Oracle will bring Google's AI chips to its cloud in 2027: the cloud stops being single-chip
Direct answer
On October 27, 2026, at Oracle AI World in Las Vegas, Oracle and Google announced that Google TPUs will be available through Oracle Cloud Infrastructure (OCI) in calendar year 2027. The companies describe OCI as the first hyperscale cloud platform outside Google to offer TPUs. Deployment will ramp over several years: Google supplies the chips and their software ecosystem; Oracle supplies the power, cooling and networking of its data centers to run them at scale.
Which chips arrive, and what each one is for
- TPU 8t, for training: the eighth generation for training large models. The companies say it delivers nearly 3 times the compute performance per pod over the previous generation. That figure comes from the companies, not from an independent benchmark.
- TPU 8i, for inference: built for reasoning, agents and long-context applications with high concurrency — for running models in production, not just training them.
- What is missing: the announcement gives no month, region or price for availability on OCI. One briefing also noted the chips are listed as coming soon even on Google Cloud's own page, so nothing is available on OCI today.
- The rest of the menu: at the same event, Oracle said Nvidia's Vera Rubin NVL72 systems and Vera CPUs are now generally available on OCI, and that Cerebras will add high-speed inference in 2027. CNBC reported that OpenAI's Jalapeño chip will also join OCI's accelerator lineup.
Oracle also puts ChatGPT inside its applications
- ChatGPT and Codex inside enterprise software: Oracle said it is integrating ChatGPT and Codex into its enterprise applications to help corporate customers query and automate database tasks.
- The strategic turn: for years, each cloud treated its own accelerator as a walled garden to keep developers locked in. Oracle is betting on the opposite: being the place where you rent several kinds of compute and pick the architecture per workload.
- The real competition: the chip is only one part. Power, cooling, networking, software and the ability to keep expensive equipment busy decide how useful that choice becomes.
What it means for your business
- Buy capacity per workload, not per brand: training a large model, running an agent in production and doing real-time inference call for different architectures. The useful question is which one performs best for each task, and at what cost.
- Avoid lock-in to a single provider: if your agent lives on one cloud and one chip, a price or availability change leaves you no room. Design so workloads can move when it makes sense.
- What to watch: the launch month and region on OCI, pricing, and whether your enterprise software provider offers these options inside the applications you already pay for.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- OpenAI reportedly plans to buy Google's AI chips
- OpenAI launched GPT-6.1 Terra at half price
- Uber took ChatGPT Work from 1,000 to 30,000 employees
Frequently asked questions
When will Google TPUs be on Oracle Cloud?
In calendar year 2027, per the October 27, 2026 announcement. Deployment will ramp over several years, and no month, region or price was announced.
What are TPU 8t and TPU 8i?
They are Google's eighth-generation AI chips: the 8t targets training large models, and the 8i targets inference, reasoning and agent workloads in production.
Does this mean Oracle is dropping Nvidia?
No. At the same event Oracle announced general availability of Nvidia's Vera Rubin NVL72 systems on OCI. The bet is adding architectures, not swapping one for another.
Can my company use this today?
Not on OCI: no TPU is available on that cloud today. What is useful today is planning your architecture and the portability terms for your workloads.
