In May 2026, China consumed an average of 140 trillion tokens per day. Two years ago, this number was almost zero; today it has surpassed that of any other country in the world.
QbitAI Think Tank quantified this phenomenon in the 2026 China AI Application Landscape Report released at the Fourth China AIGC Industry Summit. Monthly web visits to domestic AI applications exceeded 900 million, monthly app downloads surpassed 240 million, and daily active users surged 223% year-on-year.
For enterprise CIOs, the signal is clear: whether or not you approve, your employees are already using AI.
Among the five major trends, three are directly related to ERP
The report identifies five major trends, and I will discuss the three trends most closely related to enterprise digitalization.
140 trillion: China's daily Token call volume in 2026, growing more than a thousandfold in two years (QbitAI Think Tank)
First, agentification.AI has evolved from 'answering questions' to 'executing tasks'. According to estimates in the report, the token consumption of a single agent action is one hundred times that of traditional AI applications. This is why token consumption has surged—not because people are chatting more, but because AI is actually working behind the scenes and is performing complex tasks involving multiple steps and tool calls.
For ERP vendors, this means that the positioning of an “AI assistant” is already outdated. The “autonomous enterprise” that SAP promoted at its 2026 Sapphire conference follows exactly this logic—AI does not help you query data, but directly executes the financial closing process. The Odoo community is also discussing similar approaches, embedding AI agents into business processes rather than hanging them in the sidebar.
Second, become the entry point.Around this year's Spring Festival, ByteDance, Alibaba, Tencent, and Baidu collectively invested more than 4.5 billion yuan to compete over a very specific question: when users need AI, which platform do they open first.
This affects enterprise procurement decisions. In the past, CIOs selected ERP based on feature lists and price. Now they also need to consider: who controls the main entry point for employees to use AI within that system? If the everyday AI entry point for all employees is a large model in DingTalk or WeCom, then the ERP vendor's AI module may never be opened.
Third, commercialization.In less than 20 days after Kimi K2.5 was released, its revenue exceeded that of the entire year of 2025. After Zhipu API raised its price by 83%, call volume did not decrease but increased. This shows one thing: when AI is truly embedded into workflows, users are willing to pay, and price sensitivity is not as high as imagined.
The traditional approach of enterprise software—"first get users hooked with free access, then figure out how to monetize"—may not work in the AI era. Enterprise customers are already willing to pay for valuable AI capabilities, provided that these capabilities can genuinely save labor costs.
Apple is also about to launch
Apple officially announced that WWDC 2026 will be held on June 9, with the core highlight being the standalone app restructuring of Siri. This is not just a mobile feature update; it also has an impact on enterprise digitalization.
This time, Siri is using Google's Gemini model, but all requests run on Apple's own private cloud, so Google cannot access this data. This architecture is worth noting—it solves one of the thorniest problems in enterprise AI deployment: large models are very powerful, but data cannot leak out.
If this Apple system is released alongside iOS 27 this fall, it means hundreds of millions of devices will come with a hybrid AI entry point combining on-device and private cloud. The new question facing enterprise IT departments is: should the AI on employees' phones be integrated with the AI in the company's ERP? Can it be integrated? How should data compliance be calculated?
It is still too early to draw conclusions, but the direction is already clear: the AI entry point will move down from browsers and chat windows to the operating system level. Microsoft is already doing this (Copilot embedded in Windows and M365), and it is only to be expected that Apple will follow.
140 trillion every day, but what do businesses say about the data?
The surge in token consumption shows that AI usage has indeed increased. However, usage does not equal value. In the same period, KPMG's "2026 Global Industrial Manufacturing Technology Report" provided another set of data: 49% of manufacturing senior executives have implemented AI and created business value, but 76% admitted that unreliable data is the main risk.
Putting these two data points together can present a clearer picture:
| Size | Consumer/Developer side | Enterprise implementation side |
|---|---|---|
| Core indicators | Token consumption, DAU, downloads | Whether commercial value is generated, data quality |
| 2026 Status | Explosive growth (140 trillion per day) | 49% implemented, but 76% are stuck on data |
| bottleneck | Computing power costs, the battle for entry points | Data silos, chaotic standards, and weak cleansing capabilities |
| Next step | Monetization and vertical deepening | Data governance precedes AI modeling |
For those working in enterprise digitalization, this comparison is very useful. The consumer AI approach is to burn money to seize entry points, iterate quickly, and tolerate errors; the enterprise AI approach must prioritize data, start with small scenarios, and expand only after calculating ROI.
These two sets of logic cannot be conflated. Applying consumer internet strategies to enterprise AI often causes problems during project delivery.
Practical Advice for CIOs
Do not use “whether employees use AI” as a primary KPI.They are very likely already using it—they just haven’t told you. The real KPIs should be: which business processes have become faster because of AI, by how much faster, and how many person-days were saved.
First, take stock of the data, then discuss AI strategy. KPMG data shows that 76% of manufacturing companies are stuck with unreliable data. This statistic also applies to service and trading companies. If 30% of customer addresses in the ERP are wrong, then any AI prediction trained on this data is garbage in, garbage out.
Seize the internal AI gateway. Gateway-oriented strategies don't only apply to the consumer side. Within the enterprise, where is the main gateway through which employees use AI every day? If it's a third-party mini program within WeCom, then the money you invested in ERP AI modules may be wasted. Rather than waiting for vendors to sort it out for you, figure this out now.
Start with small scenarios and avoid large, all-encompassing ones.The QbitAI report mentions that the tools with the highest retention rates are intelligent assistant tools, but those that create the deepest business value are vertical scenarios. These two are not the same thing. When choosing the first scenario for an enterprise AI project, rather than choosing a broad scenario that 'the whole company can use' but whose value is unclear, it is better to choose a narrow scenario where the cost-reduction account can be calculated clearly.
140 trillion tokens is a signal, not the final result. The rules of the game in enterprise digitalization are not the same as those in the consumer internet. Understanding this difference is more important than chasing trends.
QuantumBit Think Tank "2026 China AI Application Panorama Report" (2026-05-20), Apple WWDC 2026 Official Announcement (2026-05-20), KPMG "2026 Global Industrial Manufacturing Technology Report" (2026-05-26), IBM Think 2026 (2026-05-07)
