The math behind
on-prem vs rent.

Pick a tier, the workload class, GPU-hours/month, your current $/GPU-hour, and the assessment window — read payback in months, cloud total vs Helmstone purchase + amortized support, and the 3-year line. Hands off to /quote with the tier preselected and the summary folded into your brief.

Run the TCO See the published starting prices →

HS-TC-02 / Run

Pick the four knobs. Read the line.

The math runs server-side against the published starting prices on /tiers ; what you see is a per-interaction TCO row, not a smoothed median. Negative savings are surfaced honestly — at very low cloud spend the math has Helmstone costing more; the page calls it.

Configuration

Pick a tier and the cloud-spend assumption.

Math reruns on every change. The Helmstone purchase line is the published starting price; the cloud side is your current GPU-hours × current $/GPU-h on the equivalent cloud SKU.

Tier

Workload class

Fine-tune / training burstsCloud discount applies on spot / reserved — baseline pinned at on-demand.Sustained inference / servingSustained 24/7 serving rarely gets the spot discount — +5% baseline.

GPU-hours per month

One full-TDP GPU × one wall-clock hour = 1 GPU-hour. Multi-GPU jobs scale equivalent hours.

Current $/GPU-hour

Default is the per-tier pinned cloud baseline — override if your quote sheet is different.

Assessment window

12 moYear-one only24 moTwo-year TCO36 mo3-year TCO

Bring this to /quote →

Pre-fills the tier and the TCO summary on /quote so a sales engineer replies with a build pinned to your spend.

Result

Computing…

For Helmstone W1 over 36 months. Numbers refresh as you change tier, workload class, GPU-hours, $/GPU-hour, or the window.

Cloud total over window: —
Helmstone purchase (one-time): —
Helmstone support (window): —
Payback period: —
3-year savings: —
Citation: —

HS-TC-03 / Sources

Where each row comes from.

The Helmstone side is the published starting price on /tiers. The cloud side is a pinned per-GPU-hour snapshot per tier — no live AWS / GCP fetch, no fabricated “median”. Override the cloud side with your own quote and the math re-runs the same way.

Helmstone starting prices →

The four published rows on /tiers — From $6,400 / $24,900 / $89,000 / $189,000. Founder-pending rows flagged on the catalog.

Pinned cloud $/GPU-h →

W1 $2.53/h baseline (A100/H100 on-demand); W2 dual-Pro 6000 cloud baseline; W4 multi-B200 cluster; RS multi-rack. Workload modifier adds 5% for sustained inference.

Support rate (amortized) →

12% of purchase / yr for W1 / W2; 10% for W4 / RS. Scales linearly with the assessment window in months.

Hand-off to /quote →

The Bring-this-to-/quote button pins the tier, the workload, GPU-hours, $/GPU-h, and the window into the brief so a sales engineer replies with a build pinned to your spend.

48-hour burn-in (the validation under the math) →

The published W1 burn-in profile grounds the workloads the calculator assumes — the same methodology applies to every tier on the catalog.

Compare vs DIY (the integrator-vs-build decision) →

The configurator page is the bridge between browsing and a quote — the TCO page is the financial justification that side-steps the build-vs-buy conversation.

HS-TC-04 / Hand-off

Bring this to /quote.

The TCO summary travels with the quote brief — a sales engineer replies within one business day with the SKU pinned to your spend, a configuration, a benchmark target, and a quote.

Open /quote →Or pin a build with the configurator →

What we hand off

•Selected tier, GPU spec, SKU
•Workload class (fine-tune vs sustained inference)
•GPU-hours per month · current $/GPU-hour
•Chosen assessment window (12 / 24 / 36 mo)
•Cloud total · Helmstone total · 3-year savings