The cheapest MI300X on the table right now is $1.11 per GPU-hour at Azure (spot), out of 7 offers from 5 providers. The MI300X carries 192 GB of VRAM on AMD's CDNA3 generation.
On-demand pricing for the same card spans $2.59 to $6.00 per GPU-hour — a 2.3× spread between the cheapest and the most expensive provider for identical silicon. That gap is the entire reason this table exists.
Cheapest by pricing model: on-demand $2.59 at DigitalOcean · spot $1.11 at Azure · reserved $1.91 at DigitalOcean.
| Provider | Tier | $/GPU-hr | Pricing | Node | Commit | In stock | |
|---|---|---|---|---|---|---|---|
| Azure | dedicated | $1.11 | spot | 8× GPU | — | — | rent |
| DigitalOcean | dedicated | $1.91 | reserved | 8× GPU | 12 mo | — | rent |
| DigitalOcean | dedicated | $2.59 | on-demand | 8× GPU | — | — | rent |
| Seeweb | dedicated | $2.74 | on-demand | 1× GPU | — | — | rent |
| Crusoe | dedicated | $3.45 | on-demand | 1× GPU | — | — | rent |
| Azure | dedicated | $6.00 | on-demand | 8× GPU | — | — | rent |
| Oracle Cloud | dedicated | $6.00 | on-demand | 8× GPU | — | — | rent |
How much does an MI300X cost per hour?
As of the latest scrape, $1.11 per GPU-hour at Azure is the cheapest MI300X offer across 5 tracked cloud providers. The cheapest on-demand rate is $2.59 per GPU-hour at DigitalOcean, and the most expensive on-demand rate is $6.00 at Oracle Cloud.
Which cloud is cheapest for the MI300X?
Azure at $1.11 per GPU-hour on spot capacity. Prices move constantly, so this page is regenerated from the live scrape every 15 minutes. Note that marketplace tiers are peer hardware and are not directly comparable to dedicated capacity.
Is spot MI300X capacity cheaper?
Yes — the cheapest spot MI300X is $1.11 per GPU-hour at Azure, -57% versus the cheapest on-demand rate of $2.59. Spot instances can be reclaimed by the provider at any time, so they suit checkpointed training and batch inference rather than long-lived services.
How many GB of VRAM does the MI300X have?
192 GB, on the CDNA3 architecture. Multi-GPU instances multiply that: an 8× MI300X node exposes 1536 GB of GPU memory in total.