JPMorgan noted that behind the Spring Festival AI promotion campaign lies a shift in user habits toward chatbots, driving token consumption into a long-term high growth phase. It is forecasted that China's token consumption will achieve a compound annual growth rate of 330%. The investment focus should be on "second-order winners": AI infrastructure providers that benefit directly and companies related to advertising businesses.
JPMorgan pointed out that while major technology giants launched an intense promotion battle for AI applications during the Spring Festival, this is only a surface phenomenon. Beyond the short-term game of 'who will win the AI application war,' the market should focus on deeper trends: structural changes are occurring in how consumers access information and consume content, which will drive Token consumption into a multi-year high-growth phase.
During the Spring Festival in 2026, $TENCENT (00700.HK)$ 、 $Alibaba (BABA.US)$ and $Baidu (BIDU.US)$ will compete for user mindshare by offering tens of billions of yuan in subsidies. On February 6, JPMorgan, in its latest research report on China’s internet and artificial intelligence industries, analyzed that these large internet platforms are investing heavily in promotions to accelerate the shift of users from traditional search engines to chatbots, while cultivating commercial habits around intelligent agents. This process has effectively increased inference demand, thereby accelerating the consumption rate of Tokens. The current GPU computing power demand, whether excessive or insufficient, is expected to persist in the coming years.
The bank forecasts that China’s Token consumption will grow at a compound annual growth rate (CAGR) of approximately 330% between 2025 and 2030. Based on this, JPMorgan believes it is still too early to determine the winners in the AI application space. More certain investment opportunities lie with 'second-order beneficiaries,' including direct beneficiaries of China’s AI infrastructure (such as Alibaba and Baidu) and companies exposed to online advertising businesses (such as Kuaishou). For investors, China’s AI infrastructure has become one of the most attractive investment themes in the TMT sector.
Giant Offensive During Spring Festival: Competing for User Habits, Not Just Traffic
Major technology companies will adopt markedly different strategies during the Spring Festival, with core objectives all aimed at capturing user entry points and mindshare in the AI era.
According to JPMorgan's observations, $TENCENT (00700.HK)$ has adopted a classic 'growth-first' strategy, investing 1 billion yuan in red envelope campaigns. By leveraging WeChat’s social ecosystem to lower user barriers, it aims to rapidly expand the installation and activation volume of 'Yuanbao' within a short period.
However, this campaign also exposed the boundaries of platform governance, as some incentivized sharing behaviors led to link restrictions. Analysts believe that Tencent may need to transition from link-driven viral spread to product-driven retention loops in order to integrate AI into the WeChat ecosystem in a smoother, experience-first manner.
Alibaba's strategy demonstrates structural differences. Alibaba announced a 3-billion-yuan 'Spring Festival Hospitality Plan,' integrating Taobao, Alipay, Fliggy, and AutoNavi Maps. JPMorgan noted that Alibaba’s goal is not only to increase adoption of 'Tongyi Qianwen' but also to foster commercial habits initiated via AI interfaces. The strategic objective is to shorten the consumer funnel and improve conversion efficiency; if results can be sustained post-holiday, Alibaba will strengthen its position in high-friction categories like travel planning.
In comparison, Baidu invested approximately 500 million yuan, continuing its integrated strategy of 'AI assistant + search distribution,' aiming to embed AI into search sessions with stronger intent. Although ByteDance did not announce large-scale cash subsidies, its DouBao series models reportedly reached 100 million daily active users, thanks to product iterations and distribution capabilities via platforms like Douyin. JPMorgan believes that even though monetization models remain uncertain, losing mindshare at the traffic-entry layer comes at a high cost—this is a key reason why tech giants are willing to 'buy time' through marketing.
Token Consumption Outlook: A Five-Year Infrastructure Boom with 370x Growth
JPMorgan provided a striking growth forecast in its report: China's AI inference token consumption is expected to surge from approximately 100 quadrillion in 2025 to around 3,900 quadrillion by 2030, representing a growth of about 370 times over five years.
This explosive growth is primarily driven by two factors: the increasing penetration of AI in both consumer and enterprise workloads, and the expansion of application scenarios from simple conversational AI to complex agents and multimodal outputs. Data indicates that as of December 2025, the daily average token consumption of ByteDance's DouBao series models has exceeded 50 trillion, marking a year-on-year increase of more than tenfold.
Structurally, the composition of inference demand will undergo significant changes. The share of conversational AI in total consumption is projected to decline from nearly half in 2025 to high single digits by 2030. It will be replaced by 'Knowledge Worker AI Agents' and 'Multimodal AI.' The former, involving multi-step reasoning and document workflows, is expected to account for one-third of total consumption by 2030. The latter, driven by increased workloads in images, audio, and video, will grow its share from the current 15% to nearly 40%.
Notably, JPMorgan cautioned investors that token growth does not linearly equate to an increase in chip demand. With next-generation hardware such as NVIDIA's Blackwell delivering improved inference performance (up to four times that of Hopper) and optimizations in software stacks, the number of tokens processed per GPU hour is also improving significantly.
Editor/Doris