What it costs

North Canterbury · © John Stroh

Version 0.1 · revised 6 September 2026 · this version as a PDF

Drafted with AI assistance, then checked and revised by the author. The judgements and the errors are the author’s. How this site is written sets out what is declared on every piece, who checks it, and where the per-piece record lives.

A framework for working it out, and worked numbers that are examples rather than findings.

Series part F · Version 0.1 · August 2026


Before you read this#

What does this cost? We do not know, and neither does anyone who has not built one. Anybody quoting you a figure for something that does not yet exist is quoting an aspiration.

So what use is this part? A structure for working out your own answer, with every line named, and worked numbers to show the shape of the arithmetic.

Which number actually decides it? One. Everything else ladders to it.

And where does the arithmetic usually go wrong? In four places, all of them in year three rather than year one.

Every figure in this part is marked ILLUSTRATIVE. They are placeholders showing the shape of a calculation, not estimates of anything. Replace all of them. If you quote one of them back to anybody, we have failed to mark them clearly enough.


The number that decides it#

Cost per site to install and sustain for five years.

A minister and a board of trustees need the same figure. A funder asks for it in the first meeting. A member compares it to what they pay now.

Not cost to build. Not total programme cost. Not cost per member per month, which hides the sustaining problem in an average. Per site, over five years, because year three is where arrangements like this fail and a one-year figure conceals exactly that.

If you can produce this number and defend it, you have a proposition. If you cannot, everything else is elaboration.

Four cost pools#

Four cost pools and the one number that decides it FIG-16FOUR POOLS · KEEP THEM SEPARATE · ILLUSTRATIVE FIGURES ONLY 1 · Site capitalmachine, storage, installonce, owned by the member$9,500 2 · Site sustainingreplacement, updates,a visit, support$5,200 / year 3 · Cluster sharedpooled compute, exit drill,heterogeneity premium$70,000 / year 4 · The service people at the desk —model this one first $900 – $5,400 per member THE NUMBER EVERYTHING LADDERS TO Cost per site to install and sustain for five years — $9,500 + (5 × $5,200) = $35,500 Year three is where arrangements like this fail, which is why a one-year figure conceals the thing that matters. If your case only works at 5% escalation, you are betting on a forecast.
FIG-16 Keep the pools separate — mixing them hides the sustaining cost in an average. All figures illustrative, none an estimate.

Keep them separate. Mixing them is how the sustaining cost disappears into an average and reappears as a crisis.

1 · Member site — capital, once

What sits in the member’s building, owned by them outright.

Line ILLUSTRATIVE Basis
Compute node $6,000 A workstation-class machine with a current consumer GPU
Storage and networking $1,500 Local storage, boundary device
Installation and commissioning $1,200 Local subcontractor, one day
Initial training, two people $800 Half a day each
Site capital $9,500

What moves this. Hardware falls in price and rises in capability; this line should shrink in real terms every year, which is the whole argument for measuring shares in capability rather than dollars. Organisations with existing server rooms need less. Organisations in remote districts pay more for the day.

2 · Member site — sustaining, annually

The line that kills projects, because nobody budgets it.

Line ILLUSTRATIVE Basis
Hardware replacement provision $1,900 Straight-line over five years
Model and software updating $600 Central cost apportioned
Remote monitoring $300
One on-site visit $600 Local rate, half a day
Support and escalation $1,800 See pool 4 — this is the uncertain one
Site sustaining, per year $5,200

Five-year cost per site: $9,500 + (5 × $5,200) = $35,500. ILLUSTRATIVE.

That is the number. Every argument about this proposition is really an argument about that figure and what it buys.

3 · Cluster — shared

Whatever a group of sites holds in common: pooled compute, the shared knowledge base, the exit-drill capacity, the boundary attestation.

Scale it per cluster and divide by members, and do not hide the exit guarantee inside it. Holding complete state at every cluster costs more than centralising. State the premium as its own line and defend it, because a guarantee absorbed into overhead is a guarantee an efficiency review removes.

