Briefing · Following The Great NZ AI Roadshow

AI's carbon question, with a Taranaki lens

AI adoption carries a real and measurable environmental cost. Most of it occurs in overseas data centres, in the part of carbon reporting almost nobody tracks. This briefing summarises the research evidence, the New Zealand position, and why the question lands differently in an energy region.

Prepared by Cameron Visagie For Jenn Patterson, Venture Taranaki June 2026
Use ← → arrow keys, swipe, or the buttons below to navigate
01 / The global picture

Emissions are already material, and growing

Peer-reviewed estimates vary, but they agree on direction and scale: AI is driving data centre expansion on grids that remain mostly fossil-fuelled.

32–80 Mt CO₂
Projected global AI emissions in 2025, roughly the carbon footprint of New York City
de Vries-Gao (2025), Patterns
2.18%
Share of total US emissions already from data centres (105 Mt CO₂e), 56% fossil-powered
Guidi et al. (2024)
3% → 9%
Data centre share of global electricity, now versus projected by 2030
Hankendi, Coskun & Sovacool (2025), iScience
800 TWh
Projected global data centre electricity by 2026, including AI and crypto; almost twenty times NZ's entire annual generation
IEA forecast, cited in Hankendi et al. (2025); MBIE energy statistics
02 / Where they sit

Four in ten of the world's data centres sit in one country

These are general-purpose facilities, not AI-specific: Data Center Map tracks roughly 11,100 across 174 countries. The hyperscale tier that does most AI work is far smaller, 1,360 facilities at the end of 2025 with almost 800 more in the pipeline (Synergy Research), and not all of those run AI either.

World map of data centres by location, Data Center Map

United States 4,328 facilities; Canada 290; Europe's dense cluster led by Germany, the UK, the Netherlands and France; China 368; India 301; Australia 278; Japan 256. The cautionary tale is Ireland: data centres already take 22% of national electricity, up from 5% in 2015, heading for ~31% by 2034 (CRU). Source: Data Center Map, datacentermap.com; map © Mapbox © OpenStreetMap.

Map of New Zealand data centres, Data Center Map

New Zealand: 62 facilities. Auckland 32 (roughly half), Wellington 8, Hamilton 5, Christchurch 5, Rotorua 4, Invercargill 3, Dunedin 2, and exactly one in Taranaki. Source: datacentermap.com/new-zealand; map © Mapbox © OpenStreetMap.

03 / Where the emissions hide

The reporting blind spot

When a New Zealand organisation uses an overseas-hosted AI tool, the emissions are real, but they fall outside everything we currently count.

A NZ organisation uses AI
Copilot, ChatGPT, Claude, Gemini. Every prompt consumes compute.
→
Compute runs offshore
Mostly in US, Asian, or European data centres, frequently on fossil-heavy grids.
→
Emissions occur there
They never appear in NZ's national inventory or the organisation's electricity bill.
→
Scope 3: untracked
Cloud and AI emissions sit in Scope 3, where standard reporting rarely isolates them (Reynolds, 2025).

The research literature contains no New Zealand-specific method for accounting AI-related emissions. The frameworks exist; nobody here has applied them yet.

04 / Whose impact is it?

Offshore is not the same as no impact

A blind spot in reporting does not make the harm hypothetical. New Zealand's AI usage creates real environmental pressure in other people's regions, and responsibility follows consumption, not geography.

Water, drawn in stressed regions
Cooling AI servers could consume hundreds of billions of litres of water globally each year, and many data centre hubs sit in regions already under water stress (de Vries-Gao, 2025; Xiao et al., 2025).
Hardware burdens, exported
The mining, manufacturing, and disposal of AI chips and servers carry major embodied emissions and ecological harms, frequently borne by communities in the Global South (Falk et al., 2024; Schneider et al., 2025).
Host communities, carrying the cost
Grid pressure, heat, land use, and energy price effects land on the communities hosting the infrastructure, who often see few of the benefits realised elsewhere (Falk et al., 2024; Guidi et al., 2024).

This is precisely what Scope 3 exists for. The logic New Zealand already applies to the embodied carbon of imported goods should apply to imported compute: count it, name it, and report it as our own.

