Bare-metal GPU clusters, regional data centres and a model token API — one local partner for the whole stack, on infrastructure we own and operate end to end.
Rent the hardware, site it in our data centres, or skip the cluster entirely and call a model by the token.
H100, H800/A800, L40S and RTX 4090 — bare-metal clusters or elastic cloud instances, billed hourly.
View GPU optionsFour owned and partner sites from Auckland to Frankfurt, with NZ data residency and clean hydro power.
See the regionsOne OpenAI-compatible API for Qwen, DeepSeek and Llama, billed by the token with no minimum.
Read the pricingWe already run fibre, voice and managed IT across New Zealand — compute sits on the same engineering team.
The Auckland node keeps training data and inference inside New Zealand — no offshore transfer for privacy-sensitive work.
We own the racks, the fabric and the peering. That is why provisioning takes minutes instead of weeks.
Chinese and English speaking architects on the same 24-hour response promise as the rest of our business.
Their own data centre capped training at 64 GPUs. We deployed a 512-GPU H800 bare-metal cluster with 3D parallelism and a self-healing scheduler that recovers failed nodes in minutes.
Sensor labelling, training and simulation rendering with sharp demand peaks. An elastic A800 cluster plus parallel file storage let them scale weekly instead of quarterly.
Whether it's a thousand-GPU training cluster, a regional deployment or a token API integration, our architecture team replies with a proposal and quote within 24 hours.
Book a technical consultation