The official Chinese naming of "Token" as "词元" has sparked an investment boom in "Token Economics" within capital markets. During the recent China Development Forum 2026, National Data Bureau Director Liu Liehong used "词元" as the Chinese translation for Token in his speech.
Director Liu revealed that China's average daily Token calls reached 100 billion in early 2024, surged to 100 trillion by the end of 2025, and exceeded 140 trillion this March - representing over a thousand-fold growth within two years. Liu emphasized that Token serves not only as the value anchor in the intelligent era but also as the "settlement unit" connecting technological supply and commercial demand, providing quantifiable possibilities for business model implementation.
The industry has emitted strong signals as well. Last week, NVIDIA CEO Jensen Huang introduced the concept of "Token Factory Economics" at the GTC conference, stating that Token will become the new commodity of the AI era. Future data centers will transform into factories producing Tokens, with performance-per-watt becoming the core competitive advantage for commercial realization.
The Token concept has ignited capital market enthusiasm. Investors are focusing on how to capture opportunities in this "Token Economics" wave. Recently, multiple securities firms have published research reports exploring investment themes in Token Economics, with computing infrastructure, model globalization, and computing-power coordination emerging as frequently mentioned investment lines.
From technical concept to market focus, why has Token gained such prominence? Technically, Token represents the smallest information unit processed by large models, segmenting natural language text into AI-understandable components. Commercially, it serves as the measurement unit for AI computing costs, significantly influencing AI service pricing.
The frequent appearances of this dual-nature concept reflect profound transformations in AI industry business logic. Demand-side explosion appears most evident, with the phenomenal success of AI agent frameworks like OpenClaw driving rapid Token demand expansion. Data from third-party AI model platform OpenRouter shows that during March 9-15, 2026, OpenClaw contributed 20% of platform Token consumption, equivalent to 60% of the platform's average weekly consumption during Q4 2025.
Price changes have already commenced quietly. Since early 2026, computing power rental markets have entered a price increase cycle. By February-end, high-end GPU rentals including NVIDIA H200 and H100 models saw 15%-30% month-over-month price increases. In domestic markets, Token demand expansion has driven collective price increases across model layer and cloud service providers. Major model companies like Zhipu and cloud service providers including Alibaba Cloud and Baidu Cloud have recently announced price adjustments for AI computing products.
Xiangcai Securities computer industry analyst Zhou Cheng describes this trend as "volume and price rising simultaneously," noting that non-linear Token demand growth in the AI Agent era has disrupted computing power supply-demand balance, directly affecting procurement costs for upstream components including GPUs, enterprise storage, and CPUs. Sandwiched between rigid downstream demand and upstream hardware cost inflation, cloud computing industry pricing logic is shifting toward premium realization.
Multiple institutions believe supporting factors may persist in the short term. CITIC Securities computer team led by Ying Ying judges that with OpenClaw driving higher-frequency inference requests and longer context requirements, future cloud resource utilization will further improve. Demand explosion and upstream cost transmission are expected to continue pushing cloud service prices upward.
Kaiyuan Securities communications chief analyst Jiang Ying suggests that AI application proliferation combined with OpenClaw frameworks may trigger inference demand explosion. Coupled with NVIDIA production constraints, rising hardware costs, and domestic supply gaps, markets are entering a "seller's market" where price increases may continue.
Long-term perspectives indicate industry consensus that Token market prosperity represents not short-term pulses but inevitable trends under AI application popularization. At GTC, Jensen Huang proclaimed "Token is King," envisioning future data centers as Token production factories where performance-per-watt becomes the core commercialization competency. Traditional architecture designs centered on server quantity and storage capacity will gradually yield to new architectures focusing on Token generation rates and energy efficiency ratios.
Regarding commercial implementation, Huang believes Token will become a new commodity, eventually featuring tiered pricing based on speed and intelligence levels - from free tiers to ultra-high-speed tiers (approximately $150 per million Tokens) - creating broader commercialization space for inference scenarios.
From investment perspectives, Guolian Minsheng Securities computer chief analyst Lü Wei indicates that Token demand "inflation" may become this year's core AI investment theme, with related opportunities likely revolving around inference Token demand. Multiple securities firms have identified relevant beneficiary sectors in recent research reports.
Computing infrastructure and hardware segments stand as most direct beneficiaries from Token call volume surges, representing high-consensus sectors among institutions. Kaiyuan Securities' Jiang Ying categorizes "Token Factory" core themes into AIDC (AI Data Centers), computing power leasing, and CDN (Content Delivery Networks). Jiang views "Token = AI chips (domestic computing power + computing leasing) = AIDC," while anticipating substantial CDN demand growth alongside continuous Token expansion.
Focusing on these three themes, Jiang identifies five noteworthy sub-sectors: AIDC facilities, AIDC liquid cooling, AIDC power supply, CDN, and AIDC computing and networking.
CITIC Securities TMT communications chief Yan Guicheng's team maintains that "short-term fluctuations don't alter long-term growth logic in computing sectors," continuously recommending AI computing industry chain participants including GPU/CPU, optical modules, optical chips, liquid cooling, and fiber optic cables.
Beyond fundamental computing infrastructure, major model manufacturers as upper-layer application entities may encounter new investment opportunities. Lü Wei notes that model manufacturers are transitioning to "selling Token fuel + selling outcomes." When inference consumption becomes production material, model manufacturers can transform "computing scarcity" into gross margins and cash flow through tiered pricing and subscription products.
Notably, domestic models demonstrate strong competitiveness under this logic, with "Token globalization" frequently mentioned in broker reports. Shenwan Hongyuan Securities computer team led by Huang Zhonghuang calculates that domestic models show extreme cost-effectiveness compared to overseas counterparts, with comprehensive costs at approximately 1/6 to 1/10 of foreign models. This advantage stems from architectural improvements brought by innovations like DeepSeek, particularly MLA and sparse architectures that significantly reduce inference costs.
Lü Wei recommends continued focus on quality model manufacturers, suggesting that companies maintaining subscription retention and enterprise seat expansion in high-ROI scenarios like programming, Agent, and enterprise processes - while converting "Token usage" into "labor/time/rework savings" delivery value - possess capabilities to withstand open-source competition and price wars.
Concurrently, Lü mentions "AI firewall" targets deserve attention. As enterprises integrate AI into workflows, risks like data leaks and agent over-authorization may drive "AI security/governance platforms" into essential requirements.
Additionally, "computing-power coordination" is considered crucial industrial advantage supporting "Token globalization." Soochow Securities computer team believes green power hubs effectively reduce electricity costs, making low-cost power core competitiveness for Token globalization. The new digital trade form of "power staying domestic while computing value crosses borders" is becoming China's core barrier in global AI competition.
The team further identifies four valuable target types in computing-power coordination: traditional power companies transitioning through energy advantages to build data centers; green power operators providing long-term renewable energy for computing clusters; scheduling software service providers using algorithm models for real-time load-price matching; and power engineering leaders leveraging ultra-high voltage and source-grid-load-storage experience. These four categories collectively construct the "energy-computing" closed loop.
Comments