Central Circle Observation | Hong Kong Stocks' New AI Narrative: Embracing "Token Economics"

Southern Finance, 21st Century Economic Herald, Reporter Yuan Sijie, Hong Kong Report

In early spring 2026, Hong Kong’s AI sector is undergoing a valuation-logic overhaul, becoming a focal point for global capital. As of the midday session on March 24, the Hang Seng Tech Index stood at 4,779.52 points, up 1.42%. Notably, the Hang Seng Hong Kong Stock Connect AI thematic index, which reflects the overall performance of the AI industry, has gained about 15% year-to-date, indicating that funds are accelerating their concentration into the AI track.

Amid this wave, Hong Kong-listed homegrown AI large-model “twin stars” are especially eye-catching. At the midday session on March 24, MiniMax’s latest share price was HKD 999, up 9%, with a total market value of approximately HKD 313.322 billion; while Zhipu AI’s latest share price reached HKD 627, up 6.27%, with a total market value of HKD 279.544 billion. Since listing, the year-to-date share price gains of both companies have been over 400%.

Behind the market’s fervent demand, a clear valuation main line is emerging: when AI agents such as OpenClaw spark a “deployment-in-applications” wave, Token consumption driven by single tasks is growing exponentially. As the smallest unit through which AI large models process information, Token is becoming the core pricing unit for measuring both the depth of AI business deployment and commercialization potential.

In this competitive landscape reshaped by the AI large-model wave, Hong Kong investors and analysts are also searching anew for the “yardstick” to measure value.

Competition Landscape Reshaped

Changes in the competition landscape in the Hong Kong stock market have reshaped the valuation anchor for Hong Kong’s AI sector.

Market analysis suggests that the Hang Seng Tech Index still relies mainly on traditional internet and software companies as its core weight. However, in this round of the AI revolution, large models are becoming the new traffic entry point and value hub, while the user time and commercial space of traditional platforms and software are being continuously eroded.

In January 2026, MiniMax and Zhipu AI became among the world’s first publicly listed large-model R&D companies. After the “twin stars” of domestic AI large models saw their share prices rise by several multiples, their market valuations are already on par with Baidu and comparable to JD.com.

“The listing of technology companies such as MiniMax and Zhipu AI is expected to allow investors to directly participate in the development of China’s AI, while also getting rid of the limitations of being passively exposed to traditional businesses such as e-commerce and food delivery when investing in the eight major technology conglomerates.” Chen Mingkang, a senior equity strategy analyst at Bloomberg Industry Research, told a reporter that the large language models developed by these two companies rank among the top globally, and their products have demonstrated strength in both performance and cost efficiency.

Chen Mingkang believes that since the listing of the domestic AI large-model “twin stars,” their share prices have risen by several multiples. This may reflect the market’s optimistic expectations about the prospects for early adoption, as well as demand for high-performance, cost-effective models.

The market expects that the year-over-year revenue growth rates for MiniMax and Zhipu AI will both exceed 150%, far above the single-digit high-growth level of China’s eight major technology conglomerates, and also above the low double-digit growth level of the United States’ seven major technology conglomerates.

If the story of AI’s new forces is “from 0 to 1,” then what traditional internet giants are staging is a “jump from 1 to 10.”

This year, capital expenditure plans from long-established tech giants such as Tencent, Alibaba, and Baidu have repeatedly drawn market attention. To compete for AI entry points, these players with strong cash flow are willing to spend heavily on AI compute infrastructure and model research and development.

“Old players are meeting a new windfall. Increasing capital expenditure must mean they are bullish about the development of the industry track. Using new business models alongside existing ones to achieve geometric growth is a pattern that traditional internet giants either choose proactively or find themselves unable to avoid.” Wang Xinjie, chief investment strategist at Standard Chartered China Wealth Solutions, told a reporter that in the current industry business models, all kinds of application scenarios will certainly emerge. For internet giants, increasing capital expenditure followed by quickly monetizing is the direction for development right now.

Chen Mingkang predicts that cloud service giants control full-stack AI infrastructure and may also roll out service bundling strategies, forming advantages in pricing and profit margins; competition in the industry may intensify further in the future.

