This paper presents an Artificial Intelligence of Things (AIoT) system featuring an LSTM recurrent neural network ASIC and the I3C serial communication protocol for temperature monitoring and forecasting. The proposed system employs the I3C protocol as its primary communication method and integrates an EEPROM memory chip to store both sensor-collected data and prediction results. Additionally, the LSTM recurrent neural network utilizes hardware-based Sigmoid and Tanh activation functions for efficient computation. Prediction results are stored in the EEPROM and subsequently uploaded to the CockroachDB database. The computational core of the neural network is implemented using IEEE 754 single-precision floating-point arithmetic, enabling efficient computation with low energy consumption through specialized hardware design. FPGA verification was conducted, followed by design and implementation using TSMC’s 90-nm cell-based design process. Simulation results indicate a power consumption of 0.7459 mW, a gate count of 89,440, a chip area of approximately 0.8984 × 0.8972 mm2, and a throughput of 8.88 Mbps. A key contribution of this work is the compression technique, which is applicable to mobile devices and achieves small measurement errors. The compression rate, approximately 8.4, significantly reduces data quantity while maintaining performance.