Atlas pricing. Per-second metered. No hidden per-vCPU markup.
Transparent, published SKUs across every Atlas service. You activate a service per Stripe subscription — until then, your tenant lives on free shaped stubs (real APIs, deterministic data).
GPU Compute (per node-hour, 8x GPU unless noted)
| SKU | On-demand | Spot | Reserved 3y |
|---|---|---|---|
| HGX H100 (8x SXM5, 80 GB) | $49.24 | $19.71 | $6.16 |
| HGX H200 (8x SXM5, 141 GB) | $50.44 | $20.93 | $6.31 |
| HGX B200 (8x SXM6, 192 GB) | $68.80 | $34.11 | $8.60 |
| HGX B300 (8x SXM6, 288 GB) | $70.40 | $34.88 | $8.80 |
| GB200 NVL72 (per GPU) | $42.00 | — | $10.50 |
| GB300 NVL72 (per GPU) | $42.00 | — | $10.50 |
| HGX A100 (8x SXM4, 80 GB) | $21.60 | $9.65 | $2.70 |
| GH200 (per Grace+H100) | $6.50 | — | $6.50 |
| L40 | $10.00 | $6.27 | $1.25 |
| L40S | $18.00 | $7.88 | $2.25 |
| RTX Pro 6000 | $20.00 | $11.09 | $2.50 |
CPU Compute (per node-hour)
| SKU | On-demand | Spot |
|---|---|---|
| AMD Genoa 9274F (24c, high freq) | $6.42 | $3.50 |
| AMD Genoa 9454 (48c) | $7.78 | $3.32 |
| AMD Genoa 9454 high-stream | $8.86 | $4.36 |
| AMD Genoa 9654 (96c) | $7.54 | $4.23 |
| AMD Turin 9655P (96c) | $8.18 | $4.62 |
| AMD Turin 9655P high-stream | $9.31 | $5.25 |
| Intel 8562Y+ (32c, high freq) | $5.31 | $3.27 |
| Intel 6342 (24c) | $3.36 | $2.29 |
Storage
| SKU | Price |
|---|---|
| Object Hot | $0.060 / GB-mo |
| Object Warm | $0.030 / GB-mo |
| Object Cold | $0.015 / GB-mo |
| Object Archive | $0.0125 / GB-mo |
| Parallel file system (POSIX scratch) | $0.070 / GB-mo + throughput |
Networking
| SKU | Price |
|---|---|
| Public IP / EIP | $4 / mo |
| BYOIP setup (one-off) | $4 / range |
| Direct Connect 10G — primary | $1,250 / mo |
| Direct Connect 10G — redundant | $1,500 / mo |
| Direct Connect 100G — primary | $12,500 / mo |
| Direct Connect 100G — redundant | $15,000 / mo |
| Direct Connect 400G — primary | $50,000 / mo |
| VPC / NAT / egress | Included |
Inference (fast-weight-loader backed endpoints)
| SKU | Price |
|---|---|
| Endpoint overhead (per replica-hour) | $0.10 |
| GPU time (per replica-hour) | See GPU SKU |
Ops & Observability
| SKU | Price |
|---|---|
| Fleet Lifecycle Controller (orchestrator) | Included |
| Node Lifecycle Controller (health monitor) | Included |
| Observability — metrics ingest (AMP) | $0.30 / GB |
| Observability — logs ingest (CW Logs) | $0.50 / GB |
| Observability — traces ingest | $0.20 / GB |
| Fast weight loader — serialize job | GPU time only |
Free / included
- Atlas Kubernetes control plane
- Fleet Lifecycle Controller
- Node Lifecycle Controller