The two leading AI models in the Hong Kong stock market, Zhipu and MiniMax, have surged fourfold! Token's overseas expansion ignites global computing power arbitrage. Which other companies are expected to benefit?

During the Spring Festival,Token's overseas expansion has become a hot topic in the market.

China’s large-scale models are capturing the global developer market. OpenRouter data shows that last week, the top ten models on the platform totaled approximately 14 trillion tokens. MiniMax M 2.5 ranked first with 2.57 trillion tokens, followed closely by Kimi K2.5 with 1.04 trillion tokens. Zhipu GLM 5 and DeepSeekV 3.2 ranked fourth and fifth, respectively. This indicates that China’s large language models are...Widely authorized and applied tooverseas tech companies, gradually gaining an advantage in the underlying infrastructuregaining dominance。

Source: OpenRouter

And asthe direct beneficiary of Token's globalization — $KNOWLEDGE ATLAS (02513.HK)$、 $MINIMAX-W (00100.HK)$it has surged over 400% since its IPO, becoming one of the 'AI Model Duo' in the Hong Kong stock market.

This article will analyze for fellow investors the underlying logic behind Token's rise as a leader in the 'new export economy,' and guide you to uncover investment opportunities worth planning ahead for in the Hong Kong stock market.

Why has Token risen to become the 'new export economy'?

The essence of Token's globalization lies in leveraging China’s ultra-efficient 'algorithm efficiency (low cost)' to directly challenge America’s 'computing power hegemony (high cost),' thereby establishing new value distribution rules in the global digital landscape.

Breaking down the cost of a token,the two most critical components are computing power and electricity.

Computing power represents the depreciation and amortization of GPUs, while electricity constitutes the operational costs of data centers. For example: When a user in the U.S. calls an API of a large Chinese AI model, data travels from California, passes through trans-Pacific undersea cables to reach GPUs at a Chinese data center, where electricity is consumed to complete inference tasks, and the result is transmitted back. Throughout this process, electricity never leaves China’s power grid, but its value is delivered across borders via tokens.

In other words,Tokens are the purest derivative of electricity.

From a medium- to long-term perspective, what allows domestic models to maintain their edge? The key lies in 'computing-electricity synergy.'Backed by the massive infrastructure push of 'East Data West Compute,' every token we export will always be anchored to China’s unique 'energy dividend.'This forms the core foundation supporting pricing power.

Which companies are expected to benefit?

'Token exports' serve not only as a litmus test for domestic large AI models but also mark a major turning point in the industry's investment logic:The capital market is transitioning from the early 'arms race on the training side' to a substantive boom in 'token consumption on the inference side.' The future market pricing focus will shift from merely tracking GPU hardware procurement volumes to evaluating token generation rates (TPS) and commercialization conversion rates (ARR). The expansion of Chinese models in overseas markets has preliminarily demonstrated their ability to generate global competitive advantages. Based on this industrial evolution logic, this article will delve into the following three core investment themes:

Main theme one: leading large model manufacturers

In this wave of 'Token going global,' $KNOWLEDGE ATLAS (02513.HK)$ and $MINIMAX-W (00100.HK)$ there is no doubt that they are the most representative big winners. Their ability to offer highly 'disruptive' pricing in the global market is rooted in China's severely undervalued electricity infrastructure dividend。

Currently, the input pricing for MiniMax M2.5 and Zhipu GLM-5 has been compressed to 0.3 USD per million tokens, compared to overseas giant Claude Opus 4.6’s quote as high as 5 USD, which is practically a devastating blow; output prices are less than one-tenth of their competitors. This stark pricing strategy perfectly illustrates how China’s 'cheap power grid' transforms into a significant competitive edge globally.

Main theme two: fiber optic industry

The core physical path for Token expansion overseas (overseas demand → undersea cable transmission → domestic inference) is transforming massive computational loads into a rigid demand for underlying transmission capabilities, directly igniting a new high-growth cycle for the fiber optic industry. $YOFC (06869.HK)$ With its top-tier positioning, it has become a core target in this space.

First, overseas demand is directly triggering new construction of transoceanic undersea cables; second, AI-driven fiber demand is 5-8 times that of traditional levels, and amid an 18-month production expansion cycle constraint, fiber prices have surged from RMB 17/core-kilometer last mid-year to the current RMB 50, with extremely high certainty; third, Changfei possesses 'hollow-core fiber' technology, priced 2000 times higher than ordinary fiber, meeting the low-latency needs of Token expansion and deeply tied to cutting-edge sectors like CPO, building an irreplaceable technological moat.

As a result, since the beginning of this year, the rise in Changfei’s fiber optic cable shares has approached 150%, with its market cap surpassing HKD 100 billion.

Main theme three: Computational power leasing

The ultimate destination for Token expansion lies inthe centralized handling of overseas inference services by domestic computational clusters.As massive overseas orders pour in, the demand for domestic inference computational power has exploded; compounded by persistently high pricing for overseas computational power, the domestic computational power leasing sector has officially entered a golden phase of both volume and price increases!

In the Hong Kong stock market, attention can be paid to $SENSETIME-W (00020.HK)$ 、 $GDS-SW (09698.HK)$ 、 $SUNEVISION (01686.HK)$ are notable.

SenseTime is not just an AI software company; its trump card is the vast 'SenseCore'. This is a supercomputing center equipped with tens of thousands of GPUs, supporting not only its own models but also providing large-scale computational power leasing services to external enterprises and major model manufacturers. This company aligns perfectly with the 'explosion of inference-side Token consumption'; when domestic and international small-to-medium developers need computational resources, SenseTime's computing pool becomes the most direct 'water-selling platform,' directly benefiting from rising volumes and prices in computational leasing.

As a leading high-performance third-party data center (IDC) operator in China, GDS Holdings is fully transitioning to an AI-focused computing center. To handle massive inference workloads, large model manufacturers will need to rent their top-tier cabinets that feature high density and liquid cooling capabilities.

Sun e-Info Group has been深耕 in the IDC industry in Hong Kong, China for 25 years, boasting the number one network interconnection service in Asia. Its major shareholder, Sun Hung Kai Properties, is one of Hong Kong's largest real estate developers. Backed by Sun Hung Kai Properties, the company enjoys advantages in land, power supply, financing, engineering, and more.

Beyond the three main themes mentioned above, Kingsoft Cloud, as an important computing power partner for companies like Zhipu, also stands to benefit significantly from the growing demand for token inference. Additionally, previously“Wall Street Pricing Restructuring: J.P. Morgan Names Zhipu and MiniMax as Top Picks! What Other 'Hard Tech' Stocks in Hong Kong Are Worth Investing In?”This article also outlined the four AI battle lines currently present in the Hong Kong stock market. Fellow investors who are interested can click to view more details~

1M Views