← All hardware
Google's widely available 6th-gen TPU for efficient training and inference.
Pros
- Generally available and widely proven
- Cost-efficient training/inference via cloud
- Tight Google Cloud and JAX integration
Cons
- Lower per-chip memory than v7 Ironwood
- Cloud-only, Google-ecosystem-locked
- Superseded at the high end by Ironwood
✓ Where it shines / best for
- Training and serving large language and generative models on Google Cloud
- Teams already on JAX or PyTorch/XLA wanting cost-efficient throughput
- High-volume batch inference at scale
✕ Not the best fit for
- On-premises deployments (cloud-rental only, not sold as hardware)
- Small experiments where a single GPU is cheaper
- CUDA-locked workloads that can't port to XLA
Features
- ✓ LLM
- ✓ API access
- ✓ Inference
- ✓ Training
- ✓ High Throughput
- ✓ Hbm
- ✓ Jax
- ✓ Cloud TPU
- ✓ Free trial
- ✓ Pytorch Xla
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| On-demand (us-central) | ~$2.70 | per chip-hour | Cloud TPU v6e (Trillium) on-demand list price; varies by region. No purchase of hardware — rented via Google Cloud. |
| 1-year commitment | ~$1.89 | per chip-hour | Approx. 30% discount vs on-demand with committed-use. |
| 3-year commitment | ~$1.22 | per chip-hour | Approx. 55% discount vs on-demand with committed-use. |
| Spot/Preemptible | varies | per chip-hour | Discounted preemptible pricing available; subject to reclamation. |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| memory | 32 GB HBM per chip, ~1,600 GB/s |
| cooling | cloud-only |
| pod_scale | 256 chips per pod |
| performance | 4x training / up to 3x inference vs prior gen |
| architecture | Trillium (TPU v6e), Google custom ASIC |
Sponsored
A full review is being generated for this product and will appear here shortly.