In the first half of 2026, China's AIoT market officially broke through the trillion-yuan mark, up 26.3% year-on-year. What does this number mean? The average number of IoT terminals deployed per enterprise grew from 320 in 2024 to 680, more than doubling. But what is more noteworthy is the structural change: IoT platforms that purely do device connectivity and data display saw their market share drop by 12 percentage points over the past 18 months, while platforms with edge AI inference capabilities and digital twin engines are eating into the former's share. IoT platforms are undergoing a technological leap from "connection tools" to "intelligent foundations."
Five years ago, the core proposition of the IoT industry was "how to get devices connected." The standardization of the MQTT protocol, the maturity of NB-IoT and Cat.1 modules, and the refinement of cloud vendors' IoT Core services have turned device onboarding from a technical challenge into a standard operation. Today, any IoT platform can connect tens of thousands of devices without it being anything worth bragging about.
The real bottleneck has shifted to the data closed loop: can the data generated by devices be processed in real time? Can the processing results drive business decisions? Can decisions be automatically issued back to terminals for execution? These three question marks form the new capability pyramid for IoT platforms.
Take a typical smart community solution as an example. A medium-scale smart community typically deploys 300-500 terminals—access control, barriers, cameras, environmental sensors, energy consumption monitors, fire protection equipment. If the IoT platform only does data aggregation, what the management backend sees is a bunch of charts and alarm lists, and the property staff's operating logic is still "see a problem → make a phone call → wait for repairs." But if the platform has data closed-loop capabilities, the situation is completely different: access control detects abnormal tailgating behavior, the edge gateway analyzes the video stream in real time and triggers an audible and visual warning, while writing the event into the property work order system to automatically dispatch a ticket—the entire process changes from "people looking for issues" to "issues finding people."
Industry data:According to IDC's Q2 2026 report, IoT platforms equipped with edge AI inference modules reduced average alarm response time from 15.8 minutes previously to 2.1 minutes, with false alarm rates dropping by about 40%. In the three scenarios of smart parks, smart communities, and smart manufacturing, the deployment penetration rates of edge AI reached 47%, 32%, and 58%, respectively.The architectural evolution of IoT platforms is essentially a process of computing power sinking step by step from the cloud to the edge. This sinking has already occurred in three rounds:
The first round was the sinking of protocol adaptation. Early IoT platforms placed MQQT Broker, CoAP gateways, and Modbus conversion entirely in the cloud, and device data had to be transmitted over the public network before it could be seen on the platform. Around 2019, edge gateways began to take on protocol conversion work, and the cloud was only responsible for managing a unified data model. This step compressed network latency from seconds to milliseconds.
The second round was the sinking of the rules engine. From 2022 to 2024, mainstream IoT platforms successively deployed alarm rules engines to the edge side. This meant that even with the network cable unplugged and 4G disconnected, edge nodes could still independently execute local automation policies such as "temperature exceeds threshold → close valve." This step solved the reliability problem.
The third round is happening now: the sinking of AI inference. Lightweight large models (such as 1-3B parameter vision models and time-series forecasting models) are beginning to run on edge computing boxes. An edge AI device priced at 2,000 yuan can already complete tasks such as license plate recognition, facial access control, and abnormal behavior detection without being connected to the internet. This means the application development logic of IoT platforms is changing—developers no longer just write data dashboards and alarm rules, but need to design a complete MLOps pipeline of "model deployment → data labeling → incremental training → model update."
This also explains why, over the past year, vendors with both IoT platform capabilities and AI platform capabilities have grown the fastest. Pure device connectivity platforms are easily replaced by cloud vendors' standard IoT Core products, but the "IoT + AI" combination barrier is much more solid.
Digital twins have taken a detour in the IoT field. From 2019 to 2022, the industry built a large number of "3D visualization big screens"—they looked cool, but aside from catching the eye when visiting customers came to tour, actual operations staff never used them. The fundamental problem was: the visuals were exquisite but the data was not real-time, or the data was real-time but there was a lack of actionable entry points.
