The preparedness–adoption split
IMF DataMapper scores New Zealand’s overall AI Preparedness Index at 0.75 on a 2023 base. Component scores show Digital Infrastructure at 0.19, Innovation and Economic Integration at 0.16, and Human Capital and Labor Market Policies at 0.18. Staff said performance is particularly strong on regulation and human capital, with scope to improve digital infrastructure and innovation outcomes.
The World Bank’s Mandon research lists New Zealand among ten high-income global overperformers whose AI readiness exceeds what economic complexity alone would predict. Peers in that set include Australia, Singapore, Denmark, Finland, the Netherlands, Norway, Japan, Korea and Hong Kong SAR.
Adoption surveys tell a different story. Datacom’s 2026 Index, covering leaders in organisations with more than 100 employees, puts some AI use at 91%, up from 66% in 2024. Only 15% are scaling organisation-wide. Some 81% remain exploratory or in implementation. The share saying AI has transformed core operations fell to 4% from 8% a year earlier.
Datacom AI director Lou Compagnone has described scattered use-cases as “AI confetti.” Reporting on the Index puts dedicated AI leadership at about 22% of businesses, including 13% with a chief AI officer. Investment is still rising: 79% increased AI spend in the past year and 73% expect to invest more in the next 12 months. Only about one-third report returns exceeding costs.
Prime Minister Christopher Luxon struck the same note at the June 2026 China Business Summit. He said New Zealand is missing a large opportunity on AI.
I think there's a lot of opportunity - I think we're missing a big trick on AI in New Zealand.
He added that chamber hand-raises for tool use leave him “constantly underwhelmed by how little we are adopting them.” He also cited an R&D spend rank near 19th per capita against a commercialisation rank near 46th. Those ranks are speech attributions, not a matched OECD table verified here, but they track the commercialisation friction the IMF also named.
PwC’s NZ AI Jobs Barometer shows AI-related ads rising from about 3,900 in 2024 to about 9,600 in 2025, with AI skills in 2.7% of postings. In several sectors more than 99% of roles are AI user rather than AI developer roles. The IMF noted somewhat less local demand for AI-related skills than in leading advanced economies. New Zealand is wiring tools into workflows faster than it is building a thick developer and product layer.
Why AI collides with the productivity trap
Stats NZ’s productivity statistics for the year ended March 2025 show measured-sector labour productivity up only 0.8%. Multifactor productivity fell 0.9%. Capital productivity fell 3.0%. Since 1996, measured-sector labour productivity is up 39.1%, an index of 1391 on a 1996=1000 base.
Primary industries are the long-run standout. Their labour productivity index sits at 1722—up 72% since 1996. Agriculture, forestry and fishing labour productivity rose 7.3% in the latest year, with multifactor productivity up 5.7%. Goods-producing labour productivity is only 16.6% above 1996 and fell 0.9% in the latest year. Services labour productivity is up 47.7% since 1996.
Mandala and Microsoft (2025) stress that applications plus datacentre infrastructure account for more than 90% of New Zealand’s realistic AI economic potential. Frontier models and chips are not the comparative advantage. Beehive and MBIE mainstreamed the $76 billion figure in the July 2025 AI Strategy. Those scenarios remain contingent on complementary inputs the diffusion data say are incomplete.
Cropsy, Halter and Hectre as the transmission channel
Cropsy Technologies, a University of Auckland engineering spin-out founded in 2019, mounts computer-vision scanners on existing vineyard vehicles. Company and university sources report about 22 million vines scanned and 40 scanners across New Zealand, California and Cognac. Each system analyses up to 8,000 vines per hour and about 50,000 images per hour. Insights typically arrive the next morning; some pipelines deliver in as little as 30 minutes.
Applications include pruning quality, mildew and disease maps, yield estimates and vine counts. A service-provider model—Flowerday Contracting covering nearly 1,000 hectares for more than 90 growers—opens SME access without each grower buying hardware. Co-founder Leila Deljkovic set out the operating philosophy in a May 2025 University of Auckland profile.
