Rising power consumption and dramatic thermal issues are becoming the bane of data centres as increasing computational loads and the adoption of Application-Specific Integrated Circuits (ASICs) as the leading high-performance processing methodology necessitate novel thermal management methods and techniques. ASIC operation at a fixed voltage is also a common cause of thermal hotspots, which reduce the device's reliability, increase the cooling cost, and lower the system's overall efficiency. The traditional methods of Dynamic Voltage Scaling (DVS), although helpful in reducing power, are generally not as thermally sensitive and responsive as they need to be to respond to the dynamic and highly complex thermal characteristics of actual data centre operational conditions. This paper presents an innovative AI-based thermal-conscious DVS framework implemented within ASIC-based designs to address these problems. The framework combines on-chip thermal sensors distributed throughout and a lightweight machine learning model, namely a recurrent neural network, capable of predicting the thermal states soon after the basis of real-time temperature sensors, workload measurement, and ambient data centre conditions. A dynamic AI controller is used to dynamically set the voltage and frequency settings at a fine granularity throughout the ASIC modules in a proactive manner, eliminating thermal hotspots and minimising power consumption without affecting performance. The proposed approach, as demonstrated by simulation results, achieves a power usage reduction of up to 25% and a nearly 30% reduction in peak chip temperature compared to conventional voltage scaling methods based on heuristic and static models. These enhancements result in reduced cooling overhead, improved device reliability, and extended ASIC life. The predictive control capability of the framework enables real-time, adaptive control of the voltage, representing an excellent improvement over reactive thermal management strategies. In general, the presented AI-based thermal-conscious DVS solution is a sustainable, efficient, and intelligent hardware-software co-design solution. Therefore, it is beneficial in next-generation data centres that aim to balance performance, energy efficiency, and thermal reliability in a more challenging computational environment.