Cost model

What a reference run costs on each platform, where those numbers come from, and why they do not decide the recommendation.

Read this before reading the numbers

The three figures are priced in different units. Quantum is billed per shot; AI/ML and HPC are billed per compute-hour. Putting them side by side is useful for order of magnitude and actively misleading if read as a like-for-like comparison.

More importantly: cost does not determine quantum advantage. A problem with an exponential speedup can be expensive per run and still be the right answer. A problem with no speedup is not made suitable by being cheap. Feasibility and the speedup class decide the verdict; these rates are context.

Your last evaluation

How each platform is billed

Quantum

Per shot

You are billed per circuit execution. A job is many shots, so cost scales with how many times you repeat the circuit, not with wall-clock time.

Azure AI / ML

Per compute-hour

You are billed for the VM while it is running, whether it is busy or idle. Cost scales with time and instance size.

Azure HPC / GPU

Per compute-hour

Same billing shape as AI/ML, at much higher hourly rates for large multi-GPU or InfiniBand-connected nodes.

Where the rates come from

Live retail prices. Compute rates are looked up from the public Azure retail pricing service at evaluation time, so they track list price rather than a number typed into this repository months ago.
Pinned fallback. If that lookup fails or times out, a pinned rate table is used instead. The figures stay sensible, but they can drift from list price.
Quantum rates are per-shot provider rates, applied to a hardware-grounded representative circuit whose width and depth are capped to what current devices actually expose.

All of it is list pricing. No reservations, no negotiated discounts, no egress, no storage, no support plan. Treat the output as an order of magnitude for a single reference run and validate against the pricing calculator before anyone commits a budget.

The trap this page exists to avoid

A quantum figure can look enormous next to an HPC hourly rate and produce a confident-sounding "HPC strongly preferred" on pure cost, for a problem where quantum has an exponential advantage and classical simulation is simply intractable at scale. That per-unit comparison is a separate, narrower statement than the verdict, and it is labelled as such wherever it appears.

Back to the Evaluator How the system is built