Innovation Policy

Context

Government policy plays a critical role in shaping whether AI innovation benefits New Zealand broadly or concentrates wealth and opportunity.

Sovereign inference infrastructure

New Zealand should operate its own AI inference infrastructure running open-source models. If our public services, healthcare, education, and economic systems increasingly depend on AI, that AI should not be controlled by foreign governments or corporations who can change terms, restrict access, or surveil usage.

The proposal:

• Government-funded inference clusters running open-weight models (Llama, Mistral, and successors)
• Hosted domestically in NZ data centres or in trusted jurisdictions
• Available to government agencies, public services, and NZ businesses
• Builds domestic AI expertise, operations capability, and employment
• Reduces per-query costs over time vs perpetual commercial API dependence
• Complements (not replaces) commercial AI — use sovereign infrastructure for sensitive and critical workloads

This aligns with data sovereignty principles and Te Mana Raraunga. It is also a practical resilience concern — NZ cannot afford critical systems that stop working if a US or Chinese company changes its pricing, politics, or terms of service.

Open source policy simulations

Innovation policy should be evidence-based, not aspirational. We propose building and running open-source simulation models to test policy ideas before advocating for them — agent-based models, economic simulations, labour market projections calibrated with real NZ data.

Results would be published alongside the policies they test, both in this document and as interactive explorations. Source code published in the nz-to-utopia org for anyone to audit, fork, and improve. See Methodology for details.

Key areas to explore

• R&D incentives — Tax credits, grants, and funding for NZ AI research
• Public sector AI — Government as an early adopter and testbed
• Sovereign inference — Government-operated AI infrastructure on open-source models
• International partnerships — Research and trade collaborations on AI
• Intellectual property — IP frameworks for AI-generated work in NZ
• Competition policy — Preventing AI-driven market concentration
• Open data — Government data as a platform for innovation
• Simulation infrastructure — Open-source modelling tools for policy testing

Questions for contributors

1. How does NZ's current innovation policy framework need to change for AI?
2. What is the realistic cost and timeline for sovereign inference infrastructure?
3. What open-source models are mature enough for government use today?
4. What international innovation policies should NZ consider adopting?
5. How do we balance supporting innovation with managing risks?