The AMD MI300X (CDNA 3, 192 GB HBM3) rents from $1.84/hr, averaged across 12 cloud providers, updated 2026-11-15. On-demand pricing has fallen 2% over the past 90 days.
AMD's H100/H200 competitor — its large 192 GB memory makes it attractive for memory-bound inference of big models. CDNA 3 · 192 GB HBM3 · 5.3 TB/s memory bandwidth · launched 2023.
▮ on-demand /hr · ▮ spot /hr
| Date | On-demand | Spot |
|---|---|---|
| 2026-11-15 | $2.86 | $1.84 |
| 2026-11-14 | $2.86 | $1.84 |
| 2026-11-13 | $2.86 | $1.84 |
| 2026-11-12 | $2.88 | $1.84 |
| 2026-11-11 | $2.88 | $1.84 |
| 2026-11-10 | $2.87 | $1.84 |
| 2026-11-09 | $2.87 | $1.84 |
| 2026-11-08 | $2.93 | $1.84 |
| 2026-11-07 | $2.87 | $1.84 |
| 2026-11-06 | $2.87 | $1.84 |
| 2026-11-05 | $2.86 | $1.84 |
| 2026-11-04 | $2.85 | $1.84 |
| 2026-11-03 | $2.74 | $1.84 |
| 2026-11-02 | $2.84 | $1.84 |
| GPU | On-demand /hr | Spot /hr | Memory | |
|---|---|---|---|---|
| MI300X | $2.86 | $1.84 | 192 GB HBM3 | |
| B200 | $8.28 | $4.80 | 192 GB HBM3e | MI300X vs B200 → |
| H200 | $4.59 | $3.36 | 141 GB HBM3e | MI300X vs H200 → |
| H100 | $4.01 | $2.40 | 80 GB HBM3 | MI300X vs H100 → |
| MI355X | $8.25 | $3.56 | 288 GB HBM3e | MI300X vs MI355X → |
| MI325X | $3.11 | $— | 256 GB HBM3e | MI300X vs MI325X → |
The MI300X averages $2.86/hr on-demand, $1.84/hr spot, and $2.18/hr reserved across 12 cloud providers (updated 2026-11-15).
The lowest tier is spot at about $1.84/hr on average; reserved (1–3 year commitment) runs ~$2.18/hr. Compare live per-provider rates on the Bargo GPU price tracker.
At on-demand $2.86/hr, one MI300X runs about $2,088/month (730 hours); reserved pricing lowers that to roughly $1,591/month.
On-demand, the MI300X is $2.86/hr vs the H100 at $4.01/hr — cheaper per hour. See the full MI300X vs H100 comparison for specs and cost-per-job context.
Spot ($1.84/hr) is interruptible spare capacity at a discount to on-demand ($2.86/hr); use spot for checkpointed or fault-tolerant jobs and on-demand or reserved for long, uninterrupted runs.
Cross-provider averages from public cloud GPU listings. Informational only.