Google and AMD explore hybrid TPU design

Google is reportedly working with AMD on a next-generation Tensor Processing Unit designed for AI workloads that blend accelerator performance with conventional computing. Industry reports say the companies are exploring a hybrid AI accelerator that could integrate CPU cores directly into the TPU processor package.

The design is intended in part for reinforcement learning and emerging agentic AI workloads, where accelerator and CPU resources may need to exchange data frequently. Google currently uses TPU processors in servers alongside standard X86 CPUs, but moving CPU and TPU resources into a single package could reduce the physical distance data must travel and potentially cut latency.

The effort would align with Google’s broader work on specialized AI hardware and AMD’s push into more integrated AI systems. AMD recently announced a partnership with Cerebras that combines AMD EPYC processors and Instinct accelerators with Cerebras’ wafer-scale AI processors. Analyst Jon Peddie said the proposed Google partnership makes sense because AMD has proven itself in custom, high-performance APUs and because Google needs supplier diversity.