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
Blog · How this market actually works

Notes from the buyer’s side of the table.

How AI infrastructure actually gets bought: allocation, terms, borders and the traps in between. Written from live deals, not from press releases.

Pricing and the arithmetic

What things cost, with the numbers shown. The pages buyers screenshot.

GPUaaS or own it: the B300 numbers
GPUaaS or own it: the B300 numbers$840,960 to rent a node for 3 years against $677,784 to own, colocate and cable it.
What a 4,096-GPU B300 cluster costs
What a 4,096-GPU B300 cluster costs$288m of hardware, $50m of colocation, and the 17% where the schedule risk lives.
Reserved or on-demand: the break-even
Reserved or on-demand: the break-evenRoughly 71% utilisation at a 40% on-demand premium. Above it reserve, below it do not.
Landed cost, and the cheap quote that is not
Landed cost, and the cheap quote that is notEx-works stops at the factory door. The border is where buyers find out.
Compute is the door, not the building
Compute is the door, not the buildingThe GPU line is the biggest single cost and rarely the decisive one.

Contracts and terms

The paper around the deal, where more money moves than in the rate.

Five GPUaaS clauses that beat the rate
Five GPUaaS clauses that beat the rateFloor, exit, preemption, refresh, egress. In that order of impact.
The term trap
The term trapEvery reserved quote improves with length. That is exactly why to be suspicious.
Rent, reserve or own
Rent, reserve or ownThree doors, each with a salesman in front of it not paid to mention the other two.
How to vet a sourcing desk, including ours
How to vet a sourcing desk, including oursSix questions. The answers, and how long they take, tell you everything.
The middleman myth
The middleman mythFollow who pays them. Then watch how long the answer takes.

Facilities, power and delivery

Where the hardware physically goes, and what it takes to get it running.

Colocation above 40 kW per rack
Colocation above 40 kW per rackWhere the shortlist collapses, and what high-density space actually costs.
GPU lead times in 2026, honestly
GPU lead times in 2026, honestly7 days ex-works is routinely a 5 to 8 week project. Here is the gap.
The network nobody prices
The network nobody pricesOptics bought late and in a hurry are bought at the worst possible price.
Servers, not cards
Servers, not cardsThe market quotes, ships and racks 8-GPU machines. Card-price budgets meet that later.

Access, compliance and placement

Why some buyers get allocation and others get screening quotes.

Who gets GPU allocation
Who gets GPU allocationSuppliers optimise for risk, not budget. Allocation follows the destination file.
Compliance is the moat
Compliance is the moatWhatever is expensive to fake becomes a moat for whoever does it properly.
Sovereignty as a buying decision
Sovereignty as a buying decisionDecide the map first and everything after it moves faster.
Australia has the GPUs and no one to run them
Australia has the GPUs and no one to run themHalls going up faster than the tenant queue forms.
What this is

Written from the desk, not from a keyword list.

These pieces come out of conversations that actually happen during procurement: the question a buyer asks three weeks in, the line item nobody budgeted for, the reason a supplier went quiet. They are written to be useful to somebody mid-purchase rather than to rank for a phrase.

That means they occasionally argue against our own commercial interest, because a buyer who understands the market is a better counterparty than one who does not. Where a piece cites a number, the source is named on the page.

Straight answers

Asked first, answered straight.

How often do you publish?

When there is something worth saying from live deal flow rather than on a content schedule. Every piece comes out of a real procurement conversation.

Can I republish or cite these articles?

Yes, with attribution and a link back to the original. If you would like a comment or a quote for something you are writing, contact the desk directly.

Do you cover markets outside GPUs?

The adjacent layers, yes: networking and optics, power and cooling, storage, and the import and compliance path. They are part of the same purchase and they carry most of the schedule risk.

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