Bringing Google TPUs to Oracle Cloud Infrastructure (cloud.google.com)

Oracle Cloud Infrastructure (OCI) will soon host Google Tensor Processing Units (TPUs), making it the first hyperscaler outside Google to offer these advanced chips. This partnership aims to meet the growing demand for efficient AI compute options, allowing organizations to benefit from TPUs' high performance in training and inference workloads. Set to roll out in 2027, the integration promises to optimize throughput and reduce costs for enterprises, enabling them to run complex AI models more effectively. The significance of this collaboration lies in TPUs’ co-designed architecture, which integrates silicon, software, and advanced cooling techniques to achieve unprecedented performance metrics. The latest TPU generation delivers up to 12.6 PFLOPS and offers efficient goodput, ensuring enhanced resource utilization during training. OCI's data centers, designed for AI workload density, combined with Oracle’s secure Acceleron networking, position this offering as a game changer. By enabling direct assignment of workloads to the most appropriate computing resource—TPUs or GPUs—teams can manage their AI processes more effectively, fostering innovation across the tech ecosystem and ultimately accelerating advancements in AI/ML development.