My most important lesson is that the only sustainable growth hack is a slow one.
She also framed the national opportunity and the capital constraint in plain terms.
I think New Zealand has a genuine opportunity to be a global leader in agritech. We have talented people, we know a lot about agriculture – it’s central to our economy.
Agritech is globally underfunded, with New Zealand being even more underfunded... We cannot cut funding in technology – we need more investment in technology to create value.
Publicly cited early funding remains modest, on the order of $1.5–1.7 million with Seraph Group and local angels. That scale underscores the shallow risk-capital problem the IMF listed among adoption impediments.
Halter’s March 2026 Series E raised US$220 million led by Founders Fund at a $2 billion valuation, according to the company’s BusinessWire release. More than one million collars and more than 2,000 farms across New Zealand, Australia and the United States make it the scale counterpoint. Virtual fencing and AI herd management target pasture and milksolids productivity.
Hectre’s February 2026 NZ$12 million Series A, led by Punakaiki Fund with Active Investor Plus capital, targets industrial-scale fruit sizing and quality data. Representation is cited around 17 billion apples and 37 billion cherries annually. The Primary Sector Growth Fund is co-funding hyperspectral internal defect detection work with staged accuracy and packhouse prototype milestones.
Wine provides the export backdrop. MPI’s Situation and Outlook for Primary Industries (December 2025) put wine export revenue at $2.1 billion in the year to June 2025. Forecasts sit near $2.29 billion in 2026 and $2.36 billion in 2027 after a 519,000-tonne 2025 crush on about 42,520 producing hectares. NZ Winegrowers reported exports above $2.10 billion. Global oversupply and bulk share pressure prices. Vine-level maps that defend yield and quality are margin tools for exporters, not vanity pilots.
Primary labour productivity’s long-run outperformance is the channel through which vine-, cow- and fruit-level AI can compound into national multifactor productivity—if capital, power and skills allow diffusion beyond frontier firms.
Capital, power and the datacentre bet
The IMF listed structural impediments to AI adoption, especially among SMEs: high costs and administrative burdens of starting and scaling a business; low R&D intensity relative to peers; high and volatile electricity prices; and shallow capital markets constraining risk finance. It pointed to continued progress on the Commerce Act review, Capital Markets Reforms and R&D tax incentive settings as routes to convert preparedness into gains.
Capital Markets Reforms Phase One made IPO prospective financial information voluntary. Phase Two remains open on director liability, Catalist, crowdfunding and wholesale investor settings. KiwiSaver contribution increases expand the long-term private savings pool. Whether that pool intermediates into productive scale-ups, rather than only liquid beta and housing, is the market test that matters for the next Halter-class round.
The Research and Development Tax Incentive’s five-year Motu evaluation for MBIE projected an economy-wide benefit about 4.2 times government investment—around a $6.8 billion GDP boost at midpoint. It supported 1,752 firms with $1.074 billion in credits and 65% of business enterprise R&D by 2023. Bang-for-buck of 1.4 beat the prior Growth Grants regime. That is a rare positive evaluated return. The IMF still wants settings kept under review so fiscal cost and deadweight stay visible against half-OECD GERD intensity.
Invest New Zealand briefings reported in mid-2026 pitched $25–35 billion of private investment over five years into data centres, renewables, fibre and transmission. BCG-linked materials cite potential to unlock up to about $70 billion of activity over a decade. Treat the envelope as stated ambition from briefings, not a gazetted hard target.
Datagrid’s Makarewa campus in Southland received full resource consent in March 2026. Coverage describes a campus on the order of 78,000 square metres and about 280 MW scale ambition, project costs commonly cited around $3.4–3.5 billion, a 140 MW Mercury electricity option, groundwater take near 220 million litres a year, and a target open around 2028. Consent detail includes multiple data halls and a large emergency-generator fleet. Local debate over water, generators, noise and transparency is a legitimacy test.
