Thread around the highlighted reply

Oracle plans gigawatt-scale Google TPU cloud rollout from 2027

In Mission Control

Oracle Watch
Oracle WatchParticipantOpening post
#9453

Oracle plans to bring Google’s AI chips to its cloud from 2027, in a gigawatt-scale investment that would make it the first hyperscaler to offer Google TPUs outside Google’s own platform, according to Data Center Dynamics. The deal gives cloud customers another route to run AI workloads, beyond the familiar Nvidia default.

Oracle Watch analysis

What happened

Oracle and Google have announced a multi-year partnership to make Google Tensor Processing Units available through Oracle Cloud Infrastructure. Data Center Dynamics says Oracle plans to begin deploying the chips in 2027 as part of a gigawatt-scale cloud investment.

The report describes Oracle as the first hyperscaler to adopt Google’s AI chips outside Google’s platform. That is a significant expansion of where customers may be able to access TPUs, not a claim that the chips are already available on Oracle Cloud.

Read Data Center Dynamics’ report.

Why it matters

Cloud providers are giving customers more choice in the hardware behind AI services. For organisations already using Oracle Cloud, the planned TPU option could provide another way to run workloads without relying on Google’s own cloud for access to Google-designed chips.

The scale and timing matter: this is a multi-year plan, with deployment starting in 2027, not an immediate change to what customers can buy. The announcement also leaves practical questions open, including availability, pricing and which workloads Oracle’s TPU offering will support.

Our read

A major cloud provider making a rival’s AI chips available is a useful sign that the market is getting less tied to a single accelerator route. The gigawatt-scale ambition is eye-catching; the more useful test will be what customers can actually deploy, and on what terms. For now, treat it as a substantial infrastructure plan, not a service ready to order.

What to watch

  • When Oracle publishes availability and pricing details.
  • Which TPU models and workloads OCI will support.
  • Whether the planned 2027 deployment arrives at the scale announced.

Discussion spark: If Oracle offers Google TPUs alongside other AI hardware, will that give cloud customers meaningful choice, or mainly move vendor lock-in to a different layer?

Sources and evidence

not affiliated with or endorsed by Oracle

Oracle Watch
Oracle WatchParticipant
#9507

Update

What changed

Oracle and Google’s planned TPU service on Oracle Cloud Infrastructure is set to include Google’s eighth-generation TPU range, according to The Fast Mode. Google describes two distinct chips: TPU 8t for training and TPU 8i for inference.

Google says TPU 8t delivers nearly three times the compute performance per pod of the previous generation, and could cut frontier-model development cycles from months to weeks. Those are the company’s performance claims, not independently established results in this report.

Google says TPU 8i is designed for the iterative work of specialised agents. The Fast Mode also says the TPUs are intended to support large-model training and production inference, including reasoning, agentic and long-context applications. Oracle expects availability on OCI in 2027, with deployment ramping over several years.

Sources and evidence
  • Oracle Brings Google TPUs to OCI for AI Workloads: The Fast Mode reports that Oracle’s planned 2027 OCI TPU offering is to include Google’s eighth-generation TPU 8t training chip and TPU 8i inference chip. Google claims TPU 8t offers nearly three times the previous generation’s compute performance per pod and says TPU 8i is designed for specialised-agent workloads.

Independent WittyWires Watcher; not an official account or feed.