NVIDIA A100 80GB rental price per hour
Series 1 · GPU prices · NVIDIA · snapshot 16 Oct 2026
The A100 is the legacy workhorse of the index: $1.75 per GPU-hour at the median, down from $3.20 in Jan 2024, with marketplace hosts offering it for about $0.53.
Figure 1A100: median rental price and published range, USD per GPU-hour, 2024-01-01 to 2026-10-16
§1Reading the A100 market
Vast.ai's marketplace sets the floor at ~$0.53, but the tracker records a $0.39–0.67 band depending on each host's reliability rating. Thunder Compute's own $1.09 is the lowest managed-cloud anchor, and CoreWeave's $2.21 for A100 PCIe is a newer data point in the middle.
Civo publishes the clearest term ladder for any model in the record: $1.79 on demand, $1.59 for 12 months, $1.39 for 36. Google Cloud's a2-ultragpu, at ~$5.07, has been the ceiling throughout the series, so the 9.6× spread is mostly one hyperscaler against the open market.
GetDeploying's tracked median moved from $1.69 to ~$1.75 in the last cycle, which the tracker treats as noise. The longer arc is unambiguous: −45% since Jan 2024, with most contract demand having moved to H100 and H200.
Table 1A100 prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Vast.ai | Marketplace | ~$0.53 | range $0.39-0.67 depending on host reliability | 2026-09-25 |
| Thunder Compute | On-demand | $1.09 | 2026-09-25 | |
| Civo | Reserved, 36 mo | $1.39 | 2026-09-25 | |
| Civo | Reserved, 12 mo | $1.59 | 2026-09-25 | |
| GetDeploying median | Third-party index | ~$1.75 | 2026-09-25 | |
| Civo | On-demand | $1.79 | 2026-09-25 | |
| CoreWeave | On-demand | $2.21 | A100 PCIe | 2026-09-25 |
| Google Cloud | On-demand | ~$5.07 | a2-ultragpu | 2026-09-25 |
Rows are the prices recorded in the tracker note for this model; 6 of the 8 come from the provider's own price page, the rest through comparison trackers. Third-party index rows are shown for context and not counted as providers. Every page behind this table: Sources, A100.
§2Specifications
Table 2A100 80GB SXM (A100 80GB PCIe in brackets)
| Memory per GPU (index) | 80 GB |
| Memory | 80GB HBM2e |
| Memory bandwidth | 2,039 GB/s (PCIe: 1,935 GB/s) |
| Max TDP | 400W (PCIe: 300W) |
| FP16 Tensor Core, with sparsity | 624 TFLOPS (PCIe column: 312) |
| Interconnect | NVLink 600 GB/s; PCIe Gen4 64 GB/s |
Quoted from NVIDIA — A100 Tensor Core GPU, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the A100 price
- Age: the A100 is the oldest data center part in the index; its median has roughly halved since the first point of its series.
- Form factor: NVIDIA lists the SXM part at 400W and 2,039 GB/s, the PCIe card at 300W and 1,935 GB/s — a listing's form factor changes both the power it needs and the bandwidth it delivers.
- Memory: 80GB of HBM2e per card, the same capacity as the H100 at roughly half its median hourly rate.
- Term ladders: Civo's 36-month rate is −22% against its on-demand price — a modest discount for a three-year lock-in.
§4Rent or buy
Running one A100 for a month at the median costs $1,278; an eight-GPU node $10,220, or $3,095 at the marketplace floor.
Older parts like the A100 also trade as used or refurbished servers, where price depends on condition; we have no cited figure to publish. Ownership makes most sense when utilisation stays high for years and power is cheap — at 300–400 W per GPU, the A100 fits ordinary air-cooled racks, which widens the hosting options compared with Blackwell parts.
Other models with their own page: GB200, B200, MI355X, H200, H100, MI300X, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.