NVIDIA H100 Systems
NVIDIA H100 Servers
The H100 remains the workhorse for production training and high-throughput inference, and it is sold in two quite different forms. SXM5 modules sit on an NVLink baseboard with NVSwitch, giving 900GB/s between GPUs — that is what large model training needs. H100 NVL PCIe cards drop into a standard server and are usually the better economics for inference and fine-tuning, where jobs fit inside one or two GPUs.
Which of those you buy changes the chassis, the power envelope and the network around it, so the question is rarely "how much is an H100 server" in isolation. We quote the platform, the fabric and the rack power together, because ordering the node without the InfiniBand or the PDU capacity is the most common way an AI build slips a month.
11 configurations below · quotes returned within 48 business hours · purchase orders accepted
NVIDIA H100 Servers we configure and supply
Every system below is quoted to your workload — accelerator count, CPU, memory, storage and fabric are specified together rather than sold as a fixed SKU.
H100 GPU Server
The proven data-center standard for large-scale AI training and inference.
- Proven at enterprise scale
- High memory bandwidth
- Partitionable with MIG
NVIDIA DGX H100
Eight H100 SXM5 GPUs and 640GB HBM3 — the Hopper-generation DGX workhorse.
- Matches an existing estate
- Proven platform
- Cluster-ready networking
Supermicro SYS-821GE-TNHR 8x H100 SXM5 (8U HGX)
Flagship 8-GPU SXM5 node for large-model training and dense inference.
- Maximum GPU bandwidth
- Tested before it ships
- Authorized-channel assurance
Dell PowerEdge XE9680 8x H100 SXM5
Dell-engineered 8-GPU HGX platform with enterprise serviceability built in.
- Fits Dell-standard ops
- Enterprise serviceability
- Authorized sourcing
Supermicro 8x H100 SXM5 Direct Liquid-Cooled (8U HGX)
Liquid-cooled 8-GPU H100 density that cuts cooling power and noise.
- Lower cooling power
- Sustained peak clocks
- Integrated and validated
Supermicro AS-4125GS-TNRT 4x H100 NVL PCIe (Dual EPYC)
Flexible 4U PCIe GPU server on dual EPYC for inference and mixed AI.
- Cost-efficient acceleration
- Configuration flexibility
- EPYC core density
Dell PowerEdge R760xa 4x H100 NVL PCIe (2U)
Dense 2U four-GPU H100 server for space-conscious enterprise deployments.
- High density per U
- Enterprise management
- Authorized provenance
HGX H100 8x SXM5 Rail-Optimized InfiniBand Training Node
Cluster-ready 8-GPU H100 node with eight NDR400 InfiniBand rails.
- Linear scale-out
- Fabric-tuned on delivery
- Cluster-ready building block
Nexus H100 16-GPU NDR InfiniBand Cluster (2× HGX H100 Nodes)
Two tightly-coupled H100 nodes — the right first step into multi-node training.
- Real distributed training
- Delivered fully tested
- Clean path to a pod
Nexus H100 64-GPU SuperPOD-Class Cluster (8× HGX H100 Nodes)
Rail-optimized 64-GPU H100 fabric engineered for serious foundation-model runs.
- Predictable scaling efficiency
- Single accountable supplier
- Owned-economics at scale
Nexus H100 256-GPU Liquid-Cooled SuperCluster (32× 4U HGX Nodes)
256 liquid-cooled H100s in five racks — maximum density, lower power draw.
- Density without the heat ceiling
- Lower operating power
- Plug-and-play delivery
What we need to quote accurately
A configuration quote takes minutes when these are known and days of back-and-forth when they are not. You do not need all of them to start — send what you have.
- The workload: model sizes, training or inference, and expected concurrency
- Node count, and whether this is a single system or a scaling cluster
- Rack power and cooling available per rack, and inlet temperature
- Existing estate — the vendor and management tooling you already run
- Network fabric: InfiniBand, Ethernet, and the speed you are standardised on
- Timeline, and whether the budget is approved or being built
How we quote
- 1
Send the requirement
An email, a bill of materials, or a rough description of the workload. All three work.
- 2
We validate the configuration
Accelerator, chassis, fabric, power and cooling checked against each other before anything is priced.
- 3
Itemised quote within 48 hours
Line-by-line pricing, current availability and lead time, with warranty terms stated.
H100 server — buyer questions
How much does an H100 server cost?
It depends far more on the configuration than on the GPU count. An 8x H100 SXM5 node, the CPU and memory around it, NDR InfiniBand adapters and switching, NVMe for the dataset, and the rack power and cooling to run it are all separate line items, and the fabric alone can be a substantial share on a multi-node build. We do not publish a single figure because it would be wrong for most buyers. Send us the workload and node count and you get an itemised quote.
Should I buy SXM5 or H100 NVL PCIe?
SXM5 if your jobs span more than one GPU — NVLink and NVSwitch are what make an 8-GPU node behave as one large accelerator, and PCIe cannot match that bandwidth. H100 NVL PCIe if your work fits in one or two GPUs, if you are adding accelerated compute to existing rack-mount servers, or if you need a lower power envelope per node. Tell us the model sizes you are training or serving and we will say which applies.
Can you supply H100 systems as a multi-node cluster?
Yes. We hold rail-optimised HGX H100 node configurations and quote them as clusters with the InfiniBand fabric, cabling, management network and rack layout specified together rather than as separate orders.
What lead time should I plan for?
Accelerator lead times move with allocation and are not something anyone should quote you from a web page. We confirm current availability and a committed date with the quote, against the specific configuration you need.
Compare with other platforms
NVIDIA H200 Systems
NVIDIA H200 Servers
View systemsNVIDIA Blackwell Systems
NVIDIA B200 & GB200 Blackwell Servers
View systemsNVIDIA DGX
NVIDIA DGX Systems
View systemsDell PowerEdge
Dell PowerEdge XE9680 GPU Servers
View systemsBrowse the full range of GPU servers and AI clusters, or the component catalogue for drives, memory, optics and spares.