AMD Instinct MI355X rental price per hour
Series 1 · GPU prices · AMD · snapshot 16 Oct 2026
The MI355X rents for $7.00 per GPU-hour at the median, a little above the B200's $6.25. Its prices are the most tightly bunched of any model tracked: the ceiling is only 1.4× the floor.
Figure 1MI355X: median rental price and published range, USD per GPU-hour, 2025-10-01 to 2026-10-16
§1Reading the MI355X market
Daytona now lists the part on demand at $5.99, separate from its $3.44 preemptible tier — the first confirmed on-demand rate under $7 and a drop of about 16% from the previous $7.10 floor. DigitalOcean added the MI355X this cycle as well, but only as a spot SKU at $4.50, which does not count toward the on-demand floor.
The middle of the market is three providers within ten cents: OVHcloud and Vultr ~$7.10, CoreWeave ~$7.20. Oracle bare metal at $8.60 is the ceiling. CoreWeave's 1-year reservation at $4.25 is the only published term price and the deepest discount in the index, −39% against the median.
The last median move came on 18 Sep 2026, from $7.50 to $7.00, as the new floor pulled the cluster down. With so few sellers, a single new listing can shift the median by tens of cents.
Table 1MI355X prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Daytona | Preemptible | $3.44 | 2026-09-18 | |
| CoreWeave | Reserved, 12 mo | $4.25 | 2026-09-18 | |
| DigitalOcean | Spot | $4.50 | spot SKU only | 2026-09-18 |
| Daytona | On-demand | $5.99 | 2026-09-18 | |
| OVHcloud | On-demand | ~$7.10 | 2026-09-18 | |
| Vultr | On-demand | ~$7.10 | 2026-09-18 | |
| CoreWeave | On-demand | ~$7.20 | 2026-09-18 | |
| Oracle Cloud | On-demand | $8.60 | bare metal | 2026-09-18 |
Rows are the prices recorded in the tracker note for this model; 4 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, MI355X.
§2Specifications
Table 2Instinct MI355X
| Memory per GPU (index) | 288 GB |
| Memory | 288 GB HBM3E |
| Memory bandwidth | 8 TBps |
| Power | 1.4 kW; requires a liquid-cooling loop |
| FP4 | up to 10 petaFLOPS (sparsity basis: see source) |
| Air-cooled sibling | MI350X, roughly 1 kW |
Quoted from The Register — AMD MI355X, 12 Jun 2025, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the MI355X price
- Liquid cooling: The Register reports the MI355X at 1.4 kW per GPU, requiring a liquid loop; the air-cooled MI350X runs at roughly 1 kW. Hosting is restricted to liquid-ready halls.
- Memory: 288 GB HBM3E at 8 TB/s per GPU — 1.6× the B200's 180 GB and more than two H200s combined (2 × 141 GB).
- Concentrated supply: the on-demand market is a handful of clouds quoting near-identical rates, which explains the narrow spread.
- Term discount: CoreWeave's reservation undercuts every on-demand offer, including Daytona's floor.
§4Rent or buy
Eight MI355X GPUs rented continuously cost $40,880 a month at the median, or $24,820 at CoreWeave's 1-year rate — a gap of $16,060 a month for committing.
No MI355X system price appears in our sources, so none is shown. Owning one also means owning the cooling problem: at 1.4 kW per GPU an eight-GPU node exceeds 11 kW for the accelerators alone (arithmetic from the reported TDP), before CPUs and networking.
Liquid-cooled AI capacity tracked by country: AI campus census.
Other models with their own page: GB200, B200, H200, H100, MI300X, A100, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.