05 / Why location matters

The same query, ten times the footprint

Grid emission factors, kg CO₂e per kWh. Where the data centre sits largely determines what your AI usage emits.

New Zealand
~0.07
United Kingdom
~0.23
United States
~0.40
Australia
~0.72

Indicative national grid averages; the NZ figure reflects Ministry for the Environment measurement guidance. A clean home grid does not make our AI clean, because almost none of it runs here, and the hardware carries its own footprint wherever it sits. What NZ's roughly 88% renewable grid does offer is the single biggest reduction lever available: hosting location.

06 / What actually reduces it

Four levers, and the largest two are about location

The footprint is not fixed. The research identifies where compute happens as mattering more than how much.

~50%
Locate on clean grids
Siting on low-carbon electricity can roughly halve combined carbon and water footprints (d'Orgeval et al., 2026; Xiao et al., 2025).
35–56%
Carbon-aware scheduling
Shifting AI workloads to lower-carbon regions and times cuts inference emissions substantially (Dodge et al., 2022; Chien et al., 2023).
up to ~50%
Efficient design and cooling
High-density, liquid-cooled facilities deliver the lowest emissions per unit of compute (Mu et al., 2025; Patel et al., 2025).
several-fold
Hardware and model choice
Each accelerator generation cuts compute carbon intensity; smaller models suffice for many tasks (Schneider et al., 2025).
07 / New Zealand, right now

The hyperscale build-out has arrived

NZ's renewable grid is a genuine advantage, but it cuts both ways: it attracts energy-hungry infrastructure to a grid already under pressure.

  • December 2024
    Microsoft launches NZ North, Westgate, Auckland
    Backed by a $300m deal with Contact Energy supporting the Te Huka geothermal field, allowing a 100% renewable claim.
  • September 2025
    AWS launches its NZ$7.5 billion New Zealand Region
    Three availability zones, estimated NZ$10.8b GDP contribution, renewable via a Mercury NZ wind agreement (Turitea South). CDC, DCI and NextDC are also building.
  • Ongoing
    Grid roughly 88% renewable, but constrained
    Cook Strait transmission upgrades not due until 2031; new builds expected to spread toward South Island renewables.
  • Unresolved
    The accounting question
    Overseas AI usage is invisible in domestic inventories; locally hosted AI shows up in NZ's own demand and emissions. Neither is measured at organisational level.

The question nobody is answering

NZ organisations adopting AI today cannot say what their usage emits. Not because it is unknowable, but because nobody has connected the three pieces: usage data, energy estimates per query, and grid emission factors for the hosting region.

MfE publishes the emission factors. Providers publish partial sustainability data. The literature publishes per-query energy estimates. The method exists; the application does not.

08 / The Taranaki lens

A live regional development question, cutting both ways

The case for leaning in

  • World-class offshore wind resource in the South Taranaki Bight
  • Energy-intensive industry gives new generation an anchor customer, strengthening the offshore wind investment case
  • Existing transmission infrastructure and a deep-water port
  • Oil and gas workforce with roughly 68% skills overlap with offshore renewables
  • Compute on Taranaki renewables would carry among the lowest carbon intensities available anywhere

The costs that need eyes open

  • A single large data centre can draw as much electricity as a town, on a constrained grid
  • Cooling can consume substantial water (de Vries-Gao, 2025)
  • Infrastructure costs often land on local communities while benefits accrue elsewhere (Falk et al., 2024)
  • Few permanent jobs relative to capital: Microsoft's Westgate region, ~50 ongoing FTEs
  • Iwi and community engagement must precede investment attraction, not follow it
Taranaki Offshore Partnership CIP and the NZ Super Fund. Up to 1 GW, around $5 billion.
Seeking government risk-sharing; FID targeted by 2030.
South Taranaki Offshore Wind BlueFloat, Elemental Group, Energy Estate. ~900 MW, 600+ jobs.
First of four proposed projects totalling up to 5 GW.
The strategic link Offshore wind needs anchor demand. Data centres need clean power and social licence. Taranaki could solve both halves together, deliberately.
The conversation worth having now, before decisions are made elsewhere.
09 / The other side of the ledger

New Zealand is building generation from the bottom up

While hyperscale demand arrives from the top, supply is being added household by household, school by school, paddock by paddock. The question is who that freed-up headroom ultimately serves.

