Locally woven.
Distributed compute.
Sovereign AI compute, distributed across Aotearoa.
Thatch is a distributed modular AI inference cloud — self-contained compute units, from heat-pump-sized to commercial-scale, running useful AI workloads from many sites across New Zealand, rather than one large data centre.
AI is being built offshore.
So is the dependency.
The world's AI infrastructure is consolidating in a handful of hyperscale data centres, mostly outside New Zealand. Every workload we run, every dollar we spend, every byte of data we process leaves the country — and depends on offshore networks staying available. As AI moves from optional to operational, that dependency becomes a strategic risk for every NZ business, council, and public-sector organisation.
Aotearoa has a rare advantage. Roughly 85% of our electricity is already renewable. Our distribution networks are among the most resilient in the hemisphere. We have the power, the connectivity, and the capability to host meaningful AI compute on our own soil — without waiting for someone else to build a hyperscaler here.
What’s missing isn’t the energy or the demand. It’s the model.
A distributed network of AI inference nodes, built on the infrastructure we already have.
Thatch deploys self-contained, enterprise-grade compute units across sites with available power and fibre — alongside existing infrastructure rather than instead of it. Each unit runs asynchronous AI inference workloads close to where they’re needed. A central orchestration layer ties the fleet together so it operates as a single, coherent cloud — and so individual nodes can flex with the local grid.
- 01
Sovereign by default
Workloads run on New Zealand soil, on New Zealand renewables, under New Zealand law. Data, inference, and infrastructure spend stay onshore.
- 02
Distributed by design
Many small sites, not one large facility. Faster to deploy, more resilient against single-point failure, easier to scale incrementally as demand grows.
- 03
Grid-friendly load
Asynchronous AI inference jobs can be queued, paused, or shifted in seconds. Each Thatch node is a digital load that works with the network, not against it.
- 01
Sovereign by default
Workloads run on New Zealand soil, on New Zealand renewables, under New Zealand law. Data, inference, and infrastructure spend stay onshore.
- 02
Distributed by design
Many small sites, not one large facility. Faster to deploy, more resilient against single-point failure, easier to scale incrementally as demand grows.
- 03
Grid-friendly load
Asynchronous AI inference jobs can be queued, paused, or shifted in seconds. Each Thatch node is a digital load that works with the network, not against it.
Sized to the site.
Enterprise-grade.
Built for the outdoors.
Thatch nodes come in two outdoor form factors — a compact unit roughly the footprint of a domestic heat pump, and a larger commercial cabinet for higher-demand sites. Both are fully self-contained: they handle their own power, cooling, and connectivity, and are designed to install quickly and run reliably at sites across the country.
-
Built for
- Outdoor cabinet installation
- Light-industrial, commercial, council, and infrastructure sites
- Long service life with redundant cooling and enterprise hardware
-
Compute
- 4× NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs
- 2× AMD EPYC CPUs
- 1.5 TB high-speed memory
- Dell PowerEdge XE7745 liquid-cooled server platform
-
Connectivity
- 24-port gigabit switching
- Fibre-ready uplink
- Secure remote management
-
Cooling
- Integrated HVAC with outdoor heat-pump unit
- Precision air supply and return for sustained AI workloads
- Quiet operation suitable for commercial and residential proximity
Every watt, put to work twice.
At commercial scale, the larger Thatch cabinet runs its AI workloads and hands the leftover heat straight to the building — warming water for pools, hospitals, and industry that would otherwise burn fuel or pay to produce it.
Asynchronous AI inference, hosted close to home.
Thatch is purpose-built for the workloads where latency tolerance and locality both matter — the jobs that benefit from running in New Zealand, on flexible distributed capacity, without paying the round-trip to an offshore hyperscaler.
-
Batched AI inference
Document analysis, summarisation, classification, RAG over local data sources.
-
Software development workloads
Code generation, test automation, build acceleration, agentic development tooling.
-
Domain-specific inference
Industry-tuned models for legal, healthcare, government, education, and primary-sector use cases.
-
Sovereign and sensitive workloads
Workloads that need to stay onshore for compliance, governance, or contractual reasons.
-
NZ businesses
Onshore inference for software companies, AI-native startups, and operational AI teams.
-
Councils and public sector
Sovereign inference for government workloads where data residency and governance matter.
-
Network and infrastructure partners
Distribution businesses and grid operators looking to monetise existing footprint and add a flexible, controllable load.
-
Site hosts
Commercial property, industrial parks, and developers with available power, fibre, and outdoor space.
Our own products run here first.
Thatch starts with the workloads we know best. Agentic Intelligence and its product companies are moving their AI inference onto Thatch — real jobs running onshore, proving the platform as we open it to more customers.
Software Factory runs
on Thatch.
Software Factory is our agent platform for building software — a steering agent that runs coding agents around the clock, answering their questions and reviewing what they ship. Its planning and review loops run on Thatch: patient, queued work that tolerates a wait and never leaves the country.
-
Asynchronous by nature
Planning, review and the long loops queue happily. Exactly the shape of work distributed capacity is built for.
-
Onshore by default
A run pinned to a Thatch model stays on New Zealand soil, on New Zealand renewables, under New Zealand law.
-
Capacity we own
No per-token bill on the jobs that run longest, so the work can take the time it needs.
- plan
- implement
- validate
- review
- merge
Common questions.
Asynchronous AI inference. Batched document processing, RAG over local data, code generation and agentic development tooling, domain-tuned models for regulated industries, and any workload that benefits from running onshore on flexible distributed capacity.
Many small sites instead of one large facility. Each Thatch node is a self-contained outdoor unit — from a compact heat-pump-sized footprint up to a larger commercial cabinet. Faster to deploy, more resilient against single-point failure, easier to scale incrementally, and far easier on local grids.
Sites with available power and fibre. Light-industrial, commercial, council-owned, and infrastructure sites are all suitable — alongside existing buildings, substations, or yards rather than instead of them.
Workloads run on New Zealand soil, on New Zealand renewables, under New Zealand law. Data, inference, and infrastructure spend stay onshore by default. No data leaves the country unless a customer explicitly chooses to route it offshore.
Asynchronous inference jobs can be queued, paused, or shifted in seconds. That makes each node a controllable digital load — one that flexes with local generation and demand rather than fighting it.
The compute sheds heat whatever it runs, so capturing it is near-free thermal energy — offsetting whatever a host already pays to heat water, whether that is electricity, gas, or diesel. The case is strongest where it displaces fossil fuel, cutting running costs and emissions together.
We are in active development with anchor customers and network partners. If you have a workload that fits the profile, the fastest path is to talk to us — early customers help shape capacity and rollout.