May 8, 2026, is destined to be written into the history of China's AIoT industry. On the same day, multiple national ministries and commissions will unveil two strategic milestones that will determine the next decade.
The first document is the "Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents," jointly issued by the Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology. For the first time, it defines intelligent agents as intelligent systems with autonomous perception, memory, decision-making, interaction, and execution capabilities at the national policy level, and proposes 19 typical application scenarios in areas such as scientific research, industrial development, boosting consumption, people's well-being, and social governance.
The second standard is the "Intelligent Classification of Artificial Intelligence Terminals" (GB/Z 177—2026) series of national standards jointly released by the Ministry of Industry and Information Technology, the State Administration for Market Regulation, the Ministry of Commerce and other departments. It establishes a four-level capability ladder from L1 response level to L4 collaborative level, initially covering seven major categories: mobile phones, computers, televisions, glasses, car cabins, speakers, and headphones.
Image source: Ministry of Industry and Information Technology WeChat account
The simultaneous release of the two documents is no coincidence. This represents a two-way effort at the policy level: intelligent agents are moving downwards to find physical carriers, while intelligent terminals are moving upwards to find the intelligent core. One document defines the main body of intelligent software, and the other defines the carrier of intelligent hardware, together forming a top-level design of "dual-track" approach.
This leads to the core judgment of this article: China is defining AIoT as a new type of infrastructure— intelligent infrastructure , which is of the same importance as electricity and the Internet were defined as infrastructure in the past.
This article will share three progressively deeper observations regarding this ongoing industry race:
What does double standard reveal (see the signal)?
What exactly does L4 mean (understanding patterns)?
What should AIoT companies do next (seize the window of opportunity)?
Dual-track benchmarking: A unique top-level design for AIoT globally
What was released on May 8th wasn't just two policies, but a dual-axis coordinate system: the "Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents" defines "spirit," while the "Intelligent Classification of Artificial Intelligence Terminals" defines "flesh." Understanding this coordinate system is crucial to understanding the next decade of Chinese-style AIoT.
The industrial implications of this design are threefold.
In the first layer, AI capabilities were transformed from conceptual terms into engineering metrics for the first time.
Over the past two years, the biggest pain points in the AIoT industry have been the overgeneralization of concepts, the accumulation of parameters, and the disconnect between publicity and user experience. The "Classification" standard uses a four-level capability ladder from L1 to L4 to transform intelligence from a vague adjective into a measurable, comparable, and verifiable product attribute. This is equivalent to issuing a unified "health check" to the entire industry, bidding farewell to pseudo-intelligence and parameter-driven competition, and providing a basis for evaluation.
In the second layer, intelligent agents are defined as product forms rather than application-layer added value.
The "Implementation Opinions" explicitly define intelligent agents as an important form of artificial intelligence products and services, and emphasize guiding companies in the fields of complete machines and software to develop products and services based on intelligent agents. The policy implications of these two sentences are extremely important. Intelligent agents are no longer functional modules dependent on hardware, but rather primary industrial entities on par with PCs and mobile phones. This represents a redefinition of the power structure of the entire AIoT industry chain.
The third layer consists of the drafting unit, which is essentially a strategic deployment plan for the industry.
Among the main drafting units of the "Classification" standard, industry players include Huawei, Honor, Xiaomi, OPPO, vivo, Lenovo, Unisoc, and others—all hardware giants. Meanwhile, the implementation path of the "Implementation Opinions" simultaneously involves large model manufacturers, open-source communities, chip manufacturers, and operating system manufacturers. This means that the key bargaining points in the AIoT industry chain over the next five years will arise at the intersection of two areas: how hardware players become carriers of intelligent agents, and how intelligent agents penetrate hardware operating systems.
From a global perspective, the scarcity of this dual-track standard becomes even clearer.