~250 MW
Taranaki's solar pipeline
Te Matakupenga at New Plymouth Airport (opened 2025, named by Puketapu Hapū), plus consented and proposed farms at Ōpunake (Harmony 75 MW; EFL 80–110 MW with battery), Kaponga and Stratford, against barely 13 MW operating today.
$30m
Solar on Schools, announced June 2026
Up to 500 schools with panels and batteries, ~10.1 GWh a year, $8,000 annual savings per school and 5–7 year payback (EECA and Ministry of Education).
10 kWh
Household exports unlocked
The Electricity Authority has doubled rooftop export limits, removed building consent for panels, and mandated fair buyback and time-of-use plans from mid-2026.
2%/yr
Demand growth to 2050
MBIE expects steady growth from EVs and electrification, while large energy users are already constrained by lack of supply. Every distributed kWh frees grid headroom that large loads will compete for.
The pattern worth naming: households and schools fund the panels with their own capital, earn 8–16c for exports against 28–36c retail, and the headroom they free becomes marketable to large industrial loads, data centres included. Benefits centralised, costs distributed: the same dynamic as offshore AI hosting, playing out domestically. A region that sees this clearly can negotiate rather than absorb.
10 / Measuring it

A starting framework: three steps, one formula

Precision improves over time; the discipline of measuring starts now.

Inventory the AI estate
Which tools, which providers, which data centre regions host the workloads, and how much usage (prompts, tokens, licences, cloud spend).
Estimate energy, then emissions
Apply published per-query energy estimates and the hosting region's grid emission factor. Use provider disclosures where they exist; flag the gaps.
Report it in Scope 3, visibly
A named line in Scope 3 reporting, not dissolved into general procurement. What is named gets managed.
From the Ministry for the Environment's measurement guidance:  Emissions (kg CO₂e) = Energy used (kWh) × grid emission factor of the hosting region

The hard part is not the formula. It is obtaining honest values for the left-hand side, which is why provider transparency requirements belong in AI procurement policy (Hankendi et al., 2025; Guidi et al., 2024).
References

Sources, APA 7th edition

Ranges are quoted where the literature disagrees; no figure in this briefing is invented. This is a discussion briefing, not a formal assessment.

Academic literature

Chien, A., Lin, L., Nguyen, H., Rao, V., Sharma, T., & Wijayawardana, R. (2023). Reducing the carbon impact of generative AI inference (today and in 2035). Proceedings of the 2nd Workshop on Sustainable Computer Systems. https://doi.org/10.1145/3604930.3605705

d'Orgeval, A., Sheehan, S., Avenas, Q., Assoumou, E., & Sessa, V. (2026). Generative AI impact assessment through a life cycle analysis of multiple data center typologies. Applied Energy. https://doi.org/10.1016/j.apenergy.2025.127288

de Vries-Gao, A. (2025). The carbon and water footprints of data centers and what this could mean for artificial intelligence. Patterns, 7(1), Article 101430. https://doi.org/10.1016/j.patter.2025.101430

Dodge, J., Prewitt, T., Combes, R. T. D., Odmark, E., Schwartz, R., Strubell, E., Luccioni, A., Smith, N. A., DeCario, N., & Buchanan, W. (2022). Measuring the carbon intensity of AI in cloud instances. Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3531146.3533234

Falk, S., Van Wynsberghe, A., & Biber-Freudenberger, L. (2024). The attribution problem of a seemingly intangible industry. Environmental Challenges. https://doi.org/10.1016/j.envc.2024.101003

Guidi, G., Dominici, F., Gilmour, J., Butler, K., Bell, E., Delaney, S., & Stoffi, F. B. J. (2024). Environmental burden of United States data centers in the artificial intelligence era. arXiv. https://doi.org/10.48550/arxiv.2411.09786

Hankendi, C., Coskun, A. K., & Sovacool, B. (2025). Why transparency matters for sustainable data centers and carbon-neutral artificial intelligence (AI). iScience, 28. https://doi.org/10.1016/j.isci.2025.113705

Mu, H., Lu, Y., Lepre, C., Lightfoot, C., Smiley, C., & Crenshaw, R. (2025). Artificial intelligence and its carbon footprint. 2025 Systems and Information Engineering Design Symposium (SIEDS), 31–35. https://doi.org/10.1109/sieds65500.2025.11021149