A 280 MW to 1 GW AI load on a system near 5 GW is a system-planning problem, not only a consenting one. If renewables and transmission lag, power prices rise for dairy processing, cold chain and metals. That can reverse the adoption case the IMF explicitly tied to electricity costs. The RBNZ’s AI-valuation caveat is the financial-stability mirror of the same build-out risk.
Government AI policy is deliberately adoption- and confidence-oriented. The July 2025 strategy, released by Minister Shane Reti, aligns with OECD AI Principles and light-touch enabling regulation, with Responsible AI Guidance for Business issued alongside. NZIAT channels up to $70 million into AI research and applications inside a wider advanced-technology platform. Light-touch rules reduce compliance drag for SME pilots. Enterprise transformation still needs data readiness, leadership and operating-model change—the gap Datacom measures at 4%.
Where the trade-offs bite
Growth upside competes with fiscal and monetary constraint. Public underwriting of skills, research platforms and grid capacity sits beside a medium-term consolidation path. An OCR at 2.75% raises the cost of capital for agritech scale-ups and datacentre equipment while CPI remains above the target band.
Datacentre foreign investment can anchor an applications advantage. It can also lift system costs and strain local consent. Water and noise politics are part of social licence. Dismissing them as pure obstruction ignores the price path facing energy-intensive exporters if generation and transmission lag load.
The Microsoft moderate $76 billion case assumes worker reallocation. The conservative $39 billion case is what unmanaged displacement looks like on the same modelling frame. The IMF said about one-third of New Zealand workers face displacement risk, particularly acute among women and younger cohorts. Active labour-market policy and public-service AI deployment are fiscal choices with distributional consequences for households.
RDTI generosity has an evaluated return that supports continuity. Review discipline keeps deadweight and Crown cost visible. Capital-markets deepening aims to thicken risk capital for scale-ups; calibration trades listing and wholesale frictions against investor protection and litigation cost. Neither extreme—permanent subsidy without scrutiny, nor thin markets that strand founders—serves productivity.
Second-order effects households and firms should watch
Primary AI edges defend wine and horticulture margins into MPI’s 2026–27 revenue path under global bulk pressure. Firm-level pruning and disease maps become export-price resilience for Marlborough contractors and packhouse operators.
AI user-skill demand without developer depth locks New Zealand into an applications-importer role. TIN200’s FY2025 print showed the top 200 tech exporters at $20 billion total revenue and $15.31 billion exports, with employment above 61,000. Tech is the third-largest export earner behind dairy and tourism. Whether process-AI or product-AI dominates that cohort shapes the export mix over the next cycle.
KiwiSaver pool growth only lifts productive capital if IPO, secondary and wholesale settings intermediate into scale-ups. RBNZ caution on AI-related asset valuations can tighten financial conditions for non-AI capex if equity corrects. Spillovers would hit broader business investment, not only model-training budgets.
Southland’s bargain is local rates, jobs and water against consented megawatts. Failure of financial close after consent strands political capital and chills the next campus. Gentailer power-purchase agreements and Transpower works timing become hidden industrial policy—for AI loads and for the energy-intensive export base that already carries New Zealand’s terms of trade.
Women and youth displacement without reskilling widens participation gaps. That undercuts the household-income path embedded in GDP-per-capita scenarios that assume reallocation works smoothly.
Historical context: another tech wave on thin capital
The closest analogue is the 1970–2022 divergence from Nordic peers on GDP per hour worked. New Zealand started roughly level with Denmark, Finland and Sweden. It then fell about 40% behind as capital intensity lagged. AI is another general-purpose technology. Without capital deepening, diffusion can stall the way earlier ICT waves did in measured multifactor productivity even as individual software firms succeeded offshore.
Cloud and SaaS offer a domestic warning. High preparedness and global names coexisted with slow economy-wide multifactor productivity gains. Equating flagship firms with national transformation is the recurring error in small advanced economies.
