The Hidden Bill: What New Zealand's Government AI Savings Don't Show
- David Harvey
- Jul 30
- 3 min read
A second look at the cost of borrowed intelligence — and a response to Professor Alexandra Andhov.
The case for cutting roughly 9,000 New Zealand public-service roles rests on a number: billions in savings, with government AI doing more of the administrative work for less. That number is real. It is also only one side of a ledger the public has not been shown the other half of.
Professor Alexandra Andhov, chair of law and technology at the University of Auckland, put it plainly to RNZ: the published material does not really show the cost side of AI. She is right, and the gap matters more than it looks.
The costs that aren't on the invoice
Enterprise-scale AI is not a one-off purchase — Andhov points to ongoing licence fees, model upgrades, and audit costs, plus a quieter dependency: the vendor decides when and how its tools change, and you adjust. You do not own the thing your workflow now runs on. You rent it, on terms you do not set.
And today's price is not tomorrow's. Andhov's sharpest point: today's AI pricing is a land-grab discount — providers racing to lock in market share before they set the price that reflects what these systems actually cost to run.
The subsidy won't last
This is the pattern anyone who has studied platform economics recognises. Win the market at a loss, then raise the price once switching becomes painful. When that turn comes — when the major providers have captured enough of the public sector and need to return their infrastructure spend — the pricing will move. By then the public service will have been rebuilt around tools whose cost structure it never modelled and whose pricing it cannot control.
That is what borrowed infrastructure looks like from the outside: convenient, until it is not, and expensive to leave.
Where the money goes
There is a second cost, and it compounds. As Andhov notes, if New Zealand's government AI ultimately runs on a United States provider, the payments flow to a company in California that pays no tax here. The savings booked in Wellington leave the country. The jobs they replaced were here.
Count the ledger honestly: roughly 9,000 roles gone. The savings that justified them are flowing offshore.
No New Zealand tax on the AI that enabled the displacement. No local capability built. And no governed record of how any of it was decided.
That is not a technology strategy. It is a dependency, written into the state's operating model, called efficiency.
The question procurement hasn't answered
Andhov asks the one the process has sidestepped: who are these providers? Most under consideration are neither New Zealand companies nor governed by New Zealand law. They answer to United States jurisdiction, United States data rules, and United States commercial terms.
You can use a foreign model sensibly — the country doesn't need to build its own at global scale.
But there is a difference between using a tool and routing the accountability for consequential public decisions through a tool the Crown does not own, cannot fully audit, and cannot price over time.
What's worth owning
Minister Goldsmith told Parliament that New Zealand should focus on what it does well and sell it to the world. That is the right instinct, pointed at the wrong layer.
New Zealand does not need to own every model it uses. It does need to own the decision-governance layer above those models — the part that holds the evidence, records who decided what, and answers to New Zealand institutions when the questions come. That layer needs no local chip fabrication and no frontier-scale model. It needs accountability built in, under New Zealand law.
That is something this country can build, own, preserve its value, and export. We have built a working version of it — the governance layer itself, and what it looks like in practice, is Part 2: “NZ Government AI and Decision Governance: When Copilot Isn't Enough,” at designbyzen.com/post/nz-government-ai-and-decision-governance.
The savings are real. So is the bill that has not been shown. The choice is not whether to use AI. It is whether the decisions it shapes stay accountable here — or quietly become someone else's to price.
Omega* Sensing is part of the Omega* Unified Ecosystem, developed by Design By Zen, an NZ-based AI Lab. Omega* is the algorithmic engine beneath the ecosystem. SHE ZenAI is the brand of a governed clinical intelligence framework designed for high-trust domains where evidence, not confidence, is the currency of care. Version 1.0, April 2026.
Comments