How much does an idle p5en.48xlarge cost on AWS?
A p5en.48xlarge billed on-demand and left running around the clock costs about $46,206/mo at its cheapest listed region (us-east-1). Scheduling it off outside business hours cuts that to roughly $13,752/mo — a $32,454/mo saving.
Cheapest on-demand
$63.2960/hr
us-east-1
Running 24/7
$46,206/mo
730-hour month
Business-hours savings
$32,454/mo
up to 70% off the 24/7 bill (est.)
Price by region
| Region | Hourly (on-demand) | Spot (typical) | ≈ Monthly (24/7) |
|---|---|---|---|
| us-east-1 Cheapest | $63.2960/hr | $20.8877/hr | ≈ $46,206/mo |
| us-east-2 | $63.2960/hr | $15.1910/hr | ≈ $46,206/mo |
| us-west-2 | $63.2960/hr | $18.9888/hr | ≈ $46,206/mo |
| ap-south-1 | $75.9552/hr | $51.6495/hr | ≈ $55,447/mo |
| ap-northeast-1 | $79.1200/hr | $46.6808/hr | ≈ $57,758/mo |
What a schedule saves
| Schedule | Monthly cost | Savings vs 24/7 |
|---|---|---|
| Running 24/7 | $46,206/mo | baseline |
| Weekdays only (off on weekends) | $33,004/mo | -$13,202/mo (29%) |
| Business hours only (10×5) | $13,752/mo | -$32,454/mo (70%) |
Assumes a 730-hour month at on-demand list price. Actual savings depend on when the instance is actually used.
About the p5en family
P5en carries eight NVIDIA H200 GPUs — 141 GB of memory per GPU — on a newer host platform built for multi-node jobs where the interconnect sets the pace as much as GPU memory does. It appears in accounts as short-lived training and evaluation capacity, often available in only a subset of regions. Hardware this scarce tends to be held "just in case" between runs, which is the costliest way to hold anything.
Regional pricing varies: p5en.48xlarge runs $63.2960/hr in us-east-1, the cheapest region priced here, versus $79.1200/hr in ap-northeast-1, the priciest — a 20% spread across regions for the identical instance type.
Common questions
How much does an idle p5en.48xlarge cost per month?
At the cheapest listed region (us-east-1), a p5en.48xlarge costs $63.2960/hr, or about $46,206/mo if it runs 24/7. That's the on-demand list price — an idle instance costs exactly the same as a busy one, since cloud billing doesn't know the difference.
How much can I save by scheduling it off outside work hours?
Restricting it to business hours only (10×5) saves about $32,454/mo (70%) versus running 24/7. A lighter schedule — on around the clock on weekdays, off on weekends — still saves about $13,202/mo (29%).
When does H200 memory matter versus an H100?
When the model, its optimizer state, or the serving context no longer fits in 80 GB per GPU — long-context work and large fine-tunes are the usual triggers. A job that fits comfortably on H100s pays the H200 premium for headroom it never uses; one that only barely fits with sharding gymnastics often gets the premium back in simpler, faster runs.
Stop paying for idle AWS servers
Idlefy runs a free audit of your fleet, then flips dev servers to stopped-by-default: lease one when you need it, and
when the lease ends it shuts itself off — stop, not terminate, persistent disks intact. Timers are set by your team, never
guessed, and Idlefy only manages instances tagged idlefy=enabled — on AWS that boundary is enforced by the IAM policy itself.
Prices are on-demand list price for Linux instances and exclude storage, network, data transfer, and any negotiated or volume discounts. Actual bills vary by usage and account-level pricing. Prices as of 2026-08-31. Spot figures are indicative — AWS values are derived from AWS's published typical savings vs on-demand for the region; GCP values are the listed Spot price. Spot capacity can be reclaimed by the provider. Stopping preserves EBS and persistent-disk volumes; data on local instance-store or local-SSD drives does not survive a stop.