'The next generation of AI will require a diverse range of compute': Oracle deal means Google TPUs are coming to a rival cloud for the first time

Oracle has announced it will be offering Google's Tensor Processing Units (TPUs) on its own cloud platform, marking the first time a rival cloud vendor will sell Google's specialized AI accelerators. This partnership comes just months after Google launched its eighth-generation TPUs, specifically designed for AI workloads.

Oracle's OCI platform will make Google's TPUs accessible to customers, allowing them to choose between Google's TPUs and Nvidia's GPUs for training and running AI models. This move is significant for Google's hardware, as it represents the first time a hyperscaler other than Google Cloud has sold its TPUs.

The partnership aims to provide customers with a diverse range of compute options, as AI workloads vary in their computational needs. Oracle's AI & Infrastructure SVP and Chief Technologist, Amin Vahdat, emphasized the importance of a diverse range of compute for the next generation of AI. OCI President Mahesh Thiagarajan explained that different AI workloads have different compute requirements, justifying the decision to offer another option to customers.

Google AI & Infrastructure SVP and Chief Technologist, Amin Vahdat, highlighted that the partnership will be launched with the eighth-generation TPUs - 8t for training and 8i for inference, both launched by Google Cloud earlier this year. These TPUs will become available via OCI in 2027.