a3-ultragpu-8g price per hour

Snapshot 2026-10-23 05:37:50 UTC · Google Cloud · H200

a3-ultragpu-8g in us-central1 costs USD 83.4920 per machine-hour or USD 10.4365 per GPU-hour on-demand. It has 8 H200 GPUs (141 GB per GPU), 224 vCPUs and 2952 GiB of system memory. The GPU count and GPU memory are from the price snapshot. The snapshot’s vCPU and RAM shares supply the vCPU and system-memory figures.

On-demand and spot by region

All regions for this exact Google Cloud SKU in the snapshot are below. A machine-hour is the snapshot's per-GPU list price multiplied by this instance's GPU count; the listed per-GPU value is already a rounded share of a machine-hour. “Not listed” is not a zero price. AWS spot is not covered by this snapshot.

RegionOn-demand / machine-hour On-demand / GPU-hourSpot / machine-hour Spot / GPU-hourPrice effective
asia-south1USD 100.1960USD 12.5245USD 56.9576USD 7.11972026-10-22
europe-west1USD 91.9472USD 11.4934USD 55.1552USD 6.89442026-10-22
europe-west4not listednot listedUSD 55.0920USD 6.88652026-10-22
us-central1USD 83.4920USD 10.4365USD 50.0856USD 6.26072026-10-22
us-east4USD 83.4920USD 10.4365USD 47.3392USD 5.91742026-10-22
us-south1not listednot listedUSD 58.9648USD 7.37062026-10-22
us-west1USD 83.4920USD 10.4365USD 50.0856USD 6.26072026-10-22

Cheapest same-chip alternative

No other on-demand instance type with this exact GPU chip and memory variant appears in this snapshot.

What the price covers

The source is Google Cloud Billing Catalog API, recorded at 2026-10-23 05:37:50 UTC in the GPU snapshot also used by the live GPU index. The list-price basis is instance_share: AWS/Azure divide the whole instance price by its GPUs; Google Cloud adds the GPU SKU and the machine's CPU and memory SKU shares. All include vCPU and system RAM; local SSD, boot disks, licences, network, tax and discounts are not included. Spot is interruptible and not a guaranteed completion price. Price presence does not guarantee available capacity.

Compare this chip in the H200 price-per-hour guide. For current filters, call the GPU prices API: POST https://tanod.dev/v1/compute/gpu-prices with {"chip": "H200", "provider": "Google Cloud", "regions": ["us-central1"], "pricing": "on_demand"}. Filter the returned rows by sku for this particular instance; the route ranks all matching sizes. Compare renting with buying a GPU.

Automated list prices, not a quote. All guides.