Google TPUs Gain Broader Reach With Oracle Deal
Google TPUs will be offered on Oracle Cloud Infrastructure, widening distribution and boosting demand for AI chips and cloud names.
KEY TAKEAWAYS
- Oracle will offer Google TPUs on Oracle Cloud Infrastructure, the first hyperscaler outside Google to do so.
- Availability begins in calendar 2027 and capacity will ramp over multiple years.
- Reported OpenAI plans to buy TPUs remain unconfirmed and contract terms were not disclosed.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
Alphabet Inc.’s (GOOG, GOOGL) Google Cloud and Oracle announced on Oct. 27, 2026, that Oracle Cloud Infrastructure (OCI) will offer Google Tensor Processing Units (TPUs) starting in calendar 2027. This expands market access for the chips as reports emerged that OpenAI plans to purchase TPUs.
Oracle to Offer Google TPUs on OCI
Google Cloud described OCI as the first hyperscale cloud platform outside Google to provide TPUs. Oracle will supply the data-center power, cooling, and networking to host the accelerators, while Google will provide the TPU systems. Oracle said customers can use the chips for large-model training, post-training, and production inference, including mixture-of-experts, reasoning, agentic, and long-context workloads. The companies framed the deal as a commercial infrastructure partnership rather than an acquisition, with capacity set to ramp over multiple years to meet demand.
Sales Momentum and Market Impact
Google began selling TPUs to external customers in the second quarter of 2026 after developing the chips internally since 2013. Reports indicate OpenAI plans to purchase TPUs, marking a possible shift from renting through Google Cloud to direct ownership. Neither company has confirmed a signed contract or disclosed terms, volume, or delivery timing. Other reported TPU renters include Anthropic, Apple, and Citadel Securities, though these names were not confirmed in the official announcements.
TPUs are custom AI accelerators designed specifically for model training and inference. Nvidia’s GPUs remain more broadly programmable. The announcements and market commentary position TPUs as optimized for AI workloads with potential infrastructure-cost advantages. Expanded distribution through cloud partners and reported buyer interest could accelerate TPU adoption and increase competitive pressure on Nvidia’s GPU business as hyperscalers and large-model operators evaluate training and inference options.