Patel, A., Mahalingam, N., & Patel, R. (2025). The environmental impact of AI servers and sustainable solutions. arXiv. https://doi.org/10.48550/arxiv.2601.06063

Reynolds, D. J. (2025). Carbon reporting practices in the NHS: Emissions and omissions relating to artificial intelligence. Journal of Medical Internet Research, 27. https://doi.org/10.2196/79174

Schneider, I., Xu, H., Benecke, S., Patterson, D., Huang, K., Ranganathan, P., & Elsworth, C. (2025). Life-cycle emissions of AI hardware: A cradle-to-grave approach and generational trends. arXiv. https://doi.org/10.48550/arxiv.2502.01671

Xiao, T., Nerini, F. F., Matthews, H., Tavoni, M., & You, F. (2025). Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA. Nature Sustainability, 8, 1541–1553. https://doi.org/10.1038/s41893-025-01681-y

New Zealand and Taranaki sources

Data Centre Magazine. (2025, September). AWS launches New Zealand data centre & infrastructure region. https://datacentremagazine.com/news/aws-launches-new-zealand-data-centre-infrastructure-region

iStart. (2025, September). Unpacking AWS' local data centre and training claims. https://istart.com.au/news-items/unpacking-aws-local-data-centre-and-training-claims/

Ministry for the Environment. (2024). Measuring emissions: A guide for organisations. Purchased electricity, heat and steam emission factors. https://measuringemissionsguide.environment.govt.nz/5_purchased_energy.html

Ministry of Business, Innovation & Employment. (2026). New Zealand energy sector greenhouse gas emissions. https://www.mbie.govt.nz/building-and-energy/energy-and-natural-resources/energy-statistics-and-modelling/energy-statistics/new-zealand-energy-sector-greenhouse-gas-emissions

Ngāmotu News. (2026, March 24). $5b offshore wind project stalled without Government sharing some risk. https://www.ngamotunews.co.nz/2026/03/24/5b-offshore-wind-project-stalled-without-government-sharing-some-risk/

South Taranaki Offshore Wind Project. (n.d.). Project overview. https://southtaranakioffshorewindproject.com/

Commission for Regulation of Utilities (Ireland). (2025). CRU publishes its decision on new electricity connection policy for data centres. https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/

Taranaki Offshore Partnership. (n.d.). About the project. https://taranakioffshorewind.co.nz/about-the-project/

Ministry of Business, Innovation & Employment. (2026, June 3). Government announces new Solar on schools programme. https://www.mbie.govt.nz/about/news/government-announces-new-solar-on-schools-programme

New Plymouth Airport. (2025). Te Matakupenga solar farm. https://airportstories.co.nz/solar-power-initiative/

pv magazine Australia. (2024, September 10). Harmony secures consent for almost 150 MW of solar in NZ. https://www.pv-magazine-australia.com/2024/09/10/harmony-secures-consent-for-almost-150-mw-of-solar-in-nz/

Datacentre220. (2025). New Zealand's data centre boom [citing NZTech, Empowering Aotearoa New Zealand's Digital Future]. https://www.datacentre.co.nz/newsroom/new-zealands-data-centre-boom-powering-nzs-digital-economy

Industry trackers (facility counts; methodologies vary)

Data Center Map. (2026). Data centers worldwide [interactive database]. https://www.datacentermap.com; New Zealand: https://www.datacentermap.com/new-zealand/

Synergy Research Group. (2026, April). Hyperscale operators to account for 67% of all data center capacity by 2031. https://www.srgresearch.com/articles/hyperscale-operators-to-account-for-67-of-all-data-center-capacity-by-2031

ABI Research. (2026, March). How many data centers are there and where are they being built? https://www.abiresearch.com/blog/data-centers-by-region-size-company

Cloudscene data, via Cargoson. (2025, November). Number of data centers by country. https://www.cargoson.com/en/blog/number-of-data-centers-by-country

Statista. (2026, April). Data centers worldwide by territory. https://www.statista.com/statistics/1228433/data-centers-worldwide-by-country/

Cameron Visagie · New Plymouth