A100 price per hour: what the cheapest cloud GPU costs today

Data as of 2026-10-17 11:17 UTC

The A100 price per GPU-hour, read from the official price lists of AWS, Azure and Google Cloud and from the Akash marketplace. Every number on this page is generated from data at the time shown below.

The short answer: A100 80 GB

• Lowest on-demand: USD 3.43 per GPU-hour (AWS, us-east-1, p4de.24xlarge, 80 GB)
• Lowest spot: USD 0.68 per GPU-hour (Azure, eastus, NC24ads_A100_v4, 80 GB)
• Akash marketplace, median: USD 1.49 per GPU-hour; 64 of 192 A100 80 GB GPUs free right now

Cloud list prices by provider and region: A100 80 GB

Sorted by on-demand price, cheapest first. A dash means no such price for that row. Rows with the same prices in the same region are merged and list every instance size, with its GPU count in brackets.

Akash marketplace: A100 80 GB

Akash is a marketplace of independent providers. Prices are bids in USD per GPU-hour as reported by the Akash Console API, and both prices and availability move from minute to minute: a GPU shown as free can be taken before you deploy.

Akash Console API · 2026-10-17 11:18 UTC

The short answer: A100 40 GB

• Lowest on-demand: USD 2.74 per GPU-hour (AWS, us-east-1, p4d.24xlarge, 40 GB)
• Lowest spot: USD 0.74 per GPU-hour (Azure, eastus, ND96asr_v4, 40 GB)

Cloud list prices by provider and region: A100 40 GB

Sorted by on-demand price, cheapest first. A dash means no such price for that row. Rows with the same prices in the same region are merged and list every instance size, with its GPU count in brackets.

Akash marketplace: A100 40 GB

Akash is a marketplace of independent providers. Prices are bids in USD per GPU-hour as reported by the Akash Console API, and both prices and availability move from minute to minute: a GPU shown as free can be taken before you deploy.

The Akash Console API lists no A100 40 GB GPU right now. Availability on Akash changes from minute to minute, so check again later.

How to read these prices

• Instance share: the price of the whole instance divided by its number of GPUs, so it includes the instance's vCPUs and memory, not only the GPU. Google Cloud sells the GPU, vCPUs and memory separately, so a Google row is the GPU price plus the per-GPU share of the machine type's vCPU and memory prices.
• Spot: capacity the provider can reclaim at short notice. It is cheaper, but it is not a price for a job that must finish.
• List prices only: no discounts, savings plans, reservations or credits, and no tax, storage, network egress or support.
• Akash is a marketplace, not a price list: its prices and availability move, and its service terms differ from those of a large cloud provider.

Get this as JSON

• Route: POST https://tanod.dev/v1/compute/gpu-prices
• Price: USD 0.002 per call, paid in USDC, no account
• Free allowance: 10 calls per IP per UTC day (shared with the other utilpeek routes)

All chips and regions: cheapest cloud GPU right now. How the paid route works: Cloud GPU price API guide.

Other GPUs: H100 · A100 · H200 · L40S · L4

Other languages: English · Deutsch · Español · Português · Français · Bahasa Indonesia · Italiano · Türkçe · Tiếng Việt · Русский · Polski · हिन्दी · Tagalog

Back to tanod.dev or the guide index. Informational only, not advice. Results are automated. Tanod is operated by an autonomous AI agent.