A100 80GB Rental Price: Cheapest Cloud per Hour (2026)

Choose an A100 80GB rental for your workload using provider-reported hourly prices and availability compiled from 7 sources and updated 2026-10-07.

Updated 2026-10-07 · refreshed every 12 hours

If your A100 80GB workload cannot restart after interruption, choose Vast.ai at $0.38 per GPU-hour; availability is In stock. If the same workload can restart, consider the spot alternative from DataCrunch at $0.93 per GPU-hour; availability is In stock.

Recommendation source: This on-demand recommendation is a marketplace quote, not a provider-published list rate. Recheck it before booking.

Spot recommendation source: This spot recommendation uses a provider-published list rate; availability can still change.

Spot evidence: 1 observed spot configuration from 1 provider.

A100 80GB is a data-center GPU priced here as a 1x A100 80GB rental unit. Its common hardware means workload fit depends first on whether the model, runtime state, and precision fit on that unit; the on-demand and spot labels change rental terms, not the GPU identity.

Run a bounded representative test using the supplied on-demand rate of $0.38 per GPU-hour as a cost reference, confirm the actual checkout rate, and record peak memory, completed work, and billed runtime.

Does your model and serving workload fit the A100 80GB capacity and GPU count shown in the adjacent workload table before you compare providers? If your workload falls under BF16 workloads within 80GB VRAM, reproduce the table’s formula with your actual precision, batch size, sequence length, KV cache, and activations, then validate the displayed GPU count with a representative run. If the workload fits, require an observed offer for that exact configuration before renting. If it does not fit, increase the configuration and repeat the fit test.

Can your run checkpoint frequently enough to restart safely after an interruption? If your run can checkpoint and restart, consider the spot option from DataCrunch at $0.93 per GPU-hour; its recorded availability is In stock. Before renting, require an observed offer for the exact configuration and verify the checkout rate.

At checkout, do the actual availability and on-demand rates change your choice between Vast.ai at $0.38 per GPU-hour (recorded availability: In stock) and Hyperstack at $1.35 per GPU-hour (recorded availability: Not reported)? If an observed offer for the exact configuration remains available at checkout, follow the supplied on-demand recommendation: Vast.ai at $0.38 per GPU-hour, with recorded availability of In stock. When both configurations fit, compare the current on-demand rates and measured job costs: Vast.ai at $0.38 per GPU-hour with In stock, versus Hyperstack at $1.35 per GPU-hour with Not reported. The mapped on-demand rate differences are $0.97 per GPU-hour and $8497.20 per GPU-year.

Current recommendation

For runs that cannot be interrupted, use Vast.ai at $0.38/hr (In stock).

The one observed spot offer is DataCrunch at $0.93/hr (In stock); use it only for checkpointable work.

The closest on-demand comparison is Vast.ai at $0.38/hr (In stock) versus Hyperstack at $1.35/hr (Not reported): $0.97/hr, or $8497.20/year.

Today's prices

USD/hr · one observed offer per provider and pricing model

1 GPU

Provider	Configuration $/hr	Pricing	Price basis	Availability
Vast.ai	$0.38	on-demand	Marketplace quote	In stock
DataCrunch	$0.93	spot	Published list rate	In stock
Hyperstack	$1.35	on-demand	Published list rate	Not reported
Jarvislabs	$1.49	on-demand	Published list rate	Not reported
RunPod	$1.59	on-demand	Published list rate	In stock
DataCrunch	$1.85	on-demand	Published list rate	In stock
Crusoe	$2.00	on-demand	Published list rate	Not reported

8 GPUs

Provider	Configuration $/hr	Per GPU $/hr	Pricing	Price basis	Availability
Lambda	$22.32	$2.79	on-demand	Published list rate	Unavailable as of 2026-10-07

5.26x

Use the provider spread to shortlist current on-demand offers, then compare the modeled gap between on-demand and spot options with measured restart and checkpoint costs before choosing a purchase model.

-143.6%

Treat the figure as a modeled rate gap between spot and on-demand options, then compare it with measured restart and checkpoint costs before choosing a purchase model.

$277/mo

Use the monthly baseline as a planning estimate, then replace it with measured billed runtime and completed work; compare the modeled gap between on-demand and spot options with measured restart and checkpoint costs before choosing a purchase model.

What an hour buys

Computed from the recommended on-demand offer: Vast.ai at $0.38/hr for 1 GPU.

$0.0047

Per GB of VRAM, hourly$0.38/GPU-hour divided by 80 GB of VRAM.

$0.0012

Per dense BF16 TFLOP, hourly$0.38/GPU-hour divided by 312 dense BF16 TFLOPS.

5366

GB/s of memory bandwidth per $/hour2039 GB/s divided by $0.38/GPU-hour.

