A striking figure emerged just before the May Day holiday: in March 2026, the national average daily Token usage surpassed 140 trillion. What does this number signify? It signifies that in just two years, the national average daily Token usage has grown by over 1,000 times. This is not linear growth; it is an exponential explosion. At the 9th Digital China Summit, Liu Liehong, head of the National Data Administration, used this data to outline a clear industrial logic: a new form of intelligent economy, with the Token as its smallest unit of measurement, is rapidly taking shape. For sectors like home appliances and other smart hardware, this is heavyweight news that demands attention.
**Token Economy: A New Value Measurement System is Being Established** To understand the AI transformation of home appliances, one must first clarify the concept of a Token. A Token is the smallest unit of information processed by a large language model. When you speak a sentence to an AI and it replies with a paragraph, every word, every symbol, and every command parsed consists of multiple tokens. Every inference call made by a large model consumes Tokens. Liu Liehong's core assertion is that Tokens make intelligent services measurable, priceable, and tradable. The industrial implications of this assertion are profound. If intelligence can be precisely measured, it means it can become a commodity. Home appliances are no longer just hardware sold on an outright purchase basis; manufacturers can charge users ongoing fees for AI services—the food recognition algorithm in a refrigerator, the energy consumption optimization model in an air conditioner, the path planning capability of a robotic vacuum cleaner—each can become a subscription unit. If intelligence can be priced, it means it can enter financial statements. The value of AI capabilities is no longer intangible but can be quantified as a clear revenue stream. If intelligence can be traded, it means a completely new industrial ecosystem is forming—data factor markets, AI model service providers, intelligent agent operators—and home appliances are positioned right at the forefront of this value chain. This is the Token economy of the AI era. Unlike the traffic economy, traffic is passive, while Tokens are active; traffic measures "how many times a user viewed something," whereas Tokens measure "how many intelligent decisions the AI processed."
**Inference Surpasses Training: A Critical Turning Point Has Arrived** Another set of data revealed by Liu Liehong at the summit contains an easily overlooked but crucial signal: in 2025, the volume of AI inference data surpassed that of training data for the first time, reaching 101.34 exabytes (EB). Understanding this shift requires clarifying the difference between training and inference: * **Training:** This is the "learning" process for large models, requiring massive data and repeated computations. It is time-consuming and resource-intensive, typically completed in data centers, and represents a one-time cost. * **Inference:** This is the "thinking" process for large models, where a user issues a command, and the model processes it in real-time and returns a result. It represents a continuous, ongoing consumption of resources. The fact that inference data volume has surpassed training data volume signifies that AI is transitioning from the "training era" to the "application era." In the training era, AI's value was primarily realized by model developers—those who could train the best models won. In the application era, AI's value is realized in scenario implementation—those who can apply inference capabilities to real-world scenarios will win. This is precisely the home appliance industry's main battleground. How many food recognition inferences does a smart refrigerator need to perform each day? How many local model calls are required for a single fresh-keeping mode adjustment? If inference becomes mainstream, the AI computing power of end-user devices, the capabilities of on-device models, and real-time response speeds transition from being optional extras to becoming the decisive factors for success.
**A Historic Shift in Data Structure: AI is "Creating" Data** The national total data production volume in 2025 was 52.26 zettabytes (ZB), of which data generated by AI accounted for 26.92 ZB, exceeding the volume of IoT sensor data for the first time. The significance of this change cannot be overstated. In the past, data generated by home appliances and IoT devices was primarily "passively recorded"—temperature sensors recorded temperature changes, humidity sensors recorded humidity changes, often with minimal human awareness. The value of this data was limited and highly fragmented. Now, AI is "actively generating" high-quality data. The model inference process itself creates data. Every AI decision, every piece of AI-generated content, every behavioral trajectory of an intelligent agent produces structured data that can be used for the next round of training. This implies that the core competitiveness of future home appliances will shift from "hardware manufacturing capability" to "data closed-loop capability." Whose robotic vacuum cleaner can accumulate more real-world home cleaning data will see its path planning model become more effective. Whose refrigerator can accumulate more data on food items and user behavior will see its fresh-keeping AI become more accurate. Once this positive feedback loop is established, it becomes a formidable competitive moat.
**High-Quality Datasets: Industrial Manufacturing is a Key Focus Area** Liu Liehong mentioned that by the end of 2025, China had established over 100,000 high-quality datasets, with a total scale exceeding 890 petabytes (PB), covering key sectors such as healthcare, industry, and manufacturing. The inclusion of manufacturing data into the high-quality dataset system signifies two key things for the home appliance industry: First, the AI upgrade of the home appliance manufacturing sector now has a data foundation. If quality inspection data, assembly process data, and energy consumption data from factories can be standardized, annotated, and incorporated into high-quality datasets, it means that home appliance manufacturers' manufacturing AI capabilities can iterate more rapidly and at lower cost. Leading companies like Midea, Haier, and Gree, which are already advancing smart factories, are direct beneficiaries in this round of data system construction. Second, data collection at the product end now has policy backing. The national push for the development of data factor markets means that the collection, cleansing, and circulation of home appliance usage data—under user authorization—now has an institutional framework. This removes some of the data compliance obstacles for home appliance manufacturers building a "product + subscription AI service" business model.
**For Home Appliance Companies: A Choice Concerning the Future** Returning to the home appliance industry itself, the opportunities revealed by this data are real, but the challenges are equally formidable. The opportunity lies in this: as the Token economy takes shape, the value chain of home appliances is being redistributed. The hardware gross margin of a refrigerator might be 20%, but for a refrigerator equipped with a subscription-based AI fresh-keeping service, the marginal cost of the AI service component approaches zero, potentially allowing gross margins exceeding 80%. Home appliance manufacturers have the opportunity to shift from "selling hardware" to "selling intelligent services." The challenge lies in this: the prerequisite for this path is a large user base and a data closed-loop. Without a sufficiently large installed base of devices, there will not be enough real-world usage data. Without enough real data, sufficiently good local AI models cannot be trained. Without good AI models, subscription fees cannot be justified. This is a virtuous cycle, but also a challenging entry point. Therefore, the significance of this AI revolution for the home appliance industry is not that all major appliance makers will benefit equally. Rather, established home appliance giants that have already built large-scale user bases and IoT ecosystems are gaining a monopolistic advantage in the AI era. Companies like Haier, Midea, and Gree, with their hundreds of millions of installed devices, are likely the true players in the Token economy. Other small and medium-sized home appliance enterprises need to carefully consider their strategic response to the Token economy. The 140 trillion Tokens are proof that this transformation is already underway, not a preview of what is to come. The home appliance industry, with its massive manufacturing capacity across numerous product categories, stands at the forefront of the intelligent economy's value chain. This transformation will wait for no one—users are already casting their votes for the intelligent economy with over 140 trillion Token calls every day.
Comments