【Trend Report】 Domestic AI Large Models Breakthroughs During the Spring Festival Box Office; Token Globalization Narrative Expected to Become a New Trend

A-shares today opened with a strong start in the Year of the Horse. By the close, the Shanghai Composite rose 0.87%, the Shenzhen Component gained 1.36%, and the ChiNext Index increased 0.99%. The combined trading volume of the Shanghai, Shenzhen, and Beijing markets exceeded 2.2 trillion yuan, an increase of over 200 billion compared to the previous trading day. Most industry sectors closed higher, with precious metals, oil and petrochemicals, glass and fiberglass, agrochemical products, chemical raw materials, and non-metallic materials leading gains. The sectors with the biggest declines were film and cinema chains, media, tourism and scenic spots, and software development. In individual stocks, over 4,000 stocks rose, with more than a hundred hitting the daily limit-up.

In the 2026 Spring Festival season, domestic large-scale AI models experienced a collective breakout, initiating a technological race. Breakthroughs were achieved across model performance, intelligent agents, and ecosystem deployment, transitioning from follow-up to parallel development. Zhispur released GLM-5, with programming capabilities comparable to Claude Opus 4.5, outperforming Gemini3 Pro, and completed adaptation for multiple domestic computing platforms. MiniMax launched M2.5, setting industry records for programming tests and significantly improving cost efficiency. ByteDance launched the full series of Doubao Large Models 2.0, with multiple general-purpose agent models comparable to top international standards, and upgraded audio-visual creation models simultaneously. Moon’s Dark Side Kimi K2.5 utilized Agent Swarm technology for multi-agent parallel collaboration, added AI agent service beta testing during the Spring Festival, expanding productivity scenarios. Alibaba open-sourced Qwen 3.5-Plus on New Year’s Eve, with an innovative architecture that greatly reduces deployment costs; API pricing is only 1/18 of similar international offerings, achieving dual breakthroughs in user scale and technological influence through ecosystem collaboration. Tencent released the image model 3.0, and the Yuanbao app built a differentiated AI social ecosystem, with a significant increase in function calls.

Meanwhile, China’s large models are capturing the global developer market. According to the latest weekly data from OpenRouter, the top ten models on the platform account for approximately 8.7 trillion tokens, with Chinese models dominating at 5.3 trillion tokens, accounting for 61%. MiniMax M2.5 leads with 2.45 trillion tokens, followed by Kimi K2.5 with 1.21 trillion. Zhispur’s GLM 5 and DeepSeek V3.2 rank third and fifth, respectively. Additionally, according to Frost & Sullivan’s report “China GenAI Market Insights: Enterprise Large Model Call Panorama Study, H2 2025,” the daily call volume of enterprise large models in China surged to 37 trillion tokens in the second half of 2025, a 263% increase from 10.2 trillion in the first half. Leading models saw increased market share, with Alibaba Cloud’s Qwen growing the most, rising to 32.1% from 17.7%, nearly doubling and expanding its lead.

Huatai Securities states that tokens are the purest derivative of electricity. When U.S. users call Zhispur/DeepSeek APIs, data travels via Pacific submarine cables to Chinese data centers, where GPUs consume Chinese electricity to perform inference—electricity never leaves China, but the value is transferred cross-border through tokens. JPMorgan estimates that China’s AI inference token consumption will grow from about 10 quintillion in 2025 to approximately 3.9 sextillion in 2030, a 370-fold increase over five years.

Huatai Securities: China’s Low Electricity Prices Are Transforming into Global AI Service Pricing Power

As AI develops rapidly in 2025, cloud computing demand continues to expand, with AI training and inference significantly boosting general computing power. Tokens are the purest derivative of electricity. When U.S. users call Zhispur/DeepSeek APIs, data travels via Pacific submarine cables to Chinese data centers, where Chinese electricity powers GPU inference—electricity remains within China, but the value is transferred cross-border through tokens. Electricity and computing power account for over 70% of token costs. China’s low electricity prices are turning into global AI service pricing power.

JPMorgan: Token Consumption to Enter Several Years of Rapid Growth

The AI application promotion during the Spring Festival is just the surface. The deeper trend is that consumers’ methods of obtaining information and consuming content are undergoing a structural transformation, which in turn drives inference volume and will propel token consumption into several years of rapid growth. JPMorgan forecasts that China’s AI inference token consumption will grow from about 10 quintillion in 2025 to approximately 3.9 sextillion in 2030, a 370-fold increase over five years.

Changjiang Securities: The True Explosion of Token Volume Expected in 2026

As programming and multimodal models mature, downstream applications are expected to open up, creating high demand for quality tokens. Based on overseas AI industry development patterns, there is about a two-year lag from capital expenditure to token demand explosion. Domestic tech giants’ AI capital spending cycles lag about a year behind overseas, starting in late 2024. Therefore, domestic cloud providers’ revenues have begun to grow, and the true explosion of token volume is expected in 2026.

Guolian Minsheng: Cloud Computing and Supporting Service Providers May See Valuation Reconfiguration

With rapidly increasing token demand, price hikes are gradually passing from upstream to CPUs and cloud services. Global cloud leader AWS has begun raising prices, breaking industry norms. Cloud and supporting service providers may see valuation restructuring. Key stocks to watch include: 1) Cloud computing: Alibaba, Kingsoft Cloud, UCloud, Capital Online, Sangfor, Hongjing Technology, Shunwang Technology, Wangsu Science & Technology, QingCloud; 2) CPUs: Hygon Information, Great Wall, Loongson Zhongke, Heshen New Materials; 3) Databases: HeteroDB, Damo Data, Massive Data, Taiji Shares, SuperMap Software, Tuoersi, etc.

Kainuo Securities: Global AI May Continue to Resonate

Looking ahead to 2026, the “siphon effect” of AI will be prominent, and global AI may continue to resonate. Overseas, giants like Google and Meta are continuously raising AI capital expenditure guidance, with models like Google Gemini significantly increasing token consumption, highlighting a positive AI cycle. Domestically, leading companies such as ByteDance, Alibaba, and Tencent are entering a phase of large-scale AI computing power investment.

Dongguan Securities: Domestic Large Models Are Expected to Accelerate Token Consumption Growth

As domestic models improve in programming and intelligent agent capabilities, their usage will increase substantially. Domestic large models in programming and agents are expected to further accelerate application deployment, boosting token consumption.

(This article does not constitute investment advice. Investors operate at their own risk. The market carries risks; please invest cautiously.)

(Source: Oriental Wealth Research Center)