SSP, Strategic Supply Partners Live inventory
The desk
GPU procurementNodes and clusters, landed to your rack ● Live inventoryVerified lines, indicative pricing
By platform
B300 and GB300Current-generation Blackwell H200 and H100New stock and fleet rotation All platformsThe full catalogue
Rest of the stack
Memory, servers and storageThe kit around the accelerators Networking and opticsTransceivers, switching, cabling
Capacity without capex
GPU-as-a-serviceReserved, committed spend, on demand Bare-metal leasingDedicated clusters on a term
Decide first
Cloud vs bare metalWhere each one actually wins Reserved vs on demandThe break-even, calculated Cluster cost calculatorThe whole bill, not the GPU line
Space and power
Colocation and powerSingle racks to 100 MW Above 40 kW per rackWhere the shortlist collapses
Where
United States Europe Asia Pacific Middle East and Africa All regions →
Supply side
Have capacity to fill?Tell us what you hold
Analysis
InsightsWhat things cost, with the numbers Sizing guidesMemory, nodes, power, cost ComparisonsBuy, lease or rent
Start here
How an engagement worksThe first week, held to dates Find your fitThree questions, the right desk Glossary
Talk to the desk
Integration · InfiniBand

InfiniBand fabric for AI clusters.

InfiniBand remains the reference interconnect for large-scale distributed training, and it is also where cluster budgets most often go wrong, because almost nobody counts the optics before they commit.

Tell us what you need See live inventoryAcknowledged within one hour, first sourcing pass within one business day.
At a glance

What decides this requirement.

Wins on
Large-scale distributed training with heavy collective operations
Competes with
Modern high-speed Ethernet, which has closed much of the gap
Budget trap
Transceivers and cabling in quantities that surprise buyers
We handle
Topology, switching, optics and spares, priced with the cluster
When InfiniBand is the right answer

Scale and collectives decide it.

For large multi-node training where gradient synchronisation dominates, InfiniBand's latency and collective offload capabilities remain the reference standard, and at scale the difference in realised throughput is real rather than theoretical.

Modern high-speed Ethernet has closed much of that gap, particularly with the lossless and congestion-control features developed for AI fabrics, and it frequently wins on cost and on operational familiarity. For many workloads short of frontier scale, it is the better commercial answer.

We are not attached to either. The honest question is job size, collective intensity and what your team can operate, and we put both configurations side by side with real numbers rather than defaulting.

High-speed switch ports fully populated with optical transceivers and fibre
The optics arithmetic

This is where budgets break.

Every node needs multiple fabric ports. Every port needs a transceiver at each end and a cable between. Multiply that across a real topology and you are buying transceivers by the hundreds, plus spares, because a cluster idle on a missing part is burning far more than the part costs.

Third-party optics markups are among the steepest in the industry, and optics bought late and in a hurry are bought at the worst possible price. Sourced with the cluster and buyer-side, this is one of the easiest large savings available.

We price the fabric with the compute rather than after it, and we say plainly where compatible third-party parts are appropriate and where support terms genuinely require OEM. Detail on the networking desk.

Where to next

Start from what is verified.

Current verified lines with quantities, lead times and indicative pricing are public on the live inventory. Anything not listed becomes a sourcing requirement with a first pass inside one business day.

Networking and opticsLLM trainingThe network nobody pricesLive inventoryProcurement desk
Tell us what you need See live inventoryAcknowledged within one hour, first sourcing pass within one business day.
Straight answers

Asked first, answered straight.

InfiniBand or Ethernet?

Job size and collective intensity decide it. Frontier-scale distributed training still favours InfiniBand; a great many workloads short of that run well on modern high-speed Ethernet at better cost. We price both.

How many transceivers will we need?

Hundreds for a serious multi-node fabric, plus spares. It is the most commonly underestimated line in a cluster budget and worth counting before you commit rather than after.

Can we use third-party optics?

Frequently yes, verified for the platform and warranted, at a fraction of OEM list. Where support terms genuinely require OEM parts we say so rather than quietly substituting.

Should we buy the fabric with the GPUs?

Yes. Optics ordered after the nodes arrive means a cluster standing idle in its most expensive month, and buying in a hurry means buying at the worst price.

Talk to the desk

Working through this on a real requirement?

Twenty minutes with the desk, no pitch and no quote at the end of it. Tell us roughly what you need and we will come back within one business day.

Acknowledged within one hour, first sourcing pass within one business day.
✓

Sent. We are on it.

Your enquiry has landed with the desk. Acknowledged within one hour.