Oracle opens cloud to Google’s AI chips, challenging GPU dominance

Oracle will offer Google’s Tensor Processing Units (TPUs) through its cloud infrastructure platform from 2027, marking the first time the specialised AI accelerators will be available on a hyperscale cloud platform outside Google’s ecosystem.

Announced at Oracle AI World in Las Vegas on October 27, the partnership will allow customers of Oracle Cloud Infrastructure (OCI) to access Google’s custom-designed AI chips alongside other computing architectures, expanding their options for training and deploying artificial intelligence models.

TPUs are processors designed by Google to accelerate machine-learning workloads, including model training and inference — the process of running trained models. They offer an alternative to graphics processing units (GPUs), widely used for AI computing.

“Different AI workloads have different compute needs,” said Mahesh Thiagarajan, president, 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.”

Under the arrangement, Oracle will provide the power, cooling and data-centre networking needed to deploy the accelerators, while Google will supply its TPUs and associated software capabilities.

The companies said the infrastructure would support large-scale AI model training and post-training, as well as production inference for applications involving complex reasoning, long-context processing and agentic AI systems.

“The next generation of AI will require a diverse range of compute,” said Amin Vahdat, senior vice-president and chief technologist, AI & Infrastructure, Google. He added that offering TPUs through OCI would give more organisations access to the technology and greater flexibility in developing and running complex AI applications.

Google’s eighth-generation TPU portfolio comprises TPU 8t, designed for training, and TPU 8i, optimised for inference. According to Google, the TPU 8t delivers nearly three times the compute performance per pod of its predecessor, while the TPU 8i targets the iterative workloads of specialised AI agents.

The service is scheduled to launch in calendar 2027, with deployments expected to scale over several years in response to demand.

The move comes as cloud providers race to expand AI computing capacity. Offering different processor architectures on a single platform could help customers match hardware to workloads and optimise costs. However, the benefits will depend on performance, software compatibility and availability.

The companies did not disclose pricing or specific deployment capacities.