"Morpheme" ignites capital enthusiasm; brokerages capitalize on the "Morpheme Economy" as an investment mainline

With the official Chinese name “词元” (Ciyuan) for Token being introduced, the capital market has ignited a wave of investment in the “Ciyuan economy.”

At the recent China Development High-Level Forum 2026, Liu Liehong, Director of the National Bureau of Statistics, used “词元” as the Chinese translation for Token in his speech.

According to Liu Liehong, at the beginning of 2024, China’s daily Token call volume was 100 billion; by the end of 2025, it surged to 100 trillion; in March this year, it has already surpassed 140 trillion, marking an increase of over a thousand times in two years.

Liu Liehong stated that the Token “词元” is not only a value anchor in the intelligent era but also a “settlement unit” connecting technology supply and business demand, providing quantifiable possibilities for the implementation of business models.

The industry has also sent strong signals. Just last week, Nvidia CEO Jensen Huang proposed the concept of “Token Factory Economics” at the GTC conference, stating that Token will be the new commodity in the AI era, and future data centers will become factories for producing Tokens, with performance per watt becoming the core competitive advantage for commercial monetization.

The Token “词元” has also ignited capital enthusiasm. Under the “Ciyuan economy” boom, how to seize this round of opportunities has become the focus of investors. Recently, several brokerages have released research reports to explore the “Ciyuan economy,” highlighting several main lines frequently mentioned, such as computing power infrastructure, model export, and computing-electricity synergy.

Image source: Reporter Liang Yuanhao’s photo

From Technical Concept to Market Focus: Why is Token “词元” Booming?

What exactly is the Token “词元” that has sparked heated discussions in the capital market?

A Token is the smallest unit of information processed by large models. Technically, it segments natural language text into a “language” that AI can understand, facilitating model computation; commercially, it is a measurement unit for evaluating AI computing power costs, significantly influencing the pricing of AI services.

This concept, which combines technical and economic attributes, has frequently made headlines recently, reflecting a profound shift in the business logic of the AI industry.

The explosion on the demand side is the most intuitive. Recently, the phenomenal popularity of the AI agent framework represented by OpenClaw is seen as a direct driver of the rapid expansion of Token demand.

Data from the third-party AI model aggregation platform OpenRouter shows that during the week of March 9 to March 15, 2026, OpenClaw accounted for 20% of the Token consumption on the platform, with its weekly consumption equivalent to 60% of the average weekly Token consumption across the platform in the fourth quarter of 2025.

Changes in pricing have also quietly begun. Since the beginning of 2026, the computing power leasing market has entered a price increase cycle.

By the end of February, the rental prices of high-end GPUs like Nvidia’s H200 and H100 had increased by 15% to 30%. In the domestic market, the expansion of Token demand has also led to a collective price increase among various vendors from the model layer to the cloud service layer. Companies like Zhipu and cloud service providers like Alibaba Cloud and Baidu Cloud have recently announced price hikes for AI computing power and other products.

Zhongtai Securities’ computer industry analyst Zhou Cheng summarized this trend as “simultaneous increase in volume and price,” and further pointed out that the non-linear growth of Token demand in the AI Agent era has broken the balance of computing power supply and demand, directly triggering changes in procurement costs for core hardware such as upstream GPUs, enterprise storage, and CPUs. Under the dual pressure of rigid downstream demand and upstream hardware cost inflation, the pricing logic of the cloud computing industry is shifting towards premium monetization.

Regarding whether this round of price increases can be sustained, several institutions believe that the supporting factors may be difficult to reverse in the short term.

The team led by Chief Analyst Ying Ying from CITIC Jianan Securities predicts that with higher frequency inference requests and longer context demands brought by OpenClaw, the utilization rate of cloud resources will further increase, and the explosive demand along with upstream cost transmission is expected to continue pushing cloud service prices upward.

Jiang Ying, Chief Analyst of the telecommunications industry at KGI Securities, also mentioned that the widespread adoption of AI applications and the OpenClaw framework may ignite inference demand, compounded by Nvidia’s capacity constraints, rising hardware costs, and gaps in domestic autonomy, driving the market into a “seller’s market,” where the price increase trend may persist.

Looking at the longer term, the industry and institutions generally believe that the uplift in the Token market’s prosperity is not a short-term pulse but a trend driven by the widespread adoption of AI applications.

