Reserved vs On-Demand Calculator
Find the break-even between paying on demand and committing to reserved capacity.
Provider profile
AWS: AI GPU cloud profile
AWS is the largest general-purpose hyperscale cloud; it rents NVIDIA GPU instances for AI training and inference alongside its full platform of compute, storage, and networking. ComputeTape tracks its public on-demand H100-class rate.
AWS is one of the hyperscalers ComputeTape links to a public price page. This profile summarizes what is verified, the on-demand list rate we track, and where it fits — each figure tied to the public source.
H100 On DemandTracked offering
The configuration ComputeTape observes on the public price page.
What this is
What AWS is
AWS is the largest general-purpose hyperscale cloud; it rents NVIDIA GPU instances for AI training and inference alongside its full platform of compute, storage, and networking. ComputeTape tracks its public on-demand H100-class rate.
- A full general-purpose hyperscale cloud that also rents NVIDIA GPU instances for AI.
- GPU capacity is one service inside a broad platform, not a standalone GPU product.
Tracked offering
The offering ComputeTape tracks
ComputeTape tracks the public on-demand configurations published on the provider price page — one row per GPU. The list rate is the published on-demand price observed there. It is a starting point, not a quote; negotiated, committed, and reserved terms differ.
| Service | GPU | Mode | Region | On-demand list rate | Availability | No change in | Last checked | Confidence | Source |
|---|---|---|---|---|---|---|---|---|---|
| EC2 | H100 | On Demand | us-east-1 | $6.88 / GPU-hr | Quota Gated | 23 days | Jul 14, 2026 | High | AWS pricing page |
List rate is the public on-demand price, not a quote. Confirm the live figure on the source page before budgeting.
Where it fits
How ComputeTape reads this provider against the rest of the directory.
- Teams already on AWS that want GPU capacity next to their existing data and services.
- Buyers who value region breadth, enterprise support, and integration over the lowest hourly rate.
Best-fit notes are ComputeTape interpretation, not provider guarantees.
Watch-outs
Things to confirm for your own workload before you commit.
- On-demand list rates are usually higher than neoclouds; large savings come through committed, reserved, or capacity-block plans.
- GPU instance capacity is quota-gated by region and account; availability can require a quota-increase request.
Committed capacity
On-demand vs committed capacity
ComputeTape tracks the public on-demand H100-class rate. Savings Plans, Reserved Instances, and Capacity Blocks are negotiated separately and are not reflected in the list price above.
On-demand vs reserved vs spot GPU pricing
Compare access terms, savings, and interruption risk.
Sources and checks
This profile renders only while the underlying row stays reviewed for display and within the 30-day recheck window. If a recheck lapses, the row and this profile drop from the directory until it is reviewed again.
AWS public price page
The provider-published page the tracked rate is read from.
Evidence labels reference
The rights, freshness, and confidence labels applied to every row, defined and explained.
- Last checked: Jul 14, 2026.
- No change in: 23 days (how long the current H100 list price has held in our records).
- Source confidence: High.
- Rights: links a public price page on the provider site.