88Z AI, 19 November — 36Kr Research reports that tokens have become the basic unit of measurement in the intelligent era. By June 2026, China's daily token calls exceeded 500 trillion, forming a bipolar structure with the US. Midstream margins remain low, while upstream non-GPU accelerator shipments surpassed GPUs for the first time, with Huawei Ascend's share rising to 28.5%.
At a glance
- China's daily token calls exceeded 500 trillion in June 2026, forming a global bipolar structure with the US
- Non-GPU accelerator shipments surpassed GPUs for the first time, with Huawei Ascend's share rising to 28.5%
- The industry chain shows a thick-end, thin-middle pattern, with midstream MaaS vendors' gross margins generally below 25%
- Tokens have officially become a state-recognized unit of measurement for intelligent services, promoting AI measurability and traceability

Tokens as the New Standard for Intelligent Economy
As artificial intelligence moves from technological breakthroughs to scaled applications, the intelligent economy has become a national strategy. In August 2025, the State Council issued opinions aiming for over 70% adoption of new intelligent terminals and agents by 2027. The 2026 Government Work Report further prioritized agents alongside terminals. In this context, the token, as the smallest semantic unit processed by large models, has established itself as the foundational, measurable, priceworthy, and tradable unit of the intelligent era. Similar to how electricity is measured in kilowatt-hours and data in gigabytes, tokens provide a unified standard for AI service billing. In March 2026, the Chinese National Committee for Terms in Sciences and Technologies officially named Token as Ciyuan (Word Element). By May 2026, relevant guidelines were issued to ensure AI services are measurable, comparable, and traceable, accelerating the formation of the token economy.
Explosion in Call Volume and US-China Bipolarity
Driven by policy support, open-source iterations, and cost reductions from inference optimization, China's token call volume has entered a phase of explosive growth. Data shows that daily national token calls surged from approximately 100 billion in early 2024 to 100 trillion by the end of 2025, exceeding 140 trillion in March 2026, and breaking 500 trillion by June 2026. Model-as-a-Service (MaaS) has become the primary supply mode, with IDC predicting China's MaaS token call scale to reach approximately 40,000 trillion in 2026. Globally, a bipolar structure between China and the US has formed. OpenRouter data indicates that Chinese model weekly calls first surpassed US models in February 2026, reaching a 63.5% share in the last 28 days by late July 2026. Stanford University's report notes that as of March 2026, the Elo score gap between top Chinese and US models on the Arena leaderboard was only 2.7%, indicating closely matched technical capabilities.
Industry Chain Shifts Focus to Efficiency and Value Conversion
China's token economy industry chain centers on midstream production and services, coordinating upstream computing power and downstream applications. On the upstream side, shipments of non-GPU accelerators in China reached 1.016 million units in the first half of 2026, surpassing GPUs for the first time, with Huawei Ascend's share rising to 28.5%. The competitive focus has shifted to enhancing effective token production capacity. As the distribution hub, the midstream sector prioritizes inference efficiency, with domestic models achieving an average output speed of 78.23 tokens per second in August 2026, roughly double that of 2025. Downstream applications are upgrading from single-point assistance to agent-based task execution. However, the value chain exhibits a thick-end, thin-middle characteristic: upstream Cambricon and downstream Kingsoft Office reported gross margins of 56.53% and 84.76% respectively in the first three quarters of 2026, while midstream players Zhipu MaaS and MiniMax recorded margins of only 24.6% and 17.9%, indicating significant room for profitability improvement in the midstream.
88Z perspective The rise of the token economy marks the AI industry's shift from a technology race to a stage of refined operations. A daily call volume of 500 trillion indicates that AI has deeply penetrated production environments, yet the lower gross margins of midstream vendors reveal the profitability challenges facing the current MaaS model. In the future, the key to competition will no longer be just computing scale, but how to reduce unit token costs through inference optimization and how to extract higher added value in agent applications. For practitioners, focusing on the adaptation efficiency of non-GPU computing power and extending into high-margin downstream application scenarios will be key strategies to cope with industry involution.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 36kr.comSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.