Each surge of market enthusiasm for AI typically triggers a sequential rally in graphics cards, power supplies, optical modules, and servers. This pattern is understandable, as these components most closely resemble "industrialized" products and are easiest for capital markets to comprehend: expansions in production capacity, new order acquisitions, and upstream price increases are straightforward to illustrate and narrate. However, the most visible segments of the AI supply chain are not necessarily the most valuable in the long run.
Over the past two years, the market often viewed AI as a race centered on "supplying shovels to large models." By 2026, this perspective is becoming insufficient. Policy direction has clearly shifted towards emphasizing practical application. Recent State Council guidelines on enhancing the service sector explicitly call for the deep implementation of the "AI Plus" initiative and support for procuring large model and intelligent agent services. This year's government work report further advocates for expanding "AI Plus" applications, accelerating the promotion of intelligent agents, and driving commercial, large-scale AI adoption in key industries.
The trend is unmistakable: the next phase of AI is not merely a technological challenge but revolves around procurement, budgeting, and organizational roles. This explains why the most noteworthy segment in Hong Kong's AI sector may not be the compute chain alone, but rather the line encompassing intelligent agents and AI-native application platforms. While compute power sells "electricity" and models sell "intelligence," intelligent agents offer what businesses are most willing to pay for continuously: tangible outcomes.
Put plainly, AI only transitions from a "demonstration technology" to "measurable productivity" when it integrates into core business processes such as marketing, sales, customer service, operational analysis, knowledge management, and R&D collaboration. This is the unique advantage of the intelligent agent sector. It does not merely sell a conversational tool but offers a new organizational unit. The focus shifts from superficial interaction to practical utility: Can it handle work orders, execute workflows, access knowledge, coordinate across multiple systems, and ultimately deliver business results? The key question is whether it enhances operational efficiency and generates real value for enterprises.
Consequently, the most valuable players in the intelligent agent space are not those with the largest model parameters, but those closest to achieving this goal: transforming AI from a functional tool into an operational role and a complete system. This logic is crucial because it defines the foundation of valuation. The compute chain often profits from cyclical demand, dependent on capital expenditure, supply-demand dynamics, and the roadmaps of overseas tech giants. In contrast, intelligent agent platforms generate returns based on adoption rates, measured by actual customer usage, repeat purchases, and expansion from single departments to organization-wide deployment.
These are fundamentally different business models with distinct valuation frameworks. The former resembles cyclical growth stocks, while the latter aligns more with platform-based application stocks. One thrives on hardware expansion elasticity, the other on the depth of AI-driven business process re-engineering.
From this perspective, Hong Kong's so-called "AI trio" exemplifies three different strategic positions. Xunce Technology acts more as a "water seller," refining raw enterprise data into high-quality fuel for models. Dipu Technology functions as a "master craftsman," embedding industry expertise and complex workflows into AI. MARKETINGFORCE increasingly resembles a "full-stack token factory," aiming to convert models, knowledge, platforms, and scenarios into sustainable, paid business outcomes for enterprises.
The critical factor for market valuation sensitivity is not simply "who is involved in AI," but who can demonstrate that AI has become an integrated "employee" or "specialized team" within an organization. This is why the intelligent agent sector represents a business with greater long-term potential than many anticipate. Traditional software often involves a one-time purchase, while intelligent agent platforms typically start with a single use case, expand to additional departments, and eventually evolve into organization-wide infrastructure, representing a gradual, deepening long-term investment.
Initially adopted in customer service, AI may later extend to sales, operational analysis, and eventually integrate knowledge bases, advertising, training, performance review, and R&D collaboration into a unified foundation. Enterprises thus appear to purchase an AI platform but ultimately acquire a comprehensive "AI workforce expansion plan."
This explains a practical observation: why the intelligent agent sector may not always surge the most dramatically but increasingly attracts sustained discussion. The market is realizing that the scarcest resource in the AI industry is not the ability to build a model, but the capability to sell AI as a long-term budgetary item. Ultimately, capital markets pay for profitability, not science fiction. Those who can transform AI from a "chatty novelty" into a "productive worker" are more likely to evolve from thematic stocks into platform stocks.
In summary, the first half of Hong Kong's AI narrative focused on supplying oxygen to models; the second half will be won by those who first equip AI with employee badges.