Line ILLUSTRATIVE, per cluster per year Basis
Shared compute and hosting $40,000 Sovereign-qualifying provider
Complete-state premium $12,000 The cost of being able to leave — never fold this into the line above
Annual exit drill $8,000 Steward and operator days, plus remediation provision
Heterogeneity premium $10,000 Second model family and hardware generation
Per cluster, per year $70,000 Across 25 members ≈ $2,800 each

4 · Central — the service, and the one to model first

People at the desk. This is not the largest pool but it is the least predictable, and it is where the whole proposition is decided.

Model this before anything else. The economics rest on a ratio: how much routine work the local system handles unaided versus how much reaches a person.

Escalation rate Contacts to a person, per member per month ILLUSTRATIVE cost per member per year
5% 3 $900
15% 9 $2,700
30% 18 $5,400

If your case only works at five per cent, you are betting the organisation on a capability forecast. Build it to survive three times your target rate. If it does, the model is robust; if it does not, you have found out for the price of a spreadsheet.

Two things reduce this pool that are not efficiency measures: members contributing accredited hours in exchange for entitlement, and a local system that improves as it reads more of the member’s own material.

Segments — and they behave differently#

Do not average across these. The averages are meaningless and the differences are the plan.

Segment Ability to pay Speed to decide Assurance cost Note
Public bodies High Slow High Want evidence before piloting — real cost before revenue. They are the anchor because their requirements force the standard to be real.
Professional practices Medium Fast Low Clear pain, understand recurring costs, decide in one meeting.
Schools and education Low Slow Medium Procurement cycles, boards, term timing.
Community organisations Low Fast Low Highest contribution potential, lowest cash.
Trade and small business Medium Fast Low Largest by number, hardest to reach individually.

Geography. Cluster density decides utilisation and field-service cost simultaneously, and they pull in opposite directions. A dense urban cluster has good utilisation and cheap service calls. A dispersed rural one has neither, and is where the need is greatest. Model at least one of each; do not build a plan that only works in a city.

Risk lines, priced#

Not a register. Lines in the model, with money against them.

Risk Effect on the model ILLUSTRATIVE provision
Escalation rate above target Central pool inflates Model at 3× target
Field service worse than modelled Sustaining pool inflates +25% on visits, years 1–2
A member cannot pay Revenue loss, no cost relief 5% of member revenue
Base model becomes unavailable Substitution project One-off, 6 weeks of engineering
Slower member growth Fixed costs across fewer members Model at 60% of plan
Hardware price does not fall Capability-pegged shares squeeze Sensitivity, not provision

That last one deserves attention. If shares are measured in capability, falling hardware cost is what funds rising member entitlement. If prices stop falling, the mechanism stops working and the constitution has committed you to something the market no longer supplies. Model it before adopting the peg.

How to use this#

  1. Replace every figure. All of them are illustrative and none is an estimate.
  2. Build the central pool first, at three times your assumed escalation rate.
  3. Produce the five-year per-site number and put it on the front page.
  4. Model two clusters — one dense, one dispersed.
  5. Keep the exit premium and the heterogeneity premium as their own lines, permanently.
  6. Publish what you find, including the parts that do not work. A costing that only closes under favourable assumptions is not a costing.

What we cannot tell you#

We have not built one, so we have no measured figures. What is above is a structure and an arithmetic shape.

Two things would change it materially and neither is knowable in advance: what members will actually pay for the exit guarantee, and what the real escalation rate is once a local system has been reading an organisation’s own material for a year. Both are measurable within months of starting. Neither can be resolved by more analysis.

If you build this and measure either, publishing the number would be worth more to everyone else than anything in this part.


Published under CC BY 4.0. Every figure marked ILLUSTRATIVE is a placeholder. None is an estimate, a quote, or a finding.

Disclaimer

Status of these claims

What this publication does not claim, and what is outstanding against it in the register.

Asserted, not yet measured: The cost of secession capability, this publication.

A question this rests on is parked: What secession premium will members bear?

We publish no pricing and no affordability claim, and we do not assert that members value the secession guarantee.

Alongside: sources and provenance · slides