Oracle will offer Google's TPUs in its cloud by 2027, a first out of Google

Announced at Oracle AI World, the agreement makes OIC the first big cloud to rent Google's home chips, alongside NVIDIA, AMD and Intel accelerators.

Oracle and Google announced on October 27, at the Oracle AI World conference in Las Vegas, that Google's Tensor Processing Units will be available on Oracle Cloud Infrastructure from 2027, with a multi-year ramp-up. This is the first time that these Google-designed chips will be released by a major cloud provider other than Google itself.

The offer will focus on the eighth generation of TPUs, available in two architectures: the 8t TPU for training, which offers nearly three times more computing power per pod than the previous generation, and the 8i TPU for inference, designed for the workload of numerous and specialized agents. For Oracle, the interest is to offer several accelerator architectures under one roof, so that customers can choose according to the performance and cost of each workload. For Google, this is another step towards making its TPUs a real commercial alternative to NVIDIA's GPUs.