kitio.aiAdded Nov 5, 2026
Cloud TPU
PaidGoogle's custom AI accelerators for training and serving models
What is Cloud TPU?
Cloud TPU rents Google's Tensor Processing Units, the chips behind Gemini, for training, fine-tuning and serving large models. It is aimed at ML teams that need large-scale compute for frontier training, inference or agent workloads on Google Cloud.
TPUs support PyTorch and JAX natively and the vLLM engine for fast inference, and deployments can be managed and scaled with Google Kubernetes Engine. Generations on offer include Ironwood, Trillium, v5p and v5e, plus older v4, v3 and v2 pods and devices.
Billing is per chip-hour and varies by region. On-demand rates sit alongside cheaper Dynamic Workload Scheduler options and one- or three-year commitments. New Google Cloud customers get free credits, and larger buyers can request quota and a custom quote from sales.
Features
- Ironwood and Trillium TPUs
- PyTorch and JAX support
- vLLM inference
- GKE integration
- Committed-use discounts
Tags
Releases
+ Submit new release- Listed
Cloud TPU was added to the directory.
Organization
🇺🇸 United States
Listed by kitio.ai · unverified owner
Pricing
- Pricing model
- Paid
- Paid options from
- Not listed
Cloud TPU plans
TPU v5e
$1.20 / chip-hour
- On demand, us-central1
- Cost-efficient training and inference
Trillium
$2.70 / chip-hour
- On demand, us-east1
- Sixth-generation TPU
TPU v5p
$4.20 / chip-hour
- On demand
- Large-scale training
Ironwood
$12.00 / chip-hour
- On demand, us-central1
- Newest TPU generation
Prices in USD, checked Nov 5, 2026. Plans change often, so confirm on the Cloud TPU site.Official pricing
Sponsored
Your AI tool here
A Featured launch puts your tool in this spot and on the home page for a week.
Top Cloud TPU alternatives
View allModels & APIs
Specialized AI cloud for large-scale GPU training and inference
Models & APIs
Google Cloud API that labels objects, scenes and text in video
Models & APIs
GPU cloud for AI training and inference with managed MLOps
Models & APIs
Cloud workspaces with GPUs for training and serving models
Compare Cloud TPU
If you liked Cloud TPU
GPU cloud for training and inference on NVIDIA H100, H200 and B200
PaidModels & APIsMade in United StatesManaged Ray platform for scaling AI training, data and serving
PaidModels & APIsMade in United StatesOn-demand NVIDIA GPU cloud for AI training and inference
PaidModels & APIsMade in United StatesGoogle Cloud's semantic-layer BI with Gemini conversational analytics
PaidData & analyticsMade in United States
Pros and cons
Pros
- Same chips Google uses for Gemini
- Several generations to match budget
- Large discounts for commitments
Cons
- Pricing varies by region and is hard to estimate
- Tied to Google Cloud
- Quota for big clusters goes through sales
Prompts & results
No prompts yet. Add your own prompt and output to help others use this tool.
Share a prompt
Reviews
Q&A
What is Cloud TPU?
Cloud TPU rents Google's Tensor Processing Units, the chips behind Gemini, for training, fine-tuning and serving large models. It is aimed at ML teams that need large-scale compute for frontier training, inference or agent workloads on Google Cloud.
Which frameworks run on Cloud TPU?
TPUs have native support for PyTorch and JAX, and they support vLLM for inference.
Can I get Cloud TPU cheaper than on-demand?
Yes. Dynamic Workload Scheduler modes and one- or three-year commitments cost less per hour than on-demand.
How much does Cloud TPU cost?
Cloud TPU is paid.
Is there a free version of Cloud TPU?
No. Cloud TPU is a paid tool.
What are the main features of Cloud TPU?
Ironwood and Trillium TPUs, PyTorch and JAX support, vLLM inference, GKE integration, Committed-use discounts.
Who makes Cloud TPU?
Cloud TPU is made by Google.
What are the best alternatives to Cloud TPU?
Popular alternatives include CoreWeave, Google Cloud Video AI, Nebius AI Cloud.