An Outside Perspective on New Zealand's AI Strategy

After more than a decade in the trenches in New Zealand tech, followed by several years working in fintech and applied AI overseas, I've been thinking deeply about our recently published national AI strategy. I'm not currently running a startup or working in government, so I'm perhaps a little less qualified to opine than some others, but as a Kiwi who wants to eventually come home with my family to a thriving economy, I do have some skin in the game and I'd like to help.

Many local tech leaders have already provided excellent commentary on the document. Rod Drury 's infrastructure and innovation vision and Michelle Dickinson, PhD, MNZM 's education roadmap are exactly right. Rather than restate their points, I'd like to amplify them, build on their thinking and add some observations from my time working in and adjacent to applied AI internationally.

What the Strategy Gets Right

First of all, I do not have a huge problem with the contents of the AI strategy or its hands-off nature. No one expects innovation to come from the government, and it is right to encourage the private sector to lead. The strategy positions the government as an enabler that removes regulatory frictions while adopting the OECD AI Principles, and that framing feels appropriate. I also like its emphasis on encouraging adoption and reducing government-imposed barriers.

Unlike some commentators, I do not mind that AI helped generate parts of the document. Most of the time it is clear when that occurred, and if it allowed faster publication in fast-moving space, more power to them.

Contingency

As Rod Drury notes, New Zealand doesn't have the resources to compete directly with the frontier labs, and he's right. But one thing the strategy seems to presume is that we'll always have access to the most advanced models available. I think that's likely, but I don't think it's guaranteed, and New Zealand would be wise to plan for all contingencies.

If international relations between superpowers sour, export controls or geopolitics could close doors, even to close friends of the United States like New Zealand. The US is nervous about its lead in AI technology, and has placed controls on access to chips and semiconductors. The Government therefore has a critical digital-foreign-policy role to keep as many model-access and hardware channels as open as possible. This is not just a trade discussion but a security/diplomatic discussion and I did not see that pointed out as a component of the strategy. Perhaps this strategy is not the right place for that, and there is some sensitive document floating around MFAT that addresses it, but it is an important consideration.

So Rod's idea of attracting hyperscale investment and renewable-powered data-centre capacity is pragmatic and in my view is dual-purpose. To his point, local capacity shortens latency for domestic businesses, and gives vendors access to compute capacity backed by renewable energy resources, all practical and wonderful things, but it also gives us a contingency option to run our own models if frontier models are cut off or constrained. Hopefully it never comes to that, though in that event, we'd be behind, but not completely left behind.

Efficiency

Today's frontier systems demand enormous investments in GPU hardware, though that may not last forever. With efficiency techniques, sparsity, Mixture-of-Experts routing, LoRA fine-tuning, and specialised tool use, newer model iterations are already cutting energy use and parameter counts while still making big leaps forward in quality. As models gradually approach the energy consumption of the human brain (an engineering challenge that, granted, may take decades to achieve, if achievable at all!), raw compute and cost per task may drop precipitously.

Still, even if individual models become lighter, total demand for cycles will climb sharply as routine mental work is automated across the economy. Either way, aggregate compute remains a national-security asset, so building and retaining domestic capacity is a sensible long game.

Education

Education is another concern, and Dr Michelle Dickinson is right to emphasise it. She calls for a funded, cradle-to-career AI-literacy roadmap, robust ethics and safety guardrails, inclusive design, and respect for Māori data sovereignty. I love it. Children should develop intuition for AI from an early age, and educators need training to guide that journey.

I am sure many educators are already quietly revolutionizing their classrooms with AI and I hope their efforts are being embraced and amplified. But we need to ensure that all students have access to the same advantages, and not just the kids that are lucky enough to have AI-fluent, maverick teachers.

It's taken 75 years, but we've finally abstracted away the complexities of interacting with computers from assembly language all the way to the English language, and the learning curve required to be able to make computers do interesting things has all but disappeared. There's no reason not to be introducing five-years-olds to voice assistants with agentic capabilities, and leveraging AI in the classroom as early as possible to assist in learning, though we must address the digital divide to ensure equitable access.

If we do this, and really invest to make it happen, we could have the most AI-literate workforce on the planet in a decade or two, though we have some catching up to do with countries like Estonia, Singapore and Finland who have already started on their upskilling journeys. In my opinion, such a programme will need a cross-party commitment so it can survive electoral cycles. These compacts are rare in New Zealand politics, and usually reserved for times of emergency, but they are essential for projects that are existential in nature and need decades to deliver full value.

However, New Zealand's ability to adopt AI quickly and painlessly is existential in my opinion. As the industrial revolution replaced physical labour, AI is set to displace mental labour. The artisans that survived and saw wages increase during the industrial revolution were those that better adapted their labour to work alongside and through the new machines. I believe the same is going to be true of professionals who deliver value through their intellect, and New Zealand has a rare opportunity, as a function of its scale, to transition rapidly and seize an early lead. The advantages will compound, and anyone who understands compounding understands that the earlier you start, the bigger the eventual gains.

