On-prem vs cloud machine intelligence: the real 3-year cost math for a 10-person firm

What a private machine actually costs against what a 10-person team pays for cloud machine intelligence over three years, using published numbers instead of a sales pitch.

Every business we talk to has done this maths badly at least once, usually on the back of a napkin, usually in the vendor's favour. So here is the version done properly, with sources.

What the cloud actually costs a small team

A busy 10-person team spending time in a cloud assistant every day typically runs somewhere between $300 and $500 a month once you add up per-seat subscriptions and API usage on top. That is $3,600 to $6,000 a year, and it rises every year the provider decides to raise prices, which they do.

Azure's own published enterprise rates make the ceiling clear: a single dedicated throughput unit for a frontier model runs roughly $3,000 to $4,500 AUD a month on its own, before anything else is added. Most small businesses never need that much capacity, but it shows where the cloud pricing model tops out.

What a private machine costs

Independent estimates for a comparable on-premises setup, published outside Norai, land in a consistent band: a software-only license around $4,500 one-time, a small-practice appliance around $7,500 one-time plus a modest monthly fee, and an all-in installed deployment for a small firm in the $10,000 to $16,000 range for the first year, hardware included.

A separate, unrelated build guide priced a local server for a similar-sized team at roughly $217 a month all-in once the hardware is amortised over two years and electricity is added, against $528 a month for the equivalent cloud usage on that comparison's own numbers. Different hardware, different provider, same shape of answer: the local setup breaks even within about five months and keeps paying for itself every month after that.

The maths, side by side

  • Cloud, 3 years: $3,600–$6,000 a year × 3 = $10,800–$18,000, and that is before any price rise.
  • Private machine, 3 years: one-time setup in the $10,000–$16,000 range, then close to nothing beyond power.

The crossover point in most of these published estimates sits between four and six months. After that, every month on a private machine is close to free, and every month on the cloud is a fresh bill.

The number that isn't a number

None of the figures above are ours. They're what the market is already charging for something adjacent to what a private machine does. If you want the real number for your business, tell us your team size and what you actually need it to do, and we'll do this maths for real instead of in the abstract.

Sources: Azure AI Services pricing (published enterprise rates), and two independent on-premises cost guides for comparable systems, published outside Norai. Figures current as of September 2026; cloud provider pricing changes without notice, which is itself part of the argument.

Fifteen minutes. No pitch.

Tell us what's eating your team's week. We'll tell you honestly whether a machine should be doing it, and what that would take. A person reads this and replies once.

Goes straight to the engineers. [email protected]