As AI agents such as OpenClaw spark the “application deployment” wave, internet giants are speeding up their shift from “model competitions” to “securing positioning in scenarios.” Tencent quickly launched WorkBuddy, an AI agent for all scenarios that supports OpenClaw skills; Alibaba Cloud launched dedicated image services; and JD Cloud is piloting in business lines such as retail and logistics. These veteran players with massive user ecosystems are trying to open up new growth space through OpenClaw.

This collective push to embrace OpenClaw quickly generated feedback in capital markets. Tencent became one of the biggest winners. On March 10, one day after WorkBuddy’s launch, the stock surged at the open; during the day, the intraday high reached HKD 556; the full-day gain was 7.27%, setting a near-3-month record for the largest single-day increase. Trading value exceeded HKD 30 billion, and market expectations for its AI commercialization deployment were fully ignited. Meanwhile, Alibaba, which belongs to Alibaba Group, did not see a single-day blowout, but thanks to the historical highest growth pace achieved by its 百炼 MaaS business as Token consumption surged, its share price steadily rebounded during the March mid-month technology stock rebound wave, confirming market recognition of its Token business layout. After the release of news that JD Group is piloting in retail and logistics scenarios, its stock price also broke away from the prior sideways trend: from March 12 to March 16, it closed higher for five consecutive trading days. Hong Kong stock prices rose gradually from HKD 108.6 to HKD 111.5.

When OpenClaw pushes AI from “Q&A interaction” to “task execution,” Token consumption driven by each individual task grows exponentially. Tokens are no longer only a unit of compute measurement at the technical level; they have become the core pricing unit for measuring the depth of AI business deployment and commercialization potential. This is precisely the key logic starting point for the valuation anchor of AI to migrate toward Tokens.

Hard Currency of the Digital Economy Era

In March 2026, at NVIDIA’s GTC conference, the company’s founder and CEO Huang Renxun introduced a new concept to reshape industry understanding—“Token Factory.” In his view, data centers are undergoing a fundamental role shift: they are no longer just the “electronic warehouse” of the past used to store files and data, but an intelligent production line running day and night. This factory takes in power and data and produces Tokens.

As Huang Renxun said, Token has become “hard currency” in the digital economy era, and its generation efficiency will directly determine the survival ability and revenue curve of technology companies.

What is Token?

In simple terms, Token is the smallest unit of measurement for each interaction between a user and a model: every word and every punctuation mark you input to the AI, as well as every word the model replies with, will be broken down and consume a certain number of Tokens.

For investors, Token has become a focal point because it is the first time it brings together the revenue end and the cost end of AI companies into the same quantifiable scale.

“For the AI sector, traditional PE valuation is not very effective. For this kind of potentially high-growth industry, investors mainly focus on its future potential profitability.” Wang Xinjie told a reporter.

Wang Xinjie pointed out that at the current deployment stage, heavy capital expenditures may make it difficult for these companies to obtain profitability immediately. Therefore, in the AI industry, the user traffic and Token consumption numbers at this stage are what investors care about more, because they can increase the probability of future profitability.

For example, when analysts issue valuation reports for AI companies, they start to look not only at traditional income statements, but also try to estimate the share of Token consumption by the company in the global large-model subscription and API markets, as well as the gross profit that can be achieved per unit of Token.

Behind this shift in valuation logic is the market’s re-understanding of the essence of AI companies’ business models: growth in Token consumption means higher user activity and deeper business penetration, while an improvement in monetization capability per Token indicates that the business model is moving toward maturity.

Specifically, native large-model companies win high valuations thanks to their technology foundation, because investors believe their Token consumption has the potential to grow exponentially. Meanwhile, traditional platform giants with massive scenario coverage are granted higher monetization imagination because existing business channels such as e-commerce recommendations, food delivery dispatching, and content distribution naturally provide pathways to convert Token calls into actual revenue.

Yang Delong, chief economist at Qianhai Open-Source Fund, told a reporter that in the AI era, Hong Kong stocks place more emphasis on new indicators such as customer traffic. Changes in this valuation logic will also have a major impact on how technology stocks in Hong Kong are priced. Investors will value technology companies’ advantages in Tokens and big data more.