Commodity technology waves in agriculture—genetics, irrigation, precision dairy—show the opposite channel. Primary labour productivity outperformance is the template manufacturing never matched. AI agritech fits that template if power markets clear and service models reach SME growers, not only estate vineyards and corporate herds.
Australia sits beside New Zealand in the World Bank overperformer set and has published a National AI Plan. EY-Parthenon models an A$95–116 billion lift to the Australian economy by the mid-2030s, with multifactor productivity gains of 2.0–2.4% and tens of thousands of additional full-time equivalent jobs. Larger capital markets and investment envelopes convert readiness faster. That scale mirror is why reforms on equity markets and electricity bite harder in New Zealand.
The counter-argument
The strongest opposing read is straightforward. New Zealand is already doing enough on the discovery side. Top-ten AIPI ranking, World Bank overperformer status, 91% organisational AI use, TIN200 at $20 billion revenue, Halter at a $2 billion valuation, primary productivity outperformance, a modelled $76 billion generative AI prize, an RDTI that evaluates at about 4.2 times return, a first national AI strategy, and a multi-billion datacentre investment pitch add up to a country leaving only execution details on the table.
On that view, Datacom’s 4% transformed-operations share is lagging-indicator noise. Pilots precede transformation. Chamber hand-raises understate quiet SME uptake. Electricity and capital-market frictions are chronic complaints that have not stopped Cropsy-class firms from exporting capability to California and Cognac.
The evidence for the conversion thesis is harder to wave away. Capital stock per hour remains about half peer levels on the IMF’s accounting. GERD is about half the OECD average on Treasury’s. The transformed-operations share fell, not rose, in the latest Datacom print. The IMF’s named impediments—SME scale-up costs, power price volatility, shallow risk finance, weak innovation outcomes—are the same frictions founders and surveys describe. Scenario models book upside that assumes complementary inputs the data say are incomplete.
Both reads can be honest at once. Vine-level AI is already productive in primary industries. National conversion of preparedness into multifactor productivity still fails on capital depth, stable power, skills transition and scale-up frictions the Fund named. Confusing chatbot pilots with a growth strategy is the risk. So is betting the grid on unfinanced hyperscale hopes without matching generation and transmission.
Open questions
Will Datacom’s 2027 Index show the transformed-operations share rising off 4%, or will scattered use-cases persist? Does Datagrid reach financial close, convert the Mercury option into a firm power-purchase agreement, and align Transpower works so electrons match consented megawatts by 2028?
Do Capital Markets Reforms Phase Two and Commerce Act modernisation pass and thicken risk capital enough for a wider TIN200 cohort beyond a handful of scaled names? Does any RDTI parameter review protect the evaluated engine or tighten eligibility in ways that shrink business enterprise R&D?
Do Stats NZ labour and multifactor productivity prints move beyond 0.8% and −0.9% within two to three years as generative AI hours and primary AI systems diffuse? What displacement package—design, fiscal scale, and female and youth targeting—addresses the IMF’s one-third at-risk share without stranding household incomes?
Near-term watches are concrete. The OCR path and the next inflation prints condition the cost of capital for every scanner, collar and server hall. MPI wine export volumes will show whether quality and yield tools defend margins against bulk-price pressure. Public-service and large-corporate procurement will shape demand for the AI-user skills PwC already sees dominating job postings.
Over two to three years the scoreboard is measured multifactor productivity, electricity system capacity and price volatility, and whether Halter-, Hectre- and Cropsy-class firms become a thick cohort rather than exceptions. New Zealand does not lack AI readiness on the IMF’s index. It lacks the capital depth, stable power and commercialisation machinery to turn that readiness into the productivity growth a nascent recovery still needs. The next Reserve Bank statements, Datacom index, Datagrid close signals and Stats NZ productivity release will show whether conversion has begun—or whether top-ten preparedness remains a ranking without a dividend.
