' environment of the AI industry. These resources primarily come from IPO proceeds (net amount approximately 655 million HKD) and accumulated operational earnings. Signals from the annual report indicate clear spending priorities for management:
First, R&D takes precedence. R&D expenses in 2025 grew 64.6% year-on-year to approximately 100 million yuan, significantly outpacing revenue growth. The company is leading national-level major projects focused on breakthroughs in next-generation multi-model database key technologies and was selected as the sole lead for a key Beijing municipal project. The new version of the AtlasGraph database achieved significant enhancements in dozens of functions, including temporal graphs and data lifecycle management, while GraphRAG technology also upgraded its capabilities for parsing structured and semi-structured text. This R&D investment is building strength for the next generation of products.
Second, territorial expansion. The company's industry footprint is expanding beyond its strongholds in finance and public safety. The annual report mentions new engagements: building a graph ontology model for unmanned devices (drones, unmanned ships, unmanned vehicles) for an unmanned space management center in the Greater Bay Area; deploying road network operation analysis agents and infrastructure maintenance agents for a city tunnel company; and a large state-owned telecom operator making repeat purchases for the third consecutive year, expanding from the graph database to ecosystem tools like GraphAgent. High-end manufacturing, healthcare, and intelligent mining are listed as key next directions.
Third, exploring international expansion. The company has already deployed an AI Agent application platform, including large model appliances, for an overseas non-bank financial institution and has explicitly stated its intention to leverage its Hong Kong listing to explore Southeast Asian markets. Overseas revenue remains small currently, but the Hong Kong stock identity naturally opens doors for international capital operations.
Concurrently, operational efficiency is improving. Excluding listing expenses, net operating cash outflow narrowed from 38.5 million yuan in 2024 to 30.2 million yuan in 2025.
Moat: Why is Replication Difficult? The market is not short of AI companies or players touting 'AI agents.' What allows HAIZHI TECH GP to defend its position? Consider a positioning datum: According to Frost & Sullivan, based on 2024 revenue, HAIZHI TECH GP ranked fifth among industrial-grade AI agent providers in China—behind giants like Baidu, Alibaba, and Huawei. However, in the specific sub-segment of 'graph-centric AI agent providers,' HAIZHI TECH GP ranks first, with a market share exceeding 50%. This difference reveals the company's competitive strategy: avoiding direct confrontation with major players on general-purpose large models, instead securing the intersection of graph-model fusion to dominate a niche market.
This positioning currently rests on three layers difficult for competitors to replicate in the short term. First is the underlying technology. The AtlasGraph database broke a world record in the 2023 LDBC international benchmark test, outperforming the previous record by 45% and ranking first in comprehensive performance. Graph databases are underlying infrastructure with long development cycles and high engineering complexity, not something achievable quickly simply by spending money. The 2025 new version further widened the gap.
Second is industry沉淀. Serving over 400 clients across 100+ industry scenarios, the most significant achievement is completing domestic replacement of graph databases in four of the six major state-owned commercial banks. The knowledge ontologies, data governance experience, and industry methodologies accumulated during delivery constitute a true隐性 barrier.
Third is the商业飞轮. The 'Graph → Agent' upgrade path is already proven, with the 50% client conversion rate and repeat purchases forming a positive feedback loop. Pure large model companies lack the graph client base and data foundation, while pure database companies lack the application capabilities for upper-layer agents. Very few players can打通 the full stack.
Risks must be acknowledged. Microsoft has released its GraphRAG framework, and major large model vendors are enhancing their graph computing capabilities. If the graph capabilities of general-purpose tools improve significantly, it could compress HAIZHI TECH GP's technological differentiation. However, within the foreseeable 2-3 years, it is almost impossible for general frameworks to match HAIZHI TECH GP's industry depth in highly specialized, security-critical scenarios like energy grid dispatch, financial fund supervision, and high-end manufacturing, which require extensive domain knowledge.
How Will the Market Price It? Focus on These Five Numbers Returning to the initial question—how much of the expectation behind the 5,065-times oversubscription has this annual report fulfilled? The direction is validated. The graph-model fusion technical route has achieved a commercial闭环 on the industrial end, supported by a 50% conversion rate and tangible repeat orders along the 'sell graph first, then sell agents' client upgrade path. The pace is steady. A 23.4% revenue growth rate, 43.3% gross margin, and 24.15 million yuan in adjusted net profit—against a backdrop of普遍 revenue growth without profit growth in the AI industry, HAIZHI TECH GP delivered results balancing growth and profitability. Sustaining profitability for two consecutive years is a rare attribute within the sector.
The potential lies in the evolution of the revenue structure. Starting from 23%, AI Agents are expected to become the largest business within 3-5 years. The 10 billion yuan cash reserve, continued increases in R&D investment, and expansion into new industries and overseas markets provide upside elasticity to the growth curve. Whether this logic can persist will be verified by each subsequent financial report. Long-term tracking of five core metrics is recommended: AI Agent revenue growth rate; the pace of expansion of the AI Agent client base beyond the current 40; whether the overall gross margin can continue approaching 50%; when operating cash flow turns positive; and when overseas revenue achieves further breakthroughs. For a company in the AI agent赛道 that is profitable, has defensible barriers, and earns repeat business from clients, investors might consider giving it some time.
Comments