The United States takes a market-driven approach, neither defining what an intelligent agent is nor classifying the capabilities of AI terminals, leaving it entirely to the leading companies such as OpenAI, Anthropic, Apple, and Google to compete at the product level; the European Union takes a risk-based regulatory approach, with its AI Act regulating only the risk level of the application and not touching on the product form; Japan and South Korea follow the corporate ecosystem.
China has chosen a third path, establishing a coordinate system for both the software core and the hardware carrier using national standards. This approach of simultaneously setting standards for both software and hardware is unique in the global AI policy landscape of the same period.
The most compelling historical example is the dual-credit policy for new energy vehicles. Released in 2017 and implemented in 2018, the dual-credit policy, seemingly just a technical industry management measure, simultaneously tied the production and sales targets for new energy vehicles to the fuel consumption targets for gasoline vehicles. By setting standards on one hand and creating pressure on the other, it directly reshaped the competitive landscape of the entire Chinese automotive industry. Over the past decade, China's production and sales of new energy vehicles have ranked first globally for many consecutive years, transforming it from an industry follower to a global leader.
The dual standards for AIoT announced on May 8th are highly similar to the dual-credit policy in their policy design philosophy. Both use a combination of simultaneous hardware and software initiatives, along with a focus on both capabilities and direction, to leverage a trillion-dollar industry's overall leap forward. The difference is that this time, it's not a single industry that's being leveraged, but rather a new type of infrastructure.
Intelligent Mobility for All: L4 is Rewriting the Value Anchor of AIoT
In the four-level capability hierarchy outlined in the "Intelligent Classification of Artificial Intelligence Terminals," the L4 collaborative level is deliberately left blank. The standard explicitly states that it will be further clarified and improved in subsequent revisions based on the level of industry development. What appears to be a technical omission is actually a clear acknowledgment by policymakers: L4 is not yet clearly visible, but it will inevitably arrive.
This unclear level is precisely the biggest variable for the future of the entire AIoT industry.
Looking back at the value evolution path of AIoT, a clear curve can be drawn.
The core value of the IoT 1.0 era is connectivity ; device networking brings data feedback and remote control.
The core value of the AIoT 2.0 era is cognition ; devices possess local AI capabilities, enabling them to identify, judge, and respond.
The core value of the AIoT 3.0 era is assistance , corresponding to the transition from L2 to L3. Devices possess multimodal understanding and contextual judgment, upgrading from passive tools to proactive suggestions. This is the current position of AI PCs and AI phones.
The core value of the AIoT 4.0 era is collaboration , corresponding to L4, where devices become the user's alter ego in the physical world, proactively sensing scenarios, collaborating across devices, and autonomously executing tasks.
I summarize the endpoint of this curve in four words: Intelligent Mobility for All Things.
The Internet of Everything describes the story of the past decade, where the relationship between devices is one of connection; the Internet of Everything describes the script for the next decade, where the relationship between devices acting on behalf of users is one of agency.
The disruptive nature of L4 lies not in being smarter, but in fundamentally rewriting the relationship between users and devices, transforming them from operating tools into delegated agents.
This paradigm shift is happening simultaneously on both the consumer (C) and business (B) sides, but in different forms.
The shift for consumers (C-end) is from operating tools to delegating and acting as agents.
The product logic from L1 to L3 is to sell hardware and offer smart features as a bonus, while the product logic of L4 is to sell agency capabilities, with hardware merely serving as an access point. The "Classification" standard explicitly mentions in its highest-level capability points that it is necessary to rely on personal big models and knowledge bases to achieve autonomous learning and continuous evolution of terminals. This means that whoever masters the user's personal big model masters the user's long-term value.
Lenovo launched Tianxi AI Personal Intelligent Agent, and Huawei continues to upgrade Xiaoyi to an agent-like model. Essentially, both are positioning themselves early in the L4 space.
The power in the industry chain will shift from terminal brands to intelligent agent service providers, and the business model will evolve from one-time hardware sales to a three-element structure of hardware entry points, capability subscriptions, and data assets.
