Google TPUs on Oracle Cloud in 2027: What It Means for AI Costs
Oracle will offer Google TPU 8t and TPU 8i chips on OCI starting in 2027, with no pricing yet. Plan AI costs on hardware you can rent today.
Google's AI chips are leaving Google's cloud. At Oracle AI World on October 27, Oracle and Google announced that Google TPUs will be offered on Oracle Cloud Infrastructure, which Oracle and Google say makes OCI the first hyperscaler outside Google to offer them. The catch is in the date: calendar year 2027, ramping over several years.
What actually happened
Per the joint announcement:
- OCI will offer Google's eighth-generation TPUs: TPU 8t for training and TPU 8i for inference, which Google says is built for many specialized agents working in loops.
- Availability is 2027, with a multi-year ramp.
- Oracle provides the power, cooling and data center networking. Customers will pick TPUs inside OCI alongside Oracle's other AI compute.
- Target workloads include training dense and mixture-of-experts models and production inference for reasoning, agent and long-context apps.
- No pricing and no regions were published. The release talks about "optimized economics" without a number.
Why Google TPUs on OCI matter for your business
You will probably never rent a TPU directly. You will pay for them anyway, inside the per-token price of whatever model API you call. More chip supply across more clouds is good for that price over time. More competition to NVIDIA is good for it too. That is the long view.
The short view is that nothing changes for your 2026 or early 2027 budget. Here is how we would treat it:
Price on today's rates. Build your AI cost model on API prices you can pay now. Treat any 2027 drop as upside, not as the plan.
Keep the model layer swappable. Chip announcements like this move model prices across vendors. If your app can switch providers with a config change, you capture the drop when it lands. If it is hard-wired to one SDK, you wait for your vendor to pass savings on.
Watch agent inference, not training. TPU 8i is pitched at agent workloads: many small calls in loops. That is exactly the pattern that makes agent bills grow. Measure your tokens per completed task now so you know what a cheaper chip is worth to you.
Key takeaways
- Oracle will offer Google TPU 8t and TPU 8i on OCI starting in calendar year 2027
- OCI is the first hyperscaler outside Google to offer TPUs, per both companies
- No pricing or regions were announced
- Budget AI on today's API prices; treat future chip supply as upside
- A swappable model layer lets you capture price drops when they arrive
Know what an agent task costs before the price moves. We build AI systems with per-task cost tracking and a swappable model layer. Run the numbers on our ROI calculator, or talk to us about your AI spend.
Sources: Google Cloud Press Corner: Oracle Brings Google TPUs to Oracle Cloud Infrastructure, Oracle press release.
- #oracle
- #google-tpu
- #oci
- #ai-inference
- #cloud-costs
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Gemini 4 Carbon: cyber defenders only, no public API date
Gemini 4 Carbon tops Google's cyber benchmarks but ships only to vetted defenders, with no public API date. Plan on Argon and shorten your patch cycle.
Read itAnthropic AI exploit chains: Chrome 156 fixes 62 CVEs
Anthropic's AI model built exploit chains for Chrome, Safari, Android, Windows and Linux. Chrome 156 fixes 62 CVEs. Update every browser and device this week.
Read it