Reserved clusters
Pre-deploymentDedicated HGX B300 nodes for sustained training, fine-tuning and inference. Choose the capacity and contract term around your workload.
From US$4.60/ GPU-hour · 48-month base rate
View pricingNVIDIA Blackwell UltraPre-deployment
Dedicated GPU clusters for training and inference.
Plan a reserved NVIDIA B300 cluster around your workload. Hardware, service scope and delivery are agreed before deployment.
Complete nodes. Reserved capacity. Phased expansion.
GPU products
Reserved clusters are the first planned offering. On-demand nodes and inference endpoints follow on the product roadmap.
On-demand nodes
RoadmapWhole eight-GPU nodes for evaluation and burst workloads, planned after reserved deployment.
US$6.90/ GPU-hour · indicative rate
View roadmapInference endpoints
In developmentPlanned dedicated and serverless endpoints for supported open-weight models and customer fine-tunes.
At launchPricing to be published
View roadmapIndicative planning rates. Final scope, availability and pricing are established in a signed agreement.
Cluster design
The hardware is one part of the system.
The planned cluster combines complete HGX B300 nodes, high-speed fabric, storage and operating support. Each layer is scoped against the workload.
Explore the architecture
- Compute
- Whole eight-GPU nodes with a planned 288 GB of HBM3e memory per GPU.
- Fabric
- GPU interconnects, storage networking and management access considered together.
- Storage
- Local scratch, parallel file storage and object storage scoped around the data.
- Operations
- Onboarding, monitoring, escalation and acceptance criteria defined in the agreement.
Expansion roadmap
A foundation built to grow.
We plan capacity in phases, beginning with Blackwell Ultra. Each deployment is shaped by customer commitments, infrastructure validation and agreed delivery conditions.
Initial clusters
Establish dedicated B300 clusters, with configuration and acceptance shaped by the workload.
Additional capacity
Plan further deployments around customer commitments, technical readiness and delivery conditions.
Future platforms and services
Evaluate new platforms as workloads evolve. Each phase requires its own validation and investment decision.
Forward-looking roadmap. Capacity and availability are confirmed through technical validation and signed agreements.
From requirements to acceptance
Customer requirements, system validation and commercial terms move forward together.
Define the workload
Confirm GPU count, model requirements, storage and expected use.
Validate the system
Agree hardware, fabric, cooling and workload acceptance tests.
Agree the terms
Align reserved capacity, contract term and procurement conditions.
Accept the deployment
Verify the agreed configuration and service scope before handover.
Compute questions
Product scope, availability and the next step.
Is capacity available today?
Altavast is in pre-deployment planning. Delivery and capacity remain subject to technical validation, project readiness and signed agreements.
What is included in a reserved cluster?
The planned scope combines complete GPU nodes, fabric, a storage allowance and operating support. Resource boundaries and service levels are defined in the agreement.
Can I review deployment details?
Detailed infrastructure information is shared with qualified customers through a confidential technical review before commitment.
How is the reserved rate determined?
Term length, reserved volume and advance structure determine the illustrative rate. The pricing calculator makes those assumptions visible.