Dual Large‑Model Leaders Propel Hong Kong Stocks Higher As AI Applications Diversify

The AI sector, now the dominant technology theme, appears to be shifting from a hardware‑centric phase toward software and application development.

On January 12, large‑model developers Zhipu and MiniMax both reached fresh highs, leading gains among Hong Kong AI application stocks. Mainland counterparts also rallied strongly, with more than 20 A‑share names including E‑Point World, Visual China, BlueFocus and Shengguang Group hitting daily limits. Market participants view the listings of Zhipu and MiniMax as a potential inflection point that could move the industry from technical validation into commercial realization, elevating attention on AI applications. In 2026, AI applications are expected to progress from “usable” to “well‑adopted,” emerging as the next major theme after compute power.

Zhipu’s shares surged intraday by over 60% on January 12, briefly lifting its market capitalization above HKD 110 billion, before settling up 31% at HKD 208.40 per share. The company announced a strategic collaboration with Didi focused on intelligent agents, committing to joint exploration of core AGI technologies and agent applications in mobility. The partnership aims to accelerate agent scenario deployment, cultivate talent in large models, and strengthen intent alignment and reasoning capabilities for complex travel‑related use cases.

MiniMax also posted strong gains, rising nearly 40% intraday and closing up 15% at HKD 398 per share. The firm debuted on the Hong Kong Stock Exchange on January 9, closing that day at HKD 345—109% above its issue price—and achieving a market capitalization above HKD 100 billion. Observers attribute MiniMax’s market reception to a combination of technical depth, product commercialization potential and scarcity value. Sources close to the company indicate its B‑side business avoids private on‑premise deployments in favor of recurring revenue models, with overseas revenue already outpacing domestic sales; on the consumer side, the company focuses selectively on content tools, platforms and agents rather than competing directly in conversational products.

The twin rallies have catalyzed interest in downstream AI use cases, with the market increasingly focused on GEO (Generative Engine Optimization). User behavior is shifting from traditional search engines toward large models such as ChatGPT and Doubao for direct answers. Where SEO sought to elevate links in search results, GEO aims to have content directly ingested and prioritized by large models. With Zhipu and MiniMax listed, investors are seeking applications that can capture and monetize large‑model capabilities, and GEO has rapidly emerged as a key area within digital marketing.

Analysts at Huaxin Securities note that the successive listings underscore opportunities for digital‑marketing GEO to capture AI‑driven value. In the first quarter of 2026, AI applications combined with domestic demand could sustain momentum, and commercialization prospects span digital marketing, e‑commerce, content and experience economies. As model capabilities improve—particularly in reasoning and long‑context processing while costs decline—commercial validation across search marketing, coding, multimodal systems, agents and AI for science is expected to accelerate.

Recent strong performers in the AI concept space reflect this logic. For example, E‑Point World applies large models and algorithms to programmatic advertising and AIGC creative generation, aiming to lower costs and improve efficiency across the full overseas marketing chain.

Looking beyond search, industry leaders are debating the next strategic priorities. At a recent closed meeting at Tsinghua University, Professor Tang Jie, founder of Zhipu, argued that with the arrival of DeepSeek the contest over replacing search has effectively concluded, and the sector must now identify the subsequent battleground. Tang said his team has concentrated efforts on coding, citing the December 2025 release and open‑sourcing of GLM‑4.7, which strengthened coding capabilities, long‑task planning and tool collaboration and performed well on mainstream benchmarks.

In addition to coding, agents remain a central focus. Lin Junyang, head of Alibaba’s Qwen technology, suggested that agents must ultimately interact with the physical world—participating in experiments, manufacturing and scientific inquiry—to realize substantive workplace capabilities. Yao Shunyu, Chief AI Scientist in Tencent’s CEO/President Office, observed that agent adoption in enterprise scenarios continues to expand, driven by the simple logic that more intelligent models can address a broader set of tasks and thus deliver greater value in B‑side applications. Tang emphasized that value, cost and speed are the decisive factors: agents must solve genuinely valuable human problems, keep costs manageable and be deployed rapidly. The balance among these three dimensions will determine whether agent products transition from concept to commercial breakout.