AI Servers
Seraya supplies AI servers configured to your workload. Card count, chassis, power and cooling are agreed per order and stated in the quotation.
Open lineSeraya supplies servers, workstations, storage and networking as complete systems. Single GPU cards and SXM modules are quoted separately on the GPU line, because they are not interchangeable with a finished system.
Single cards and SXM modules are quoted on the GPU line, separately from the complete systems on this page. A module needs an HGX baseboard; a card needs a free PCIe slot.
Seraya supplies complete systems — workstations, servers, storage and networking — on a B2B quotation basis. Configuration is confirmed per order. Prices are not published.
Seraya supplies AI servers configured to your workload. Card count, chassis, power and cooling are agreed per order and stated in the quotation.
Open lineWorkstations built around one or two accelerators, for teams that need the work off shared infrastructure and quiet enough for an office.
Open lineBuilt for resellers, retailers and esports venues buying in volume. This is a B2B supply line, not retail single-unit sales.
Open lineStorage and network infrastructure sized to match the compute it serves — pools and capacity tiers for the data, leaf/spine switching or InfiniBand for the fabric.
Open line
Use the systems RFQ form, or email
felix.mm@seraya.systems with the subject tag
[Systems RFQ].
Every configuration on this page is an example, not a validated build. Compatibility, firmware and availability are confirmed per order.
General-purpose rack nodes for virtualization, databases and mixed workloads.
Reference image
1U/2U rack server · Reference
Reference image
2U rack server · Reference
Reference image
Rack server · Reference
Complete servers configured for inference, fine-tuning and large-model training.
Reference image
GPU server · 2 cards · Reference
Reference image
GPU server · 4× H100 NVL cards · Reference
Reference image
GPU server · 8 SXM modules · Reference
All-flash pools, capacity tiers and backup targets, NVMe through SATA.
Reference image
Shared all-flash pool node · Reference
Reference image
Capacity all-flash node · Reference
Reference image
High-density disk shelf node · Reference
25 / 100 / 400 GbE switching and InfiniBand fabric for AI clusters.
Reference image
1U leaf switch · Reference
Reference image
1U/2U spine switch · Reference
Reference image
InfiniBand switch and HCA set · Reference
Engineering and content workstations, quiet enough for the office.
Reference image
Single-socket tower · Reference
Reference image
Dual-GPU tower · Reference
Reference image
Compact workstation · Reference
Capacity shown is raw installed media (drive count × drive capacity), not usable host capacity. Usable capacity depends on the redundancy and file-system layout agreed for the order.
Give us the system type, how many systems, and the cards per system if that applies. The two counts are quoted separately.
| Comparison point | New | ITAD / certified pre-owned |
|---|---|---|
| Cost basis | List or allocation pricing | Recovered cost — normally a discount to new |
| Performance & life | Full service life, current firmware | Same silicon; shorter remaining life, older firmware |
| Lead time | Driven by allocation and configuration | Driven by what has been recovered and tested |
| Payment & warranty | Deposit or credit terms | Set case by case against the grading report |
A single node and a rack-scale rollout do not quote the same way.
In stock, on order and on allocation carry different lead times.
New, refurbished and ITAD are priced, paid for and warranted differently.
All three are stated in writing in every quote.