According to AMD’s Chief Technology Officer Mark Papermaster, the artificial intelligence industry is poised for a significant transformation by 2025, with AI workloads shifting predominantly from training to inference. Currently, about 80% of AI computing focuses on training large language models, but this balance is expected to flip, with 80% moving toward inference tasks. This shift reflects the maturation of AI models and their increasing deployment in real-world applications. Papermaster emphasizes that this transition will drive substantial changes in hardware requirements and computing architectures. The move toward inference will push AI processing closer to edge devices, where data is generated and consumed, rather than relying solely on centralized data centers. This distributed approach aims to reduce latency and improve efficiency in AI applications. AMD is strategically positioning itself for this shift by developing specialized chips and accelerators optimized for inference workloads. The company’s adaptive computing solutions and MI300 accelerators are designed to handle both training and inference tasks effectively. The prediction aligns with broader industry trends toward more practical AI implementations and the growing demand for edge computing solutions. This transformation will likely impact how companies deploy AI resources and influence future hardware development strategies across the tech industry.