Live-ish cloud GPU prices for H100, H200 and B200 across the providers that matter, and a three-quote service for Supermicro NVIDIA HGX systems delivered into Europe. We sit on the buyer's side of the table.
Get three quotes See cloud pricesWe do not sell hardware or cloud. We source options from integrators and providers and you pick.
Tell us the workload once. We return comparable quotes for cloud, dedicated and owned options.
UK and EU delivery, VAT, and data-residency handled in the quote rather than discovered after.
Indicative on-demand list prices in USD, checked 19 September 2026. Marketplace and spot prices move daily and reserved contracts run 30–60% lower. Always confirm on the provider's page.
| Provider | Type | From, $/GPU-hr | Notes |
|---|
Median on-demand across tracked providers this month: H100 about $3.25, H200 about $4.72, B200 about $6.52 per GPU-hour. Hyperscaler on-demand for the same GPUs runs $7–14. Sources: AIMultiple GPU index, gpuperhour.com, getdeploying, Thunder Compute. Some provider links may be referral links; they never change the price you pay.
The configurations most buyers in Europe are asking about. We source these from authorised Supermicro integrators and Taiwan-direct channels with delivery targets of four to six weeks. Prices are quoted, not listed, because they move with GPU allocation.
| System | GPUs | Best for | |
|---|---|---|---|
| SYS-821GE-TNHR 8U · Intel | 8× HGX H100 / H200 | Single-node training and fine-tuning, inference at scale | Quote |
| AS-8125GS-TNHR 8U · AMD EPYC | 8× HGX H100 / H200 | Same class as above with AMD CPUs, often better lead time | Quote |
| HGX H200 SuperCluster 32 nodes · air-cooled | 256× H200 | Large-model training, cloud-scale inference, spine-leaf 400G fabric | Quote |
| Liquid-cooled AI SuperCluster rack-scale · DLC | 256× H100 / H200 | Dense deployments where power and cooling are the constraint | Quote |
| GB200 NVL72 rack-scale · Blackwell | 72× B200 + 36× Grace | Trillion-parameter inference and frontier training | Quote |
| H200 Tensor Core GPU component | 141GB HBM3e | Upgrades and spares for existing HGX chassis | Quote |
The honest short version, before anyone sends you a spreadsheet.
SuperServer.ai started as a Supermicro reseller for Europe. It now runs as an independent sourcing desk: we compare cloud prices publicly and, when you ask for quotes, we go to integrators and providers on your behalf and return like-for-like options. We are paid by the supplier you choose, never by you, and the price you receive is the supplier's direct price.
More about how we work. SuperServer.ai is owned and operated by Lovell Group Ltd and is independent of Supermicro and NVIDIA.
One form. We come back within two working days with up to three comparable options across cloud, dedicated hosting and purchase, with lead times and total cost side by side.
We use your details only to source these quotes and to reply. No newsletter unless you ask. Privacy.