Google Cloud GPU pricing 2026
GCP exposes Nvidia GPUs through A3 (H100 / H200), A2 (A100), and G2 (L4) machine series, plus Google TPUs as a parallel accelerator family.
Published rate card
Per-GPU per-hour rates pulled from the vendor pricing page linked above, Last verified June 2026. Rates exclude storage, egress, and any managed-service uplift. Reserved-capacity contracts typically improve on these rates.
| GPU | Configuration | Per GPU per hour |
|---|---|---|
| H100 80GB (a3-highgpu-8g) | On-demand, per-GPU equiv | $11.060 |
| H100 80GB (a3-highgpu-8g) | Spot, per-GPU equiv | $3.690 |
| A100 80GB (a2-ultragpu-8g) | On-demand, per-GPU equiv | $5.070 |
| A100 40GB (a2-highgpu-8g) | On-demand | $3.670 |
| L4 (g2-standard-4) | On-demand | $0.810 |
| T4 | On-demand | $0.350 |
Hidden costs to watch
- VM machine type (vCPU and memory) billed in addition to per-GPU compute.
- Persistent SSD, Filestore, and Cloud Storage egress sit on top.
- Spot VMs are interruption-prone; multi-day training requires preemption-tolerant orchestration.
- CUDs (Committed Use Discounts) need 1 or 3 year terms for the published savings tier.
What Google Cloud is best for
Teams already on GKE, Vertex AI, or BigQuery that benefit from native Google-specific accelerators (TPU v5p / v5e) alongside Nvidia H100 / H200.
See the GPU cloud buying guideWorked example
Acme Vision Co. (illustrative example, not a real company) needs to train a 10 billion-parameter vision-language model on a fixed 8-GPU cluster for 30 days at 18 hours per day duty cycle.
At Google Cloud's cheapest published rate of $3.69 per GPU-hour, the run costs $15,941 for raw GPU compute, before storage, egress, and MLOps overhead. Add the typical 25 percent year-one uplift and the modelled spend is $19,926. Use the calculator on the homepage to model your own GPU class, cluster size, and duty cycle.
Last verified June 2026. Google Cloud rates change frequently. Always obtain a vendor quote before purchase.
Visit Google Cloud pricing page