Skip to content

SkyForge · pricing

Pay by the second. Stop, and you stop paying.

One per-second meter across everything we rent — no minimums, no tier gates, no reserved commitment to get started. One SkyForge API key across every instance, live spend in the console, terminate anytime.

per-second billing · no minimums · one key, every instance · terminate anytime

01On-demand

Two machines, one meter.

A Blackwell dev box for small models and your own code, and a 96 GB inference workhorse for the catalog — both billed per second from ready to terminate.

DGX Spark

dev & small models

$0.59/ hr

GB10 Blackwell units in 1, 2, and 4-unit shapes — 128 GB unified memory per unit, root SSH + Docker, CUDA 12.8. Priced per instance, billed per second.

RTX PRO 6000

inference workhorse

$2.25/ GPU-hr

Production inference behind an OpenAI-compatible /v1 API. 96 GB GDDR7 per GPU, 1–8 GPUs per node, billed per second — no commitment.

ShapeMemoryRate
DGX Spark · 1 unit128 GB unified$0.59 / hr
DGX Spark · 2 units256 GB unified$1.10 / hr
DGX Spark · 4-unit cluster512 GB unified$2.10 / hr
RTX PRO 6000 · 1–8 GPUs96 GB GDDR7 per GPU$2.25 / GPU-hr

DGX Spark is priced per instance; RTX PRO 6000 per GPU. Both are billed per second while the instance is running.

02How the meter works

Simple, honest, and yours to control.

No hourly rounding, no plans to graduate into, and nothing running that you didn't ask for. The meter tracks real usage — and the dashboard shows it live.

01

Pay by the second

The meter runs per second while an instance is up — not rounded to the hour. You pay for exactly the time you use.

02

Stop and you stop paying

Terminate an instance and the meter stops. There's nothing left running and nothing left to pay for.

03

No minimums, no tier gates

No monthly commitment and no plan to upgrade into. Every capability is available from your first instance.

04

One API key, every instance

A single SkyForge API key works across all of your instances and OpenAI-compatible endpoints — no per-instance keys to juggle.

05

Live spend in the dashboard

See accrued, burn rate, and projected spend update live in the console, so the bill is never a surprise.

06

Terminate anytime

Spin up and tear down whenever you like. No lock-in, no cancellation flow — the instance is yours to end on your terms.

03Reserved & committed

Steady workload? Reserved is a conversation.

Committed-use pricing and dedicated capacity for sustained production, quoted to your usage and allocated through our team — including reserved Blackwell B200 / B300. There's no public rate card; tell us the workload and we'll size it.

Talk to us

04FAQ

Questions about billing.

05Get started

Launch now, or talk to us about reserved.

On-demand is self-serve from the console — no sales call. Running steady production? Tell us what you're building and we'll size committed capacity.

Reserved or committed capacity?

Tell us the workload and we'll follow up with a quote.

What's the use case?