Per-Seat Subscription or Private LLM Build?
A per-seat AI product looks cheap in month one and keeps growing with headcount. A private deployment costs more up front and then costs roughly the same every month. This calculator puts both on the same timeline so you can see the 1, 3 and 5 year totals and where they cross.
Enter the per-seat price you have been quoted, not a list price. For the private side, our published range for a single use case deployment is $25,000 to $60,000, so the default sits inside it.
Subscription vs Private Build
Positive saving favours the private build over your chosen horizon.
Your quoted AUD price including any prerequisite licence.
Our single use case range is $25,000 to $60,000.
Compute, vector database, monitoring and a support retainer.
Positive favours the private build. Negative favours the subscription.
Cost comparison only. Excludes training, change management and delivery risk, which favour the subscription, and excludes data residency, which may rule the subscription out. Private cost is treated as flat with seat count. Seat growth compounds once a year on the subscription side. Inputs stay in your browser and are not sent to us.
Reading the Numbers
The two cost curves have different shapes. One is a line that steepens with every new seat. The other is a step followed by a flat line. Which is cheaper depends only on where you are on the timeline.
Subscription cost scales with seats
Every new starter adds a licence. The calculator compounds your seat growth percentage once a year, so a growing organisation sees the year five subscription bill sit well above year one on the same per-seat price, before any price rise from the vendor.
Private cost is mostly fixed
A private deployment has a one-off build cost and a monthly hosting and support figure that does not move much with seat count. Extra users add load, not licences. That is why the private line stays flat while the subscription line climbs.
The break-even month is the decision
If the two paths cross inside your planning horizon, the private build is the cheaper option over that horizon. If they cross after it, the subscription is cheaper for now. The break-even figure uses your current seat count, so growth only makes it earlier.
How the Comparison Is Built
Plain arithmetic over your inputs. No hidden discount rates, no assumed productivity gains, no vendor list prices baked in.
Subscription per year
Seats multiplied by the per-seat monthly price multiplied by twelve. Each following year the seat count is increased by your growth percentage.
Subscription totals
Years are summed as a compounding series: year one, plus year one grown once, plus year one grown twice, and so on out to five years and to your chosen horizon.
Private totals
The one-off build cost plus monthly hosting and support multiplied by the number of months in the period. The build cost is counted once, in year one.
Break-even
Build cost divided by the monthly gap between subscription spend and private hosting, at today's seat count. If the subscription is not the dearer of the two per month there is no break-even, shown as 240 months.
What the Calculator Leaves Out on Purpose
These items matter and are hard to put a number on without knowing your organisation. Add them to the conversation, not to the formula.
Prerequisite licences and bundles
Some enterprise AI products require a base productivity licence before the AI add-on can be purchased, and some vendors bundle the AI seat into a higher tier. Put the full incremental monthly cost per user into the per-seat field, not the headline add-on price.
- Include any licence tier upgrade the AI seat forces
- Include minimum seat commitments if the contract has them
- Use the AUD price on the quote, not a USD list price
- Check whether the price is fixed for the term or reviewed annually
What sits inside the private monthly figure
For a cloud-hosted private deployment the monthly figure covers compute or model API spend, a vector database, hosting, monitoring and a support retainer. For on-premises deployments the hardware is a capital item and the monthly figure is mostly power, support and staff time.
- Use the LLM cost calculator to size the token and infrastructure part
- Add a support retainer if you will not maintain it in-house
- On-premises hardware belongs in the one-off figure, not the monthly one
- Re-embedding a large corpus is an occasional cost, budget for it separately
Capability is not identical
A per-seat product gives every user a general assistant across the vendor's apps. A private deployment is usually built around your documents and one or two workflows. The cheaper option on this page is only the better option if it does the job you need done.
- List the tasks staff will actually use it for before comparing price
- A private build answers from your documents with citations and permissions
- A subscription product is faster to switch on and needs no build project
- Many organisations run both: a subscription for general use, a private build for sensitive data
Data residency changes the answer
If your data must be processed in Australia, some subscription products cannot meet that at all and the cost comparison is moot. Check the residency question first with the data residency quiz, then come back to cost.
- Confirm the region of inference, not only the region of storage
- Confirm whether the vendor excludes your data from model training in writing
- A private deployment in an Australian region or on-premises settles both questions
- Regulated sectors may need to evidence this to an auditor
Next Steps
Build vs Buy Calculator
The same comparison with maintenance as a share of build cost and the seat count where the paths meet.
Compare build against buy →LLM Cost Calculator
Size the monthly token and infrastructure figure that goes into the private side of this comparison.
Estimate monthly running cost →Private LLM Cost in Australia
The written guide to what small, medium and on-premises deployments cost locally.
Read the cost guide →Frequently Asked Questions
The incremental monthly cost per user on your actual quote, in Australian dollars, including any licence tier the AI seat forces you to buy. Vendor list prices are often in US dollars and exclude prerequisites, so a quote is a better input than a website. If you have not been quoted yet, use the list price as a placeholder and note that the real figure is usually higher once prerequisites are counted.
Because a private deployment is licensed by capacity, not by user. More staff means more queries, which raises compute and token spend, but that increase is small compared to a per-seat licence and it is already inside the monthly hosting figure you enter. If you expect usage to grow a lot, raise the monthly figure rather than trying to model it per seat.
Build cost divided by the monthly gap between what you would pay in subscriptions at today's seat count and what you would pay in private hosting and support. If the subscription costs less per month than private hosting, there is no break-even and the calculator shows 240 months, which is the top of the range. Seat growth is deliberately left out of this figure, so the real break-even is the same or earlier.
Our published range for a single use case, cloud-hosted deployment is $25,000 to $60,000. That covers retrieval architecture, ingestion of one document corpus, permission mapping, an evaluation set and integration with one existing system. Medium deployments with two or three use cases run higher. Enter the figure from a quote if you have one.
No. Both paths are assumed to deliver the capability, and the page compares only what they cost. Productivity claims depend on your tasks and staff and are easy to overstate. If you want to put a value on time saved, the document search time cost calculator does that from your own numbers.
Your inputs are saved in your own browser so they are still there if you come back, and nothing is transmitted to us. Use the reset button to clear them.
Sources and further reading
Want the Private Side Quoted Instead of Estimated?
Tell us your seat count, the documents you want covered and where the data must live. We will come back with a fixed build price and a monthly figure you can drop straight into this calculator.