Transparent rates. Billed by the minute.
Pricing is shown in Australian dollars, exclusive of GST. There is no minimum commitment, no spend floor, no egress charge inside Australia. Volume and reserved-capacity discounts are available — see notes below.
Per-GPU-hour rates
Instances are billed in 60-second increments. Storage and networking inside the facility are included; outbound traffic outside Australia is metered separately.
| Hardware | VRAM | Available | Best for | AUD / GPU-hour |
|---|---|---|---|---|
| NVIDIA B200 SXM6 Frontier | 192 GB HBM3e | On request | Frontier training, 100B+ inference, FP4 | $6.50 |
| NVIDIA H200 SXM5 | 141 GB HBM3e | 8 | Training, long-context inference, 70B+ models | $4.20 |
| NVIDIA H100 SXM5 | 80 GB HBM3 | 16 | Cost-effective training and batch inference | $3.20 |
| NVIDIA RTX PRO 6000 Blackwell Server Edition | 96 GB GDDR7 | 16 | Inference workhorse (FP4) — replaces L40S | $1.95 |
| AMD Instinct MI325X | 256 GB HBM3e | On request | Memory-heavy inference, non-CUDA stacks (ROCm) | $3.20 |
Inventory shown is the configured first phase. We're adding capacity as utilisation justifies; ask if you need more than is listed.
Estimate your cost
Pick hardware and a duration. The result is an instant estimate at retail rate; reserved-capacity discounts apply separately.
Estimate excludes GST and any outbound bandwidth charges. Real invoices reflect per-minute usage, not the rounded figures above.
Pay only for what you use
| Service | Unit | Best for | AUD price |
|---|---|---|---|
| Qwen-3 inference (general + code) | per 1M input tokens | Chat, RAG, code assistants | $0.50 |
| Qwen-3 inference | per 1M output tokens | — | $1.50 |
| Fine-tuning service | per job (LoRA, 7–14B params) | Domain models on private data | from $180 |
| Hosted endpoint for your fine-tune | per GPU-hour reserved | Always-on private inference | $1.95–$4.20 |
Indicative example: 1,000,000 input + 200,000 output tokens through Qwen-3 ≈ AUD $0.80.
Engineering time
Discovery + scoping
A fixed-fee 1–2 week engagement where we shape the problem, run a small technical spike, and return a written proposal. AUD $4,000 – $9,000 depending on scope.
Build engagements
Day rate AUD $1,400 – $2,200 per engineer depending on seniority. Typical small projects 2–4 weeks; larger systems 6–12 weeks. Compute is billed on top at the rates above, transparently.
What we promise
Uptime
99.9% monthly availability per instance, measured at the hypervisor. Credits applied automatically if we miss it. The underlying facility runs on a Tier III design with N+1 power and cooling.
Data residency
All workloads run at Mayfield West, NSW. No replication to US or EU regions without an explicit, written request from you. Backups live in the same jurisdiction.
Support response
Email replies inside 4 business hours. Critical incidents reach an on-call engineer in under 30 minutes during business hours; under 2 hours overnight.
Volume discounts
Reserve a GPU for 30 days and pay 25% less. Reserve for 12 months and pay 40% less. Talk to us if you need a custom shape.