GPU MODELS / A100

A100 cloud GPU pricing

13 listings across 9 providers. Cheapest published rate: $0.890/GPU/hr at Deep Infra. Sorted by per-GPU hourly price. Every row links to the provider's own published price.

WHAT IS THE A100

The A100 is the most influential datacenter GPU of the decade: the chip GPT-class models were scaled on, and still the market's price-performance reference. Panel rows under this name vary in memory (40 vs 80 GB) and form factor - check each row's config for the exact variant.

MAKER
NVIDIA
ARCHITECTURE
Ampere
LAUNCHED
2020
MEMORY
40 or 80 GB HBM2/e
MEMORY BANDWIDTH
1.6-2.0 TB/s
TENSOR COMPUTE
312 TFLOPS FP16/BF16 tensor dense (624 sparse)
POWER DRAW
250-400 W
INTERCONNECT
NVLink 3.0 (SXM) or PCIe 4.0

WHAT IT IS GOOD AT

Fine-tuning
The default for 7B-70B models.
Multi-GPU training
Real NVLink in SXM nodes.
Inference
80 GB parts serve 70B-class models per card.

WHO SHOULD NOT RENT THE A100

Workloads that fit a 24-48 GB card - consumer and L-class parts serve them for less.
FP8-native pipelines - Hopper and later only.
ProviderConfigVRAMLocation$/hr$/GPU/hrSource
Deep InfraDeep InfraA100 × 180 GBNot published$0.890$0.890receipt
Thunder ComputeThunder ComputeA100 × 180 GBNot published$1.090$1.090receipt
SesterceSesterceA100 × 180 GBNot published$1.240$1.240receipt
HyperstackHyperstackA100 × 180 GBUS / EU / CA$1.350$1.350receipt
Massed ComputeMassed ComputeA100 × 180 GBNot published$1.350$1.350receipt
Massed ComputeMassed ComputeA100 × 880 GBNot published$10.800$1.350receipt
GcoreGcoreA100 (from) × 180 GBNot published$1.436$1.436receipt
HexgridHexgridA100 (from) × 180 GBNot published$1.490$1.490receipt
LyceumLyceumA100 × 180 GBEurope$1.590$1.590receipt
Hugging FaceHugging FaceA100 × 180 GBUS$2.500$2.500receipt
Hugging FaceHugging FaceA100 × 280 GBUS$5.000$2.500receipt
Hugging FaceHugging FaceA100 × 480 GBUS$10.000$2.500receipt
Hugging FaceHugging FaceA100 × 880 GBUS$20.000$2.500receipt

THE A100 MARKET, BY THE NUMBERS

PROVIDERS
9
CONFIGURATIONS
13
MEDIAN $/GPU/HR
$1.44
SPREAD
2.8x

9 providers publish 13 A100 configurations on the panel today. Published per-GPU hourly rates run from $0.890 (Deep Infra) to $2.50 (Hugging Face), a 3x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $1.44/GPU/hr. 9 rows were reported in stock by the provider at last observation; the rest are published catalog rates. All rates are on-demand - committed-use and reserved pricing is excluded.

CHEAPEST A100 BY GEOGRAPHY
united statesHyperstack$1.350/hrcanadaHyperstack$1.350/hr
A100 PRICE TREND, DAILY PANEL SNAPSHOTS

Across every A100 row on the panel, the daily floor moved from $1.350 to $0.890/GPU/hr over 11 days of tracking (-34.1%). The median continuously-tracked listing held flat at $1.35. Snapshots run daily; the full history is public in the repo.

A100 PROVIDER NOTES
Deep Infra · 1 config · 1 in stock
Sets the panel floor for this model.
from $0.890/hr
Thunder Compute · 1 config · 1 in stock
1 of 1 configs reported in stock at last check.
from $1.090/hr
Sesterce · 1 config · 1 in stock
1 of 1 configs reported in stock at last check.
from $1.240/hr
Hyperstack · 1 config · 1 region
Published catalog rates.
from $1.350/hr
Massed Compute · 2 configs
Published catalog rates.
from $1.350/hr
Gcore · 1 config
Published catalog rates.
from $1.436/hr
Hexgrid · 1 config · 1 in stock
1 of 1 configs reported in stock at last check.
from $1.490/hr
Lyceum · 1 config · 1 region · 1 in stock
1 of 1 configs reported in stock at last check.
from $1.590/hr
BUYING A100: WHAT THE PANEL SAYS

At the current floor, one A100 running around the clock costs about $650 a month (730 hours of on-demand arithmetic). Node shape matters: the single-GPU floor is $0.890/GPU/hr against $1.350/GPU/hr on an 8-GPU node (+51.7%). Per-GPU floors by config size: 1x $0.89 · 2x $2.50 · 4x $2.50 · 8x $1.35. Compare the alternatives below before committing - the same budget often buys more than one chip class.

A100 VS THE ALTERNATIVES

A100 vs H100 SXM
Hopper is roughly 3x faster dense with FP8 support; A100 is roughly half the price. Both statements are on the panel - compare directly.

Compare live rates: A100 vs H100 SXM · A100 vs Ascend 910B

A100 PRICING FAQ

What is the cheapest A100 cloud GPU right now?
The cheapest published A100 rate on the panel is $0.890 per GPU-hour at Deep Infra (A100 x 1, Not published). Every price links to the provider's own published rate.
How much does a A100 cost per hour?
Across 9 providers tracked today, published per-GPU hourly rates for A100 run from $0.89 to $2.50, with a median of $1.44. Committed-use and reserved rates are excluded; everything shown is on-demand.
How many providers rent A100 GPUs?
9 providers publish 13 distinct A100 configurations on Compute Cafe. 9 of those rows were reported in stock by the provider's live endpoint at last observation.
Where is A100 cheapest?
By published per-GPU hourly rate, the cheapest geography for A100 today is united states at $1.35/hr (Hyperstack). The spread across 2 tracked geographies runs to $1.35/hr in canada.
Are A100 prices going up or down?
Over the 11 days Compute Cafe has tracked the panel, the A100 floor is falling: $1.350 to $0.890/GPU/hr (-34.1%). The median continuously-tracked listing moved from $1.35 to $1.35 (+0.0%). Daily snapshots are public in the repo, so the trend is auditable.
What does a A100 cost per month?
At the current floor of $0.890/GPU/hr, a A100 running 24/7 costs about $650 per month (730 hours of on-demand arithmetic on the cheapest published rate). Committed-use contracts can price lower; the panel tracks on-demand rates only.
Is a full 8-GPU A100 node cheaper than a single card?
On today's panel: the single-GPU floor is $0.890/GPU/hr and the 8-GPU node floor is $1.350/GPU/hr - more expensive per GPU on a full node (+51.7%). Check the table for the exact configs behind each floor.
Why do A100 prices vary so much?
Memory size (40 vs 80 GB), form factor (PCIe vs SXM/NVLink), and node shape (single cards vs 8-GPU nodes) all move the rate. The panel breaks each out per row.

See also alternatives to the A100 · cheapest GPU-hour per model · inference chips · training chips · by location