Oracle and Google have announced that Google's Tensor Processing Units will be available through Oracle Cloud Infrastructure, making OCI the first hyperscale cloud outside Google to offer the technology.
The service is planned to launch in 2027, with capacity increasing over several years.
The agreement adds another AI computing option to Oracle's cloud portfolio as providers compete to offer more specialised infrastructure for training and running large AI models. Customers will be able to use the chips for workloads including model training, production inference and agentic AI applications.
Google's TPUs are custom accelerators developed for machine learning tasks across training and inference. Under the arrangement, Oracle will provide the power, cooling and data centre networking needed to deploy the systems within OCI.
The move also reflects growing demand from corporate customers for a broader mix of chips beyond standard graphics processing units, as AI workloads vary in their requirements for cost, speed and energy use. Oracle said customers will be able to choose TPUs within OCI based on workload needs, performance targets and cost considerations.
Mahesh Thiagarajan, President of Oracle Cloud Infrastructure, set out the rationale for adding another chip architecture to Oracle's cloud offering.
"Different AI workloads have different compute needs," said Mahesh Thiagarajan, President of Oracle Cloud Infrastructure. "Bringing Google TPUs to OCI gives customers access to multiple leading AI compute architectures in one cloud. That choice gives them the flexibility to choose the right infrastructure for the performance, throughput and economics of each AI workload."
Chip options
Oracle is positioning the TPU rollout as part of a broader effort to give customers more choice over the infrastructure they use for AI. The addition is aimed at organisations running demanding training jobs, large-scale inference and systems that rely on multiple AI agents.
Google described the chips as part of a long-running programme in custom AI silicon. Its TPU portfolio has been developed over more than a decade and is designed for large machine learning workloads.
In the latest generation, Google has introduced two architectures: TPU 8t for training and TPU 8i for inference. According to the companies, TPU 8t is intended to shorten development cycles for frontier models, while TPU 8i is designed for inference tasks involving many specialised agents.
The announcement gives Oracle a point of differentiation in the AI infrastructure market. While major cloud providers have been expanding support for a range of processors, OCI will be the first hyperscale cloud platform outside Google to offer Google TPUs, according to the companies.
That matters because access to advanced AI chips has become a central competitive issue for cloud providers and their enterprise customers. Companies building and deploying large models are looking for ways to balance training speed, inference costs and power consumption as usage grows.
Google said the TPU architecture is designed to improve the amount of AI processing available from each watt of electricity, while also supporting rapid communication between chips and fast access to model data. Those factors matter in large-scale AI systems, where chip interconnection and data movement can affect both cost and execution time.
The chips on OCI will support dense models and mixture-of-experts models, as well as reasoning, long-context and agentic applications in production, the companies said. These are among the areas where enterprises are increasing spending as they move AI systems from experimentation to day-to-day use.
Amin Vahdat, Senior Vice President and Chief Technologist, AI & Infrastructure at Google, said the partnership is intended to widen access to TPU-based computing.
"The next generation of AI will require a diverse range of compute," said Amin Vahdat, Senior Vice President and Chief Technologist, AI & Infrastructure at Google. "TPUs give customers a powerful, purpose-built option for demanding training and inference workloads. Bringing Google TPUs to OCI will give more organisations access to this technology and greater flexibility in how they build and run their most complex AI applications."