The transformation for B-end businesses is from data dashboards to autonomous execution.
The Industrial Internet of the past decade has primarily addressed connectivity and visibility; sensors collect data, transmit it to the cloud to generate dashboards, but decision-making and execution still rely on humans. The introduction of intelligent agents fundamentally reverses this logic.
The "Implementation Opinions" explicitly propose the development of intelligent agents for production management, dynamically optimizing production scheduling, resource allocation, and process connections. It also proposes promoting the integration of intelligent agents with CNC machine tools, industrial robots, and automated production lines. Coupled with the forward-looking deployments in cutting-edge fields such as multi-agent collaboration and the intelligent internet outlined in the "Implementation Opinions," future smart factories will no longer be assembly lines, but rather an intelligent agent society composed of scheduling agents, quality inspection agents, and logistics agents. These agents will autonomously negotiate, dynamically allocate resources, and collaboratively complete complex tasks.
The value focus of B-end businesses is shifting entirely from data collection and PaaS platforms to "intelligent agents as services" in vertical industries.
The transformations of C-end and B-end differ, but they share the same singularity logic: manufacturers that cross L4 define the rules of intelligent agents and occupy the value center; those that cannot cross can only become the execution end of the rules of intelligent agents and become value channels.
This scenario has already been foreshadowed once in history, right next door in the automotive industry. Before the emergence of L0 to L5 autonomous driving levels, intelligent driving was just a concept, with each company claiming to be more intelligent; after the levels were introduced, industry order, product positioning, consumer expectations, and the division of responsibilities were all rewritten, and capital flows shifted from fragmentation to a highly concentrated focus around Level L.
Today's AIoT is repeating the same script, only this time the stage covers all device forms.
Based on this assessment, two clear industry predictions can be made: In the next 12 to 18 months, the first batch of L3-level national standard certified products will be launched, and L-level will gradually replace computing power TOPS and parameter quantity, becoming the core benchmark for the next generation of AIoT products; In the next 18 to 24 months, L4 reference implementations will appear in the flagship products of leading manufacturers, and personal intelligent agents will move from concept to large-scale production.
L4 is not just a technology level; it is the singularity of the AIoT industry.
Four Key Breakthroughs: The 18-Month Window for AIoT Companies to Position Themselves
my country's chosen dual-track approach of establishing standards and driving development through specific scenarios has opened a unique strategic window for domestic AIoT companies, but this window may only be valid for 18 to 24 months.
The key to understanding this path is to see that it is composed of three superimposed maps.
The capability map is a classification of terminals according to their capabilities from L1 to L4, which serves as a benchmark for the supply side.
The risk map is a classification and grading governance framework clearly defined in the "Implementation Opinions". For sensitive areas and key industries, the Cyberspace Administration and the industry authorities will jointly determine open scenarios and implement management measures such as filing, testing, and recall of problematic products. For low-risk areas such as life and entertainment and daily office work, efficient governance will be achieved through compliance self-testing, information reporting, distribution platform management, and industry self-regulation. This is the boundary of the demand side.
The directional map includes 19 typical application scenarios plus subsidies for trade-in of old consumer goods, which is a driving force from the industry side.
The superposition of the three maps signifies that the country has clearly defined the boundaries of the rules of the game, and the remaining tracks are open for companies to compete on.
The uncertainty of the US path lies in market competition, the uncertainty of the EU path lies in the scale of regulation, and the certainty of the Chinese path lies in the clear policy direction; companies only need to decide where to position themselves. This represents a paradigm shift from seeking opportunities amidst policy uncertainty to securing a position within a framework of policy certainty.
Next, all AIoT companies will be forced to choose one of three tracks.
The first track is that of standard definers, who participate in the drafting of national standards and the formulation of protocols to incorporate their own technical roadmap into national standards. This path has high barriers to entry but also strong competitive advantages, making it suitable for leading hardware manufacturers, large model companies, and chip manufacturers.
