Current-generation HGX nodes.
Dedicated bare metal, B300-class and current SKUs, with the fabric, storage and networking sized to the workload rather than the price list.
Dedicated HGX B300-class nodes in named Tier III data centres, on multi-year terms sized to your workload. Roughly 8 to 10 weeks to deployment, a 20% deposit secures the specific allocation, and the exit is structured before you sign. This is the model for teams whose cloud bill stopped behaving like a variable cost.
Dedicated bare-metal clusters on a term, so the hardware never lands on your balance sheet. You get the capacity and the queue position; the supplier keeps the residual and the refresh risk.
If you are still establishing what the workload actually needs, start with the sizing guides and the comparison set before committing to a term.
Related: GPU as a Service
On-demand cloud is the right answer for spiky, exploratory work and the wrong one for a sustained baseline: at steady utilisation, reserved bare metal undercuts on-demand rates by multiples, and the gap widens with every month of runway. Owning fixes the cost but brings import, deployment, operations and end-of-life onto your desk, and locks capital into an asset that depreciates on NVIDIA's schedule, not yours.
Leasing is the deliberate middle: dedicated hardware and bare-metal performance, a defined multi-year rate a finance team can plan around, no import or disposal problem, and a facility someone else is accountable for running. For most sustained training and inference baselines, it is the structure that survives contact with both the engineers and the CFO.
Because the desk is buyer-side, the sizing is honest: if your baseline does not justify a lease, or an owned cluster prices better on your numbers, that is what we tell you.

Dedicated bare metal, B300-class and current SKUs, with the fabric, storage and networking sized to the workload rather than the price list.
You know the building, the operator, the power and the cooling before signature. Sovereign and residency-constrained variants available.
Term and rate pressure-tested against your roadmap. A 20% deposit secures the specific allocation; deployment runs roughly 8 to 10 weeks.
Renewal, extension and refresh terms negotiated at signature, when you have leverage, not at expiry, when you have none.
The decision deserves more than a rule of thumb. We keep the working comparisons public: buying versus leasing GPUs and cloud versus bare metal, with the break-even logic spelled out. Bring your utilisation and runway and we will run them on your numbers in the first call.
A cleared file gets real numbers. That is the whole mechanism, and it is why the compliance work comes before the price conversation.
The desk sits at the gate on your side, so the contract runs supplier to buyer
Nothing directly. You contract straight with the supplier on their terms, and our compensation comes from the supply side under disclosed arrangements. The engagement costs you the time it takes to state the requirement.
Roughly eight to ten weeks from signature for a racked HGX deployment in a named Tier III facility, depending on the site and configuration. A 20% deposit secures the specific allocation.
Multi-year reserved terms, structured to the workload. The honest trade: longer terms price better but carry generation risk, and we pressure-test that trade against your roadmap rather than defaulting to the longest term the market will sign.
Lease when you want capacity without the balance-sheet weight, the import work or the end-of-life problem. Buy when utilisation is long and predictable enough to justify owning the asset. We run both models on your numbers; the comparison page walks through the logic.
A named Tier III certified AI data centre, identified before you sign, never a region on a map. Facility, operator, power and cooling are all verifiable up front.
Renew on the then-current generation, extend, or walk away clean. The exit is part of the structure from day one, which is exactly what an on-demand cloud bill never gives you.
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.
Your enquiry has landed with the desk. Acknowledged within one hour.