AMD Instinct MI300X rental price per hour
Series 1 · GPU prices · AMD · snapshot 16 Oct 2026
AMD's MI300X carries 192 GB of HBM3 and rents for $2.99 an hour at the median — below the H100's $3.49. Per gigabyte of GPU memory it is the cheapest data center accelerator in the index.
Figure 1MI300X: median rental price and published range, USD per GPU-hour, 2025-01-01 to 2026-10-16
§1Reading the MI300X market
The effective floor is TensorWave at ~$1.71, a sales-quote rate with no self-serve checkout. The self-serve cluster sits just above: Vultr ~$1.85, Hot Aisle ~$1.99, Runpod and GPU Outlet ~$2.39, DigitalOcean ~$2.59. DigitalOcean also publishes a 12-month rate of $1.91, −36% against the median.
Azure prices its ND96isr MI300X v5 at $48–96 an hour depending on region, $6.00–$12.00 per GPU — a regional spread wider than most providers' entire price range, and the reason the ceiling looks out of line with the rest of the market.
The median of $2.99 matches GetDeploying's tracked figure. It started the series at $4.20 in Jan 2025; the latest uptick ($2.79 → $2.99 on 25 Sep 2026) is within the week-to-week noise the tracker sees on this part.
Table 1MI300X prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| TensorWave | Sales quote | ~$1.71 | sales quote, no self-serve checkout | 2026-09-25 |
| Vultr | On-demand | ~$1.85 | 2026-09-25 | |
| DigitalOcean | Reserved, 12 mo | $1.91 | 2026-09-25 | |
| Hot Aisle | On-demand | ~$1.99 | 2026-09-25 | |
| Runpod | On-demand | ~$2.39 | 2026-09-25 | |
| GPU Outlet | On-demand | ~$2.39 | 2026-09-25 | |
| DigitalOcean | On-demand | ~$2.59 | 2026-09-25 | |
| GetDeploying median | Third-party index | $2.99 | 2026-09-25 | |
| Microsoft Azure | On-demand | $6.00–$12.00 | ND96isr MI300X v5, $48-96/hr for 8 GPUs depending on region | 2026-09-25 |
Rows are the prices recorded in the tracker note for this model; 4 of the 9 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, MI300X.
§2Specifications
Table 2Instinct MI300X (OAM module)
| Memory per GPU (index) | 192 GB |
| Architecture | CDNA 3, 8 accelerator complex dies |
| Memory | 192GB HBM3 per accelerator |
| Platform | 8 OAM modules on a UBB 2.0 baseboard, 1.5TB total |
| Memory bandwidth, board power | not stated on the cited page |
Quoted from AMD Instinct documentation — MI300X, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the MI300X price
- Memory per dollar: 64 GB of HBM per dollar-hour at the median, against 23 GB for the H100 — attractive for serving large models that are memory-bound.
- Software stack: AMD accelerators run on ROCm rather than CUDA, so teams with CUDA-only code face porting work before they can use the cheaper hours.
- Platform: AMD documents eight OAM modules on a UBB 2.0 baseboard, 1.5 TB of HBM3 per platform, so the natural unit is an eight-GPU node.
- Regional hyperscaler pricing: one Azure instance spans a 2× range by region.
§4Rent or buy
An eight-GPU MI300X node costs $17,462 a month at the median and $11,154 on DigitalOcean's 12-month term.
AMD does not publish system pricing for the MI300X platform and we have no dated quote to cite, so ownership cost is left blank rather than guessed. With the MI355X now on the market, buying MI300X hardware depends on what inventory suppliers still hold; the form below checks it.
Other models with their own page: GB200, B200, MI355X, H200, H100, A100, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.