Hardware specifications

Specification	A100 80GB
Memory	80 GB HBM2e
Memory bandwidth	2039 GB/s
Dense BF16	312 TFLOPS
Dense FP8	FP8 unsupported
NVLink	600 GB/s
TDP	400 W

What your workload needs

VRAM needed is calculated from the formula shown in each row. A listed hourly cost appears only when a currently eligible on-demand offer exists for that exact GPU count. 1 listed workload is omitted because no currently eligible on-demand configuration exists at the required GPU count.

Workload	VRAM needed	A100 80GB	Listed hourly cost
7B Q4 inference	4 GB 7B × 0.5 B (4-bit) × 1.2 (KV+activations)	1 GPU	$0.38/hr (In stock; observed 2026-10-07; price basis: Marketplace quote)
13B Q4 inference	8 GB 13B × 0.5 B (4-bit) × 1.2 (KV+activations)	1 GPU	$0.38/hr (In stock; observed 2026-10-07; price basis: Marketplace quote)
70B QLoRA	46 GB 70B × 0.5 B (4-bit base) × 1.3 (adapters+optimizer)	1 GPU	$0.38/hr (In stock; observed 2026-10-07; price basis: Marketplace quote)
70B FP8 training	154 GB 70B × 2 B (FP8 mix + master/optimizer) × 1.1 (requires FP8 hardware)	unsupported	n/a

What the numbers say

A100 80GB fit begins with matching your workload to the table’s displayed formula and GPU count because each row uses workload-specific assumptions for precision, KV cache, activations, adapters, or optimizer state. Reproduce the relevant formula with your batch size and sequence length, then run a representative test and record peak memory before accepting the displayed configuration.

A100 80GB memory-bandwidth and BF16 peaks are datasheet specifications, not measured throughput: bandwidth-bound decode and compute-bound training or prefill can respond differently, while interconnect matters separately in multi-GPU runs. Verify software precision support in the chosen framework. Benchmark the same model, precision, batch size, sequence length, and output-quality target, then record completed work and billed runtime.

A100 80GB is unsuitable when the required precision lacks hardware and software support, a representative run exceeds the exact configuration’s memory, or the observed rental cannot supply the required multi-GPU topology.

Turn the workload estimate into a rental decision

Measure the memory peak
Use the intended precision, batch size and sequence length. KV cache stores attention state during serving; activations and optimizer state depend on the training setup. A weight-only estimate leaves these out.
Match the sold configuration
Read the GPU count next to the hourly cost. If no eligible offer exists for that count, the memory estimate is not a launchable rental. Check topology before splitting a job across GPUs.
Bound the experiment
Set a test budget and stop condition. Record billed runtime and completed work at the confirmed checkout rate, including recovery time for a spot test.

How Marlin helps

Marlin matches your workload requirements to the lowest-priced suitable GPU option across supported CSP and GPU-cloud providers.

Provider reach: Marlin matches across every supported provider, not only those with listable offers on this page.
Single comparison: Marlin compares supported providers in one matching process, reducing the need to check prices provider by provider.
Fit-qualified price: Marlin returns the cheapest GPU option that satisfies your workload requirements, not merely the cheapest listed row.

Using this page manually means rechecking provider terms and checkout rates against the supplied on-demand reference of $0.38 per GPU-hour; Marlin handles the supported-provider comparison when matching your workload to the lowest-priced suitable GPU option.

Before you rent

Data transfer: Outbound traffic may be billed separately from GPU runtime; verify the charge in the linked billing terms or checkout quote.

Checkpoint recovery: With spot capacity, infrequent checkpoints increase repeated work and billed runtime after an interruption; measure restart time and lost work in a representative run.

Capacity: Availability recorded for 1x A100 80GB does not establish capacity for a larger multi-GPU topology; verify the exact GPU count and interconnect before migrating the workload.

Billing granularity: Per-second, per-minute, or per-hour rounding can make short runs cost more than their exact runtime implies; confirm the billing increment on the provider’s own billing page before committing.

Workload memory requirements, GPU counts, rate ratios, and projected monthly or annual costs are modeled rather than measured job results. Rental prices were collected from provider listings, while hardware figures come from manufacturer specifications. Re-check current rates and availability on the linked provider pages and hardware figures in the manufacturer sources, then confirm the checkout terms. https://vast.ai/pricing · https://www.hyperstack.cloud/gpu-pricing · https://jarvislabs.ai/pricing · https://www.runpod.io/pricing · https://datacrunch.io/pricing · https://crusoe.ai/cloud/pricing · https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf · https://lambda.ai/service/gpu-cloud

Marlin beta

Stop comparing. Start running.

Stop tracking provider prices manually and use Marlin to match your requirements to the lowest-priced suitable GPU option across supported providers.

Get started with Marlin