What ‘qualifying deployments’ actually means

We quote “~$5/GPU-hour for qualifying long-term commitments” — and we deliberately do not say “all our GPUs cost $5/hr.” That honesty is a feature, not a legal department’s invention. This article explains exactly what “qualifying” means, what moves your real price, and how a quote gets built.

First, the target rate describes deployments that look a certain way: sustained, multi-GPU baseload workloads committed for 12–24 months, in regions where capacity is secured and contracted. If your deployment matches that profile, ~$5/GPU-hour is the starting point of the conversation — not the final number, but the anchor.

01 / THE FIVE THINGS THAT MOVE THE NUMBER

Five variables dominate real pricing. In order of impact:

• 01GPU type. A B300 is more compute than a B200 — more memory, more throughput, denser deployments. The target rate is tiered by the silicon your workload is built on, not blended into one fictional number.
• 02Node configuration. Full 8×GPU nodes with high-bandwidth interconnect price differently from smaller builds. The spec we’d sign gets quoted against the spec you actually need — no paying for a node you wouldn’t run.
• 03Quantity. Larger fleets price better per unit: deployment and networking costs amortize over more GPUs. Single-digit GPUss rarely clear the “qualifying” bar for the target rate.
• 04Region. Power, cooling, datacenter availability and networking differ by region, and pricing follows the real deployment geography — wherever your capacity is actually secured.
• 05Commitment. A 24-month commitment prices deeper than a 12-month one. The longer the horizon we’re securing capacity against, the better the economics we can pass through.

02 / WHAT THE RATE CLOSES

The commitment rate covers your dedicated capacity for the full term — fixed, contracted, known. Node configuration, networking, SLA, uptime target, provisioning time and support model are specified in the agreement before you sign, so the number finance models is the number you pay.

03 / HOW A QUOTE ACTUALLY GETS BUILT

A quote starts with your reality, not our inventory: current GPUs, GPU type, what you’re paying per GPU-hour, monthly spend, workload, and commitment preference. We model that against the baseload shape of your fleet. If the deployment doesn’t qualify for the target rate, we tell you what it would take to get there — or that the model isn’t the right fit.

What we will not do: quote a rate we can’t sustain, or imply inventory we don’t have. B200/B300 capacity is available through qualifying deployments — if we can’t secure the hardware your timeline needs, you’ll hear that before anything else.

“Qualifying deployments” isn’t a loophole — it’s the definition of the deployments the model serves. Sustained, multi-GPU, committed compute gets the target rate because that compute is what the whole company is built to deliver.