When Token Becomes the New Unit of Pricing

Under this trend, Token is becoming the new valuation anchor for AI companies. Song Weiwei, manager of the China Europe Zhongzheng Robot Index Fund, told a reporter that OpenClaw represents a paradigm shift from AI “conversation” to “execution,” and Token consumption is the first principle behind the entire investment chain.

For pure AI companies, the business model is highly dependent on Token consumption. The underlying logic for which directions are expected to benefit from this multiplier effect is also derived from this.

The commercial data of Hong Kong’s AI “twin stars” is the best evidence. MiniMax, driven by a surge in its M2 series text models, saw its average daily Token consumption jump more than 6x compared with December 2025, driving a year-on-year revenue surge of 158.9% for all of 2025 to USD 790.4 million. Its gross margin also jumped significantly from 12.2% to 25.4%, demonstrating outstanding commercialization efficiency.

Zhipu AI, meanwhile, has relied on the rapid commercialization of its GLM-5 flagship model API services, with Token consumption continuing to climb—average daily Token consumption reached 4.2 trillion in November 2025, driving multiple growth in 2025 revenue. Its MaaS platform has brought together more than 3 million enterprises and application developers. API prices were cumulatively raised by 83% in the first quarter of 2026, and the trend of both volume and price increasing is clearly visible.

A shared characteristic of these two companies is that Token call volume has become a direct revenue indicator, and performance delivery drives the valuation shift from “concept-driven” to “revenue-driven.”

Duansh?—Y? (The rest is not provided)

Duan Nairong, senior investment strategist at the Royal Bank of Canada’s wealth management division, told a reporter that currently, most of the “AI new forces” are not yet profitable, so the market is paying more attention to their potential for dynamic growth.

Duan Nairong believes that under a “Token inflation” backdrop, the value anchor for Hong Kong tech stocks may be assessed with reference to the following dimensions. First is the conversion efficiency from Token consumption to revenue—how much revenue can be generated per unit of Token—which determines the true profit-making ability of the business. Second is the monetization rate of traffic and customer stickiness, including metrics such as customer retention rate and paid renewal rate, which reflect the sustainability of revenue. Third is the lock-in capability of prepayment models, meaning the extent to which companies lock in future revenue through prepayment and subscription models, which can better reflect business stability.

It is worth noting that the performance and cost-performance advantages of domestic AI models are supporting Token competitiveness, and Token pricing and cost advantages are also one of the foundations for constructing valuation premiums.

According to data on the third-party API aggregation platform OpenRouter, in late February 2026, API Token usage using China’s AI models surpassed that of U.S. AI models for the first time.

A new research report from Goldman Sachs that the reporter obtained indicates that the performance gap between China’s new generation AI models and U.S. models has narrowed significantly. It supports an ultra-long context window of up to 1 million Tokens and strong agent capabilities. Second, the Token pricing of China’s AI models is only 5%-10% of that of U.S. flagship models, making cost-effectiveness more prominent. Finally, China’s AI models perform excellently in multimodal domains such as text, video, and audio, further expanding Token application scenarios.

In the research report, Goldman Sachs analysts believe that surging Token demand will bring revenue breakthroughs for China’s AI model companies, while also driving growth in cloud services revenue.

In terms of the competitive landscape, independent AI model companies are expected to rise quickly by leveraging the Token business, competing with internet giants. Head internet companies, through capital expenditure investment, will also consolidate their advantages in Token-related infrastructure.

A research report from Bloomberg Industry Research states that when companies such as MiniMax and Zhipu AI saw their share prices soar, in essence, the valuation recognition by the market for their “low token cost + high demand growth rate” combination enabled their market caps to rapidly catch up with technology giants such as JD.com and Baidu.

For investors, they may need to accept a reality: in an AI industry that iterates quickly, perhaps there is no eternal valuation anchor. Today’s Token may be replaced tomorrow by a new unit of measurement; today’s leader may be overturned tomorrow by a crossover player.

What can be confirmed is that Hong Kong stocks are becoming the core stage for docking China’s AI industry with global capital. On this stage, the story is just beginning.

(Zhang Weize also contributed to the interview for this article.)