According to a research report released by China Merchants Securities Co.,Ltd. (ASX: 600999), in the multi-model era, Token operations are expected to become an important infrastructure connecting model supply with AI applications.

As the number of foundation models continues to increase and different models keep diverging in capability, cost, speed and stability, it is becoming the norm for downstream developers to invoke multiple models and multiple providers at the same time.

Token operation platforms aggregate models through a unified API and further provide services such as model selection, intelligent routing, Token metering, unified settlement and enterprise governance, with their role evolving from a simple model access tool into an AI traffic distribution and enterprise invocation control layer.

CMSC believes that as the scale of Token calls and usage scenarios continue to expand, the value of Token operations is expected to rise further. The main viewpoints of China Merchants Securities Co.,Ltd. (ASX: 600999) are as follows:

Industry space: Token calls are growing rapidly, and the Token operations market is still in the early stage of penetration

Gartner estimates that global spending on the Token operations market will grow from about US$251 million in 2026 to US$429 million in 2027, and is expected to reach about US$1 billion in 2028, corresponding to a three-year CAGR of nearly 100%. Currently only about 15% of software engineering leaders have deployed related platforms, 31% are in the process of deploying, and 22% are still in the evaluation stage, leaving substantial room for industry penetration to rise.

CMSC estimates that in 2026 the global GMV of model calls carried by Token operation platforms will be about US$4.3 billion to US$6 billion, accounting for roughly 15% to 21% of global generative AI model spending. Domestic demand is also expanding rapidly, with China's full-year Token call volume in 2025 at about 21,100 trillion, and the average daily call volume further exceeding 140 trillion in March 2026.

Business model: Asset-light characteristics are obvious, with traffic scale, routing data and enterprise stickiness building core barriers

Token operators usually do not directly bear foundation model training and large-scale GPU capital expenditures, and currently mainly adopt charging models such as commissions, subscriptions and "subscription plus Credits." Compared with the relatively low technical threshold of a unified API itself, the platform's real competitive advantage comes from the positive cycle formed after large-scale operations: the richer the models and providers, the more developers can be attracted; as Token call scale rises, the continuous accumulation of call data helps optimize model selection and routing efficiency, and enterprise functions such as budgeting, permissions, logs, security and governance further raise customer migration costs.

Overseas case: OpenRouter has rapidly grown from a model aggregator into an AI traffic distribution gateway

OpenRouter connects different models and providers through a unified API and has gradually established a two-tier system of model routing and provider routing. As of September 2026, the platform had connected more than 500 active models and more than 80 providers, served more than 10 million developers and enterprises, and processed more than 10 trillion Tokens per day. The platform adopts Credits for unified settlement and mainly charges a 5.5% platform service fee. As more users leave model and provider selection to the platform, what OpenRouter controls is no longer just the model access gateway, but participation in the actual allocation of Token traffic among different models and inference service providers.

Domestic layout: Extending from model access to Token supply and application scenarios

Tencent and SiliconFlow focus on model aggregation, inference and Token supply; companies such as Yeahka, Marketingforce and Jihong shares enter from marketing, enterprise services and cross-border e-commerce scenarios, embedding model calls, Token metering and Agent scheduling into business processes. Rongxin Culture relies on content and IP resources to explore Tokenized calls of vertical knowledge assets, while Paradigm Intelligence extends from enterprise AI infrastructure to model services, API calls and Agent platforms.

CMSC believes that domestic Token operations have gradually developed from single model access toward multiple paths of "infrastructure plus enterprise applications plus vertical content," and as Agents and enterprise AI applications penetrate deeper into business processes, the importance of Token metering, scheduling and scenario-based monetization capabilities is expected to rise further.

Risk warning: AI application and Token call growth may fall short of expectations; demand for multi-model calls and intelligent routing may fall short of expectations; industry competition may intensify; commercialization progress may fall short of expectations; data security and compliance risks.