NVIDIA B300 default
Blackwell Ultra silicon, dedicated to your reservation.
Sovereign AI
GPU clusters for the AI era.
Dedicated capacity for large-scale workloads. NVIDIA B300 Nodes built to Reference Architecture and sovereign by nature.
100% Australian & Compliant by Design
ISO 27001Tier 3+ Australian data centresCarrier neutral
What it is GPUs you can plan around.
GPUs you can plan around.
Reserved when it counts.
Reserved multi-node NVIDIA B300 (Blackwell Ultra) clusters built to the NVIDIA Reference Architecture. Reserve for up to 5 years for lower cost-per-hour compute.
Reservable GPU clusters
Amaze offers five reservable NVIDIA B300 (Blackwell Ultra) clusters in Sydney, from 8 to 1,024 GPUs each, from $5.00 per GPU-hour on 1 to 5 year terms, with the first cluster ready for service immediately.
| GPU model | Scale | Location | GPU memory | Cluster memory | Networking | Price and Term | Ready For Service | |
|---|---|---|---|---|---|---|---|---|
| NVIDIA B300 | 1,024 GPUs 128 nodes | Sydney, Australia | 288GB | 288TB | NVIDIA RA 800Gbps | From $5.00 / Hour 30% Prepayment 5 Years | Feb 2027 | Contact Sales → |
| NVIDIA B300 | 1,024 GPUs 128 nodes | Sydney, Australia | 288GB | 288TB | NVIDIA RA 800Gbps | From $5.00 / Hour 30% Prepayment 5 Years | Feb 2027 | Contact Sales → |
| NVIDIA B300 | 128 GPUs 16 nodes | Sydney, Australia | 288GB | 36TB | NVIDIA RA 800Gbps | From $6.80 / Hour 30% Prepayment 3 Years | Nov 2026 | Contact Sales → |
| NVIDIA B300 | 128 GPUs 16 nodes | Sydney, Australia | 288GB | 36TB | NVIDIA RA 800Gbps | From $6.80 / Hour 30% Prepayment 3 Years | Dec 2026 | Contact Sales → |
| NVIDIA B300 | 8 GPUs 1 node | Sydney, Australia | 288GB | 2.3TB | NVIDIA RA 800Gbps | From $7.50 / Hour 20% Prepayment 1 Year | Immediate | Contact Sales → |
All prices listed in USD. AUD pricing provided on quote.
Prices are indicative starting figures. Quotes are contract-term, memory-configuration and payment-structure specific. Speak to our solution architects to create a tailored quote based on your requirements.
Built for training at scale
Reserved silicon. Sovereign envelope.
Predictable per-node performance on NVIDIA B300 (Blackwell Ultra) silicon built to the NVIDIA Reference Architecture. Dedicated capacity under an Australian-governed control plane, root access available.
NVIDIA Reference Architecture
Purpose-built cluster fabric for collective ops.
Root access available
Full OS-level control on request for regulated workloads.
Sovereign by design
Australian-controlled infrastructure, Australian jurisdiction.
Robust networks and personal service.
Built to custom
Cluster shape
Silicon count, fabric and topology sized to the workload.
8 GPU Nodes
NVIDIA B300 Hardware
8 latest-generation NVIDIA B300 (Blackwell Ultra) GPUs per node.
40 Gbps
Transit per region
Carrier-grade Australian backbone.
< 15 min
P1 incident response
Australian-based on-call, 24×7.
Cluster fabric
For AI/ML training, inference, model fine-tuning and anything else.
Reserved clusters run on enterprise networking built to lower latency and minimise errors.
Catalogue
Run any open-weight model, in Australia.
Seven families, dozens of sizes, one Australian API. Pick the model, call the endpoint, ship the feature.
Qwen (Alibaba)
DeepSeek
Google · Gemma
Mistral AI
Kimi (Moonshot)
Z.ai · GLM
Microsoft · Phi
AI workloads at scale, in Australia.
Five things make Amaze reserved GPU clusters different from a generic public-cloud GPU instance.
NVIDIA B300 (Blackwell Ultra) default
Latest-generation dedicated silicon, 8 GPUs per node. Trillion-parameter model ready.
NVIDIA Reference Architecture
Purpose-built cluster fabric for collective ops. No oversubscribed paths, predictable throughput per job, every job.
Multi-node scale
Scale from a handful of GPUs to multi-node training fabrics without rearchitecting. Reference-architecture fabric end-to-end.
Root access available
Full OS-level control on request. Bring your own kernel tuning, drivers or scheduler for regulated or performance-critical workloads.
Onshore data for open-source models, end-to-end
Training data, checkpoints, fine-tuned weights and evaluation artefacts all stay in Australian regions. The end-to-end open-source training lifecycle, under Australian law.
How to start
Reserve a cluster in four steps.
All self-service. No tickets, no offshore approvals, no time-zone tax.
01
Talk to a solution architect
Workload sizing, topology, contract terms, all Australian.
02
Pick your silicon and footprint
NVIDIA B300 GPU nodes on the NVIDIA Reference Architecture, Intel Xeon CPU nodes, or both. Root access available on request.
03
Reserve and provision
Cluster live in days, not quarters. Same API as on-demand.
04
Run your training
Foundation-model training, large-scale batch, scientific compute.
Need an outcome story?
See HPC on Amaze.
GPU clusters fit inside the broader HPC overview, sovereign supercomputing, VFX rendering, scientific compute, all on the same Australian backbone.
Australian infrastructure built with leading global technology
Reserved GPU Clusters, frequently asked questions
What's the difference between GPU Compute and GPU Clusters?
GPU Compute is a reserved single-GPU-through-single-node product. GPU Clusters are multi-node reservations of NVIDIA B300 silicon built to the NVIDIA Reference Architecture — dedicated capacity for large-scale training and inference workloads.
What GPU silicon do you run?
NVIDIA B300 (Blackwell Ultra). Every cluster in the reservable inventory is B300, 8 GPUs per node, built to the NVIDIA Reference Architecture with a purpose-built cluster fabric for collective ops.
Can I get root access?
Yes. Root access is available on request for regulated or performance-critical workloads — bring your own kernel tuning, drivers or scheduler.
Is the reservation flexible?
Reserve weekly, monthly or annually. No exit penalties.
How does the 30% prepayment work?
Reserved GPU cluster contracts of 3 years or longer carry a 30% upfront prepayment against the total contract value. The 20% variant on the single-node option reflects its shorter 1-year term. Prepaid amounts draw down against monthly usage; the balance is invoiced through the term in AUD equivalent.
Can I reserve capacity before the ready-for-service date?
Yes. Every cluster in the inventory is booked against a specific ID (e.g. AU-SYD-B300-01) with an agreed ready-for-service date. Signing before RFS locks the price, term and footprint — billing begins when the cluster is delivered on the RFS date.
Is the capacity uncontended and dedicated?
Yes. Each reservable cluster is a fully dedicated allocation on NVIDIA B300 silicon — no shared kernels, no oversubscription, no time-sharing across tenants. The GPUs in your reservation only ever run your workload.
Reserve a sovereign cluster, by the week or by the year.
Talk to an Australian solution architect about the silicon, the topology, and the contract terms.