Starting in the second half of 2025, digital twins entered a new stage: transitioning from a "display tool" to an "operation tool." There are two technology drivers: first, the implementation of WebGPU means browser-side 3D rendering no longer lags; second, the interaction interface between the IoT platform's rules engine and the digital twin model has been standardized. Today's digital twin console can do this: click on a device in the 3D scene, directly pop up its real-time data panel, historical operating curves, and work order records, and issue remote control commands in the same interface.
In industrial scenarios, the after-sales service system of a certain equipment manufacturer already uses digital twins as a standard tool. After customer service receives a repair call, there is no need to ask the customer "what color is that light"—they directly locate the device in the digital twin interface, check real-time sensor data, perform remote diagnosis, and even issue parameter calibration commands. This system compressed average fault handling time from 4.2 hours to 47 minutes.
The key to digital twins changing from a gimmick into a productivity tool is not how realistic the 3D rendering is, but whether it can connect with work order systems, asset management systems, and alarm systems. This point is especially important in small and medium-sized enterprises'IoT solutionsselection—do not be fooled by cool 3D demos; look at whether the digital twin has real operation entry points and data closed-loop capabilities.
IoT platforms face a structural contradiction: scenarios are extremely fragmented, but platforms must provide standardized product capabilities. A smart community project and a factory equipment monitoring project appear on the surface to both be using an IoT platform, but the former's core needs are people flow analysis, access control linkage, and property work orders, while the latter's core needs are equipment OEE calculation, predictive maintenance, and MES system integration. This difference is so large that it is almost like building two different products.
There is already a clear solution direction in the industry: the platform provides PaaS-layer capabilities (device access, rules engine, data storage, AI inference services), and the upper-layer business logic is completed by implementation teams through low-code tools. The key lies in the degree of API abstraction at the PaaS layer—good abstraction allows implementation engineers to complete the setup of a smart community scenario within two weeks, while poor abstraction can stretch the implementation cycle indefinitely. Currently in thesoftware developmentfield, the API maturity of IoT PaaS platforms is becoming the primary evaluation criterion for enterprise selection.
Technical indicators:Industry evaluations show that the average API call latency of leading IoT PaaS platforms (Alibaba Cloud IoT, Huawei IoT, Xiaomi IoT, etc.) has dropped to within 120ms, device shadow synchronization latency is below 500ms, and the device capacity of a single instance reaches the million level. However, differences among platforms in edge AI inference, digital twin integration, and third-party system integration remain significant.Based on the current pace of technological evolution and market demand, IoT platforms have three certain trends in the second half of 2026:
First,edge AI will move from optional to standard. The maturity of lightweight model deployment tools and the continued decline in edge computing costs (currently an edge box supporting 4TOPS inference has a retail price of less than 1,500 yuan) have brought the cost-performance ratio of IoT+AI combo products to a critical point. It is expected that by the end of 2026, more than 60% of new IoT projects will require edge AI capabilities.
Second,IoT platforms will accelerate integration with low-code platforms. In the past, implementing an IoT project required three roles: device engineers, backend developers, and frontend developers. Now, through low-code tools to orchestrate device data flows + page building + rule configuration, one implementation engineer can complete 80% of the work. This trend is having a direct impact on human resource allocation in the software development industry—demand for basic CRUD-type development engineers is declining, while demand for composite talent who understand both IoT protocols and low-code component development is rising.
Third,smart communities will become an important incremental scenario for IoT platforms. The digital penetration rate of the property management industry is still less than 30%. Cities such as Shenzhen, Hangzhou, and Chengdu have already issued smart community construction standards, requiring newly built residential communities to be equipped with infrastructure such as smart access control, energy consumption monitoring, and fire protection IoT. For IoT platforms, this is a multi-billion-yuan incremental market.
Technology itself is not the end goal. Using technological means to transform device data from the physical world into actionable business decisions, thereby reducing operating costs and improving management efficiency—this is the fundamental driving force behind the evolution of IoT platforms.
Room 1409, Building C, Shenzhen Century Industrial Center, No. 743 Zhoushi Road, Hezhou Community, Hangcheng Sub-district, Bao'an District, Shenzhen
