Oracle adds Google TPUs to OCI to expand enterprise AI compute options
Network World · · g5506956
Oracle has entered into a partnership with Google to bring the latter’s Tensor Processing Units (TPUs) to Oracle Cloud Infrastructure (OCI), giving its enterprise customers an alternative to GPUs for training and inference workloads without running their workloads on Google’s own cloud platform. The deal, which was announced at Oracle AI World in Las Vegas on Tuesday, marks the first time Google TPUs will be available through another hyperscaler, the companies said in a joint statement. As part of the agreement, Oracle will provide the data center infrastructure needed to deploy the TPUs, including power, cooling, and networking, they added. The addition of TPUs will expand Oracle customers’ compute portfolio as they look for alternatives to GPUs to manage the rising costs and infrastructure demands of training large AI models and running inference at scale, said Stephanie Walter , practice leader of AI stack at HyperFrame Research. Inference workloads specifically, Walter added, could benefit as they account for the largest amount of compute consumption in an enterprise: “Inference is a daily challenge. Every time an employee uses an AI assistant or an agent calls a model, it consumes compute.” Primarily because purpose-built accelerators, such as TPUs, can offer different performance and efficiency benefits depending on the model architecture, workload, and software stack, in contrast to GPUs, which are designed to handle a broader range of computing tasks, Walter further added. However, access to the chips alone will not determine the economics of a deployment. “An enterprise may have models, tools, and skills built around a different accelerator. Before switching, it needs to know what runs on TPUs, what needs to change, and how much effort it takes to maintain. A lower infrastructure price is not much of a bargain if the development and operational costs outweigh the savings,” Walter said. Typically, Google’s TPUs are intended for machine learning workloads, including large-model training, post-training, and inference. The chips can also support reasoning-intensive and long-context applications, including those built around AI agents that make multiple model calls to complete a task. Google has spent more than a decade developing its custom AI silicon. Its eighth-generation TPU portfolio includes two architectures: TPU 8t, designed for training, and TPU 8i, designed for inference. TPU 8t, according to the hyperscaler, can deliver nearly three times the compute performance per pod of its predecessor and can reduce frontier-model development cycles from months to weeks. TPU 8i, on the other hand, is designed to handle the iterative workloads generated by multiple specialized AI agents, it added. However, the companies have not disclosed pricing, specific OCI configurations, or the initial capacity that will be available when the accelerators become available. Those details are likely to help enterprises assess whether the additional accelerator option can deliver meaningful cost or performance improvements for their workloads. For now, the only detail available is that Oracle plans to make the chips available in calendar 2027, with capacity ramping up over multiple years. Even a specific launch date has not been shared.