The second track is scenario integrator, focusing on providing in-depth "AIoT intelligent agent as a service" around 19 typical scenarios. The entry barrier is moderate, and success depends on the depth of industry know-how. This is the most realistic track for medium-sized enterprises and the track with the greatest potential to become unicorns.
The third track is that of infrastructure builders, developing intelligent agent frameworks, toolchains, open-source licenses, intelligent agent software stores, and other infrastructure. This track has a lower barrier to entry but requires a long-term perspective, making it suitable for platform startups and core contribution teams in open-source communities.
The most dangerous position is that of companies caught between three tracks, neither participating in standards, specializing in specific scenarios, nor building a foundation, but only making generalized AI-added products. These companies will face the greatest survival pressure in the next two years.
Once the track has been selected, there are four common tactical levers that should be immediately incorporated into the strategic plan for the next 18 to 24 months, which I summarize as the "Four Leverages" strategy.
The first benefit is the benchmark. The Level 3 national standard is essentially a powerful endorsement from policy for companies. Manufacturers that first achieve Level 3 and then strive for Level 4 will receive triple benefits: preferential consumer subsidies, priority in government procurement, and premium pricing for consumers. For leading manufacturers, the next competition will be about the speed of achieving Level 4 benchmarks; for small and medium-sized manufacturers, the real opportunity lies in becoming the first Level 3 benchmark in a specific product category, such as the first Level 3 AI glasses or the first Level 3 AI home appliance. Rather than competing across all seven product categories, it's better to become a Level 3 benchmark in a niche category.
The second aspect is leveraging existing platforms. These 19 typical scenarios are not just policy slogans, but a targeted roadmap for subsidies, pilot programs, and prioritized procurement over the next three years. Among these, the integration of intelligent manufacturing, intelligent agents, and CNC machine tools/industrial robots is the most certain, because China's manufacturing data and application infrastructure are globally leading. The most crucial understanding is: it's better to be among the top three in one scenario than to be among the top ten in ten scenarios.
The third aspect is leveraging existing resources. The "Implementation Opinions" explicitly promote compatibility and adaptation between intelligent agents and open-source chips, open-source operating systems, and open-source large-scale models, which is tantamount to issuing a collective cost-reduction voucher to AIoT entrepreneurs. However, a deeper difference in understanding lies in the fact that using open source is about cost reduction, while creating open source is about securing a position. The status of a contributor is an order of magnitude more valuable than that of a user. Medium-sized and larger enterprises should contribute to open source in return for gaining ecosystem dominance.
The fourth approach is leveraging existing strengths. Protocol ecosystems are becoming a new battleground for global AIoT competition, with Anthropic's MCP, Google's A2A, and ANP and ACP already forming the leading international players. Chinese AIoT companies need a two-pronged approach: one leg outward, actively participating in international protocol communities and securing a leading position; the other leg inward, verifying protocols through China's advantageous scenarios such as the Industrial Internet and smart homes, and then exporting international standards in reverse.
In conclusion
The double standards of May 8th are not the end of the policy, but the starting gun for a decade-long industrial race.
Looking back at the history of China's telecommunications industry, from a blank slate in 1G, to following in 2G, to running alongside in 3G/4G, and finally leading in 5G, it has taken thirty years to achieve a reversal in the power of standards. Today, the L-level plus protocol ecosystem of the AIoT industry has the opportunity to complete a larger-scale leap in a shorter time. The protagonists of this leap are not the state, but enterprises.
The nation has laid out the track, marked the starting line, and fired the starting gun. The only remaining question is: as a company, which track should we be on, and how should we start?
The Internet of Everything is the story of the past decade; the Intelligent Mobility of Everything is the script for the next decade.
This article is from the WeChat Official Account "IoT Think Tank" (ID: iot101), author: Peng Zhao