At the GTC conference, Jensen Huang shouted the slogan “Token is King.” In his view, future data centers will become factories for producing Tokens, with performance per watt becoming the core competitive advantage for commercial monetization. The traditional architecture design focused on server quantity and storage capacity will gradually give way to a new architecture centered on Token generation rate and energy efficiency.

Specifically for commercial implementation, Huang believes that Tokens will become new commodities, and once mature, they will be priced in tiers based on speed and intelligence level, from a free tier to a super-fast tier (around $150/million Tokens), opening up broader commercial space for inference scenarios.

Brokerages Exploring the “Ciyuan Economy”: Computing Power Infrastructure, Model Export, Computing-Electricity Synergy

From an investment perspective, Lu Wei, Chief Analyst of the computer industry at Guolian Minsheng Securities, points out that the “inflation” of Token demand is expected to become the core line of AI investment this year, with related investment opportunities likely to rapidly revolve around inference Token demand.

Under the “inflation” of Token demand, how to seize investment opportunities in the “Ciyuan economy”? Several brokerages have recently sorted out the relevant beneficiary sectors and targets in their research reports.

Specifically, the most direct beneficiaries of the surge in Token call volume are the computing power infrastructure and hardware sectors, which also have a high level of consensus among institutions.

Jiang Ying has classified the three core main lines of the “Token Factory” as AIDC (AI Data Centers), computing power leasing, and CDN (Content Delivery Network). In Jiang Ying’s view, “Token = AI chips (domestic computing power + computing power leasing) = AIDC.” Meanwhile, as Token continues to grow, CDN demand may also see significant growth.

Around these three main lines, Jiang Ying further identified five sub-sectors worth paying attention to, including AIDC data centers, AIDC liquid cooling, AIDC power supply, CDN, and AIDC computation and networking.

The team led by Yan Guicheng, Chief Analyst of TMT Communications at CITIC Jianan, also stated that “short-term fluctuations in the computing power sector do not change the long-term growth logic,” continuously recommending relevant targets in the AI computing power industry chain, including GPU/CPU, optical modules, optical chips, liquid cooling, and optical fiber cables across the computing power chain.

In addition to the underlying computing power infrastructure, the upper-level application entities of computing power resources, large model vendors, are also expected to welcome a new round of investment opportunities.

Lu Wei pointed out that large model vendors are shifting to “sell Token fuel + sell results.” When inference consumption becomes a production material, model vendors have the opportunity to transform “computing power scarcity” into gross profit and cash flow through tiered pricing and subscription products.

Notably, under this logic, domestic models are showing strong competitiveness, and “Token export” has been frequently mentioned in recent brokerage reports.

The computer industry team of Shenwan Hongyuan Securities, led by Huang Zhonghuang, estimates that domestic models exhibit extremely strong cost performance compared to overseas ones, with comprehensive costs being about 1/6 to 1/10 of overseas models. This cost performance advantage comes from improvements in model architecture brought by entities like DeepSeek, especially MLA and sparse architectures, which significantly reduce inference costs.

Lu Wei suggests that continuous attention should be paid to high-quality large model vendors in the future. In his view, those who can maintain subscription retention and enterprise seat expansion in high ROI scenarios such as programming, agents, and enterprise processes, and can convert “Token usage” into delivery value that “saves labor, time, and rework,” will have the capability to navigate through open source and price wars.

Simultaneously, Lu Wei also mentioned that “AI firewall” targets are worth paying attention to. As enterprises embed AI into workflows, risks such as data leakage and agent overreach are likely to drive the “AI security platform/governance platform” to become a necessity.

Moreover, “computing-electricity synergy” is also seen as one of the important industrial advantages supporting “Token export.”

The computer team at Dongwu Securities believes that green power hubs effectively reduce electricity costs, and low-cost electricity becomes a core competitive advantage for Token export. The new type of digital trade model of “electricity not leaving the country, computing power value crossing borders” is becoming a core barrier for Chinese AI to participate in global competition.

The team further pointed out that within the computing-electricity synergy track, there are four types of targets with core value: traditional power transformation enterprises that invest in data centers based on energy endowments, which have the most valuation uplift; green power operators relying on low-cost renewable energy to provide long-term green electricity supply for computing power clusters; scheduling software service providers that use algorithm models to achieve real-time matching of load and electricity prices, thereby improving operational efficiency; and leading power engineering firms that leverage ultra-high voltage and source-grid-load-storage construction experience to solidify the physical foundation for synergy. Together, these four elements form a closed loop of “energy—computing power.”