Finance Leaders · Spreadsheet and a short guide

On-Prem vs Cloud AI TCO Worksheet

Cost per productive GPU-hour for owned hardware against a hyperscaler, the neocloud median and the lowest neocloud, with the utilization each one needs to beat. Example inputs are dated and sourced; replace them with your own quotes.

  • B200 base case and a B300 tab, every input editable
  • The crossover utilization against three cloud benchmarks
  • Capital as your actual loan or lease payment, not depreciation
Free download

Get the worksheet

Tell us where to send it. The link arrives by email in a minute or two.

We use your email to send this and, occasionally, related material. Unsubscribe any time. Privacy Policy

what it leads to

When the document doesn't settle it: Placement Assessment

This is the same material we work from on paid engagements. When your answers raise more questions than they settle, that is where Placement Assessment starts.

More for Finance Leaders

Drawing: cost over utilization A chart of cost per productive GPU-hour against utilization. The on-prem cost line falls as utilization rises. The cloud line stays flat. The point where they cross is marked: below it the cloud costs less, above it owning costs less. Summit Logic Cost over utilization Dwg SL-0016 Sheet 1 of 1 Rev A Illustrative Cost per productive GPU-hour Utilization 1 On-prem Fixed cost, spread over more hours as use rises 2 Cloud Paid by the hour, flat at any use 3 Crossover Cloud costs less Owning costs less
also in the library

Other working documents

Rather talk it through?

Forty-five minutes. Bring the question this document is about and we will tell you which answers decide it.

Book a discovery call