By March 2026, China’s AI applications were processing 140 trillion tokens per day. That is a 1,000-fold increase from roughly 100 billion daily tokens in early 2024. The figure, cited by Wei Liang, deputy head of the China Academy of Information and Communications Technology, in a CCTV Finance program preview, is one of those numbers that sounds abstract until you realize what it implies: the Chinese AI market is not just growing. It is compounding.
Tokens are the basic unit of work in large language models. Every prompt, every query, every generated sentence, every tool invocation consumes them. A single chat message might use a few hundred. A long document summary might use thousands. An AI agent that plans, queries, reasons, and acts across multiple systems can use millions for a single user request. When you multiply that by hundreds of millions of users, billions of queries, and an expanding ecosystem of agents, you get a trillion-scale workload. China is now at 140 trillion daily tokens. That is not a vanity metric. It is a demand signal.
Agents Are the Multiplier
The CCTV report linked the surge to wider adoption of AI agents. That is the crucial detail. Agents do not just answer questions. They chain actions. A single user instruction can trigger a planning step, a memory retrieval, a tool call, a result validation, and a follow-up generation. Each step consumes tokens. A simple agent workflow might use ten to fifty times the tokens of a single chat turn. The result is that agent adoption does not just add users; it multiplies the work each user creates.
This explains the 1,000-fold increase better than raw user growth. China has plenty of internet users, but user growth alone does not produce a thousand-fold jump in eighteen months. Agentic workflows do. Chinese companies have been aggressive about embedding AI agents into search, e-commerce, productivity software, customer service, and cloud services. The token count is the bill for all that automation.
The report also noted the need for better token pricing and scheduling systems. That is a polite way of saying that demand is now large enough to require infrastructure specialization. When your daily workload is measured in hundreds of trillions of tokens, you cannot just rent GPUs from a cloud provider and hope for the best. You need routing, caching, batching, model distillation, and pricing models that reflect the actual cost of compute. The companies that build those systems will capture a lot of the value.
The Scale in Global Context
To put 140 trillion tokens in perspective, consider that OpenAI has reported its own systems process hundreds of billions of tokens daily. China’s daily figure is orders of magnitude larger. That does not mean Chinese models are better. It means they are being used more intensively, by more applications, in more integrated ways.
The gap is partly structural. China’s consumer internet giants — Alibaba, Baidu, ByteDance, Tencent — have embedded AI into search, shopping, video, payments, and messaging at scale. An American user might open ChatGPT for a specific task. A Chinese user might trigger AI dozens of times a day without leaving WeChat, Taobao, or Douyin. The model is not a separate product. It is the plumbing.
That integration has implications for global competition. Chinese companies are learning to operate AI at a scale that few American companies outside Google and Meta can match. They are building the operational experience, the cost curves, and the tooling for a token-dense future. The next generation of AI-native applications may be designed by people who have already debugged trillion-token pipelines.
What It Means for the Infrastructure Stack
Token volume on this scale is not free. It requires compute, memory, networking, and power. The 140 trillion figure is a demand signal for the entire Chinese AI supply chain. Domestic chipmakers like Huawei and Hygon, cloud providers like Alibaba Cloud and Tencent Cloud, and model-serving startups all benefit from the workload. It also puts pressure on the power grid and data-center buildout in ways that may become visible in the next few years.
The scheduling systems mentioned in the report are a clue about the next layer of competition. When demand exceeds supply, the company that can route the right query to the right model at the lowest cost wins. That is why Chinese labs are racing not just on model quality but on inference efficiency, quantization, speculative decoding, and model distillation. The model leaderboard gets the headlines. The serving stack gets the margin.
There is also a defensive dimension. The more tokens Chinese companies process domestically, the less dependent they are on foreign AI infrastructure. That matters when export controls are a constant threat. A domestic token economy of 140 trillion per day is not just a commercial achievement. It is a hedge against supply-chain disruption.
🔥 Hot Takes
1. The West is underestimating the operational learning that comes from token scale. American AI discourse is obsessed with model benchmarks. The Chinese market is learning how to serve AI at a scale that makes benchmarks almost irrelevant. The company that can run a trillion tokens a day profitably is building a different kind of advantage than the company that tops a leaderboard once.
2. Agents are the real cost explosion everyone saw coming but nobody priced for. A single agent can consume ten to fifty times the tokens of a simple chat query. Enterprise adoption of agents will turn modest AI budgets into massive compute bills. China’s numbers are a preview of what will happen everywhere as agents become normal.
3. 140 trillion tokens is a strategic asset, not just a usage statistic. Data flywheels matter. The more tokens Chinese companies process, the better they can optimize their models, serving systems, and pricing. That volume becomes a moat. It is also a signal to policymakers that Chinese AI is not a copycat industry anymore. It is a self-sustaining ecosystem.
The Bottom Line
China’s 140 trillion daily tokens are not proof that Chinese AI is superior. They are proof that Chinese AI is now operating at a scale that changes the competitive dynamics of the industry. The number captures the move from chatbots to agents, from experiments to infrastructure, and from copying to compounding.
For the rest of the world, the question is not whether to match that scale. The question is whether to build the systems that can handle it. The next phase of AI will be won by the countries and companies that can process intelligent workloads at massive volume, low cost, and high reliability. China is practicing that at 140 trillion tokens a day. Everyone else is still counting in billions.