Oracle to Deploy Google TPU AI Chips on OCI in Multi-Gigawatt Deal

Oracle will deploy Google's Tensor Processing Units (TPUs) inside Oracle Cloud Infrastructure (OCI), becoming the first hyperscaler to offer the chips since Google opened sales to a "select group of customers" in April 2026. TPUs on OCI go live in calendar year 2027, ramping over a multi-year period.

The deal is big. Karan Batta, SVP of Oracle Cloud Infrastructure, told DCD at the Oracle AI World event in Las Vegas that it would be a "multi-year, multi-gigawatt partnership with Google and it will look and feel the same as it does from a GCP standpoint." Oracle declined to specify the initial deployment's scale.

"What I can tell you is that this is not an experiment," Batta added. "This is going to be a cloud gigawatt-scale investment from Oracle over time, and the best way to think of it is that when Google launches a new TPU, we will have it on day one."

What does this change for OCI customers?

OCI tenants gain a second major AI compute architecture alongside existing options, without leaving Oracle's cloud. Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, framed it as workload economics.

"Different AI workloads have different compute needs," Thiagarajan said. "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."

Which chips are involved?

Google is on its eighth TPU generation after more than a decade of development. The lineup splits into two parts:

• TPU 8t — the training chip, delivering nearly 3x the compute performance per pod over the previous generation. It carries 19.2Tbps of bidirectional scale-up bandwidth (chip-to-chip links inside a pod) and 400Gbps of scale-out networking bandwidth.
• TPU 8i — the inference chip, built for the iterative work of many specialized AI agents. It scales to 1,152 chips in a single pod, delivers 11.6 exaflops of FP8 compute, and offers 331.8TB of total HBM capacity per pod plus 19.2Tbps of bidirectional scale-up bandwidth per chip.

The 8i's extra memory bandwidth targets latency-sensitive inference workloads.

Google also built a new networking architecture for the hardware. Called Virgo Network, it supports a 4x increase in data center bandwidth and uses high-radix switches (switches with many ports) to cut the number of network layers.

Who gets TPUs first?

Google is also working with investment firm Blackstone on a US-based cloud platform offering TPUs. Depending on which project launches first, Oracle could be the first operator to bring Google's chips online outside Google's own platform.

For now, the 2027 availability date and Batta's "day one" commitment define the roadmap: OCI becomes a standing distribution channel for every future TPU generation, backed by what Oracle calls a gigawatt-scale investment over time.