Skip to content

Owning your AI

Why run your own AI?

Rented AI bills you for every word it reads and writes. On hardware you own, most of the cost is the machine, so each extra question costs very little. Here's when that matters, and when it doesn't.

See your AI bill

Renting

Rented AI charges for every word.

Providers bill by the token, about three quarters of a word, for what you send and again for what comes back. Every request leaves your business. The provider sets the price and the terms, and decides when a model changes.

What renting gives the vendor

Your files

Client files and site records can stay on your own servers.

Clients and head contractors often set rules about where their files may go. When the model runs on your own hardware, or in an Australian private cloud you control, matter files, tender packs and site procedures don't pass to another provider. There's less to check and less to explain. At a mine site, the same setup keeps answering when the connection drops.

AI for your industry

As use grows

A small pilot bill can become a large one.

Rented AI that costs little in a pilot costs far more once the whole business uses it. AI agents push it up again, because they work through tasks on their own and send much more each time. Here's what three teams would pay each month, worked out by our AI cost calculator.

Planning estimates in Australian dollars, excluding GST, at Claude Opus 5's published prices checked 18 September 2026. Not quotes. The calculator also shows what owning the same work would cost.

Why now

Why it's practical now.

  • Models you can download

    Open-weight models, which you download and run yourself, now handle many everyday business jobs. You test that on your own examples before buying anything.

  • Enough memory in one machine

    A desk-side AI computer can now run one strong model that a team shares.

  • Software that installs like any other

    The program that runs a model is a standard install. The work goes into choosing the model and fitting it to your jobs.

What it takes

Memory decides which models fit, and how fast they answer.

A model is a large file of learned settings, called weights, and a program that runs it. Answers come back as fast as the machine can read those weights from memory.

  • Memory fitWhether the model actually fits in a machine's memory, with room for the conversations it is handling.
  • Weight precisionHow finely the model's numbers are stored. Lower precision makes the model smaller and faster, with some loss of accuracy.
  • ThroughputHow fast a machine reads and writes tokens, in tokens per second. It decides how many machines your workload needs.
What each machine suits

Your tools

The tools and AI agents you already use can point at it.

Most AI tools send requests in the same common format, so many can move to a model you own by changing an address and a model name. We check each of your tools during the pilot, because some don't allow the change.

  • AI agents

    An AI agent is a model plus the software that lets it read files and use the tools you approve, so the model underneath can be yours.

  • Private Model Service

    A single entry point for every team and AI agent. It sends each request to the right model, rented or owned, records who asked what, and keeps spending within your limits.

What each step includes

Your documents

How the model learns your business.

  • Looked up: documents that changeProcedures, client files and price lists sit in a library the model searches. It answers from the pages it finds and shows which ones it used, so an update counts as soon as it's saved.
  • Trained in: the way you workYour terms, your report formats and how your team writes. Training adds a small layer to the model, and the pilot shows whether it's worth doing.

Keep renting

What we'd keep renting.

An engineer's recent video on running AI locally lists where it fits, including

“anything you'd otherwise be paying per token for at scale.”

Maddy Zhang, senior software engineer, “Everything You Know About Local AI is OUTDATED”, YouTube, October 2026 · source

The best rented models still do better on the hardest problems. A model you own that answers slowly costs your team time.

Owning fits private files, sites that go offline and simple jobs that repeat all day.

Five questions show where your AI should run. Sometimes the answer is to keep renting.

Find your setup

Your numbers

Work out your own numbers.

Start from one of the teams above, or your own. The calculator shows what renting costs, what owning would cost and the hardware you'd need.

Open the AI cost calculator

AI Ownership Workshop

Find out what to rent and what to own.

About 90 minutes at no cost, and you leave with a one-page ranking, a rough cost and a first pilot.

Book a call