Concrete Actions

Because AI will boost productivity and international competitiveness across New Zealand society, the government should move faster to encourage adoption, and it can do more than demystifying AI and economic incentives and research grants. Here are specific levers I'd recommend that I think are missing from the strategy or not covered in sufficient detail.

1. Regulatory Reform

One lever is regulatory reform. By lowering barriers that protect incumbents, banks, telcos, insurers, supermarkets, while retaining consumer protections, we create room for local AI-native entrants. Imagine a financial institution with AI-powered loan origination and customer support from day one, instead of something bolted on. Or a supermarket chain with AI-enabled supply chain capabilities and inventory controls. There is a huge amount of activity in private equity and venture capital in the United States right now where new players are entering sleepy industries through a combination of acquisitions and new ventures, designing operations around AI from the get-go and as a result securing huge margin advantages against competition. Though local market entry attempts against big players may be quixotic for those entrepreneurs who are brave enough to try, the mere presence or even threat of competition of that kind would force incumbents to modernise more quickly.

2. Public Sector Leadership

The government can also lead in promoting adoption by auditing public-sector processes, identifying the most expensive routine, repeatable tasks, and tendering them to the private sector for AI transformation. The goal is to free staff from repetitive work so they can focus on policy, creativity, and frontline service. Natural attrition and up-skilling, not cuts, should be the lever. Shopify CEO Tobi Lütke asks whether a new role could be done by AI before approving a hire. The New Zealand public sector could apply the same test. New Zealand has a huge advantage over others here also, the state apparatus on a much smaller scale and therefore the task of transformation could be executed much, much faster.

3. Build Local Expertise

Leveraging the local private sector to assist will help seed and build a local AI transformation industry and develop important skills, which New Zealand will need, but also provide the opportunity for AI transformation experts, once established, to look outwards to increase their revenues and offer those skills internationally, gaining experience as a result. Whereas once the big international IT consultancies had all the advantages in large IT projects, now they have little true in-house expertise (despite any claims they might make to the contrary!), and frankly effective AI transformations need to be run more like iterative R&D projects, with small, nimble, experimental teams who can operate like forward-deployed engineering squads and work alongside public sector professionals. Palantir is the canonical (and most controversial) example of this, but OpenAI and Anthropic are beginning to invest in the same approach. This model is well suited to New Zealand teams that can operate locally, and again, from a sovereignty perspective I would feel better if locals were doing the work.

4. Strengthen IP Protection

Finally, the government should strengthen IP protection at home and abroad. While the strategy mentions cultural IP, much of New Zealand's advantage lies in industrial IP in viticulture, agriculture, food science, health, and environmental protection. Any IP update must sit alongside tikanga-based governance models that respect Māori concepts of kaitiakitanga and rangatiratanga. Innovators should find ways to share their processes safely with AI while guarding against IP leakage. Without such protection, other countries could match or surpass us by using our ideas with higher efficiency.

(A caveat: this alone is not enough. we need rapid AI adoption also, as there's still a risk that inferior ideas and processes running more efficiently could outpace the application of our IP, even if well-protected. In short, there's a very real risk that AI makes any IP advantages we currently have obsolete, unless we AI-enable them.)

Conclusion

In summary, the strategy is on the right track, but to make the recommended actions more concrete. I would:

1. Develop a plan to ensure foreign policy secures continued access to advanced models and critical hardware and makes them a part of trade and security negotiations.
2. Set targets around local compute capability and infrastructure.
3. Establish a cross-party, fully funded AI-literacy programme from primary school to mid-career re-skilling.
4. Redouble efforts to reduce regulations that shield incumbents from new entrants while maintaining consumer protections, allowing a wave of new AI-native innovation.
5. The Minister for Digitising Government should have a program to identify AI-transformation opportunities in the public sector and establish a budget, backed by ROI projections, to engage the local private sector to deliver them.
6. Initiate an inquiry into our IP laws and embed tikanga-based governance to protect our cultural and technological assets. It looks like the Copyright and Patent Acts have been updated fairly regularly for the past couple of decades but the most recent exhaustive review of our IP laws that I could find was conducted in 2010 by the University of Auckland, and the Law Commission last published a report on Intellectual Property in the 1990s. Given the wonkish and multi-faceted nature of the legislation It may be time for the Law Commission to take a fresh look.

Though currently overseas, I'm exploring ways to contribute to New Zealand's AI transformation. This isn't about telling anyone what to do, it's about joining the conversation and finding ways to help.

If you're working on AI transformation projects and want to collaborate, or if you think I'm wrong about any of this, I'd love to hear from you.

DM me if you want to connect.