Products Solutions Pricing Infrastructure Security Compute for equity Company

Talk to our team Get a quote

Products

Blackwell, from a single node to a training cluster.

Two NVIDIA Blackwell platforms in dedicated clusters of 8, 24 or 32 servers on a 400G fabric, reserved for a term. Bare metal today; managed platform layers follow on the same infrastructure.

See pricing Compare specifications

RTX PRO 6000

Fully reserved

Through 2026 · reservations for 2027 open

HGX B300

Reservations open

Deliveries from December 2026

Specifications

Two platforms, side by side.

Two nodes, one fabric. Clusters of 8, 24 or 32 servers, scoped per contract.

Node	NVIDIA HGX B300 Blackwell Ultra · AI factory standard	NVIDIA RTX PRO 6000 Blackwell Server Edition · workstation-class cluster
System	8-GPU HGX B300 system, built by Supermicro	8-GPU server, built by Supermicro
GPUs per node	8 × B300 SXM	8 × RTX PRO 6000 Blackwell Server Edition
GPU memory	288 GB HBM3e per GPU · 2.3 TB per node	96 GB GDDR7 per GPU · 768 GB per node · 14.3 TB/s aggregate
Compute (FP4)	120 PFLOPS	32 PFLOPS
GPU interconnect	1.8 TB/s NVLink, GPU to GPU	PCIe Gen 5.0
Node fabric	1.6 Tb/s, ConnectX-8 ready	400G high-speed fabric for multi-node scaling
Form factor & power	3,000 W redundant, Titanium-grade power supplies	4U rackmount, optimised for air cooling
Best for	Large-scale LLM training, foundation models, high-volume inference	Agentic AI, multi-app workflows, inference, professional visualisation
Availability	Reservations open · deliveries from December 2026	Fully reserved through 2026 · 2027 reservations open
Pricing	Multi-year reservation · rate on quote · see pricing	Reservation from 3 months · rate on quote · see pricing

Specifications as published by NVIDIA and Sapience AI for the configurations deployed. System memory per server is confirmed in your quote. Cluster configurations, storage and networking options are scoped per contract.

Platform

Choose how you run it.

Bare metal is live. Virtual machines, Kubernetes, SLURM and container runtimes are in development on the same fabric, so a cluster reserved today can move up the stack without moving data.

Bare metal

Live

Dedicated GPU servers with no virtualisation overhead. Direct hardware access, custom configurations and isolated tenancy.

Virtual machines

Coming

GPU-accelerated VMs with full root access, custom OS images, hourly billing and auto-scaling.

Kubernetes

Coming

Managed clusters with GPU node pools, auto-scaling, Helm charts and integrated monitoring.

SLURM

Coming

HPC workload manager with job scheduling, resource allocation, queue management and multi-user support.

Containers

Coming

Docker and Singularity support with pre-configured ML frameworks.

Developer tools

Roadmap

API, CLI, Terraform provider and SDKs for provisioning and automation.

Fabric, storage and support

Everything around the GPU, included.

Data-centre space, NVIDIA GPUs, high-speed fabric and a pre-configured ML stack, run by our team so yours stays on the models.

Networking

400G fabric for multi-node scaling; 1.6 Tb/s ConnectX-8-ready fabric and 1.8 TB/s NVLink on HGX B300 nodes.

Storage

Encrypted NVMe on every node. AES-256 at rest, TLS 1.3 in transit.

Tenancy

Single-tenant by default: dedicated compute, storage and network with hardware-level segmentation. Private clusters with contractual data residency on request.

Support

24/7 monitoring and technical support, included with every node.

Models

Run open-source or proprietary models. Full-stack infrastructure — hardware, network and software — with no hyperscaler dependency.

Ready to size a cluster?

Pick the node, the count and the start date; the quote confirms rate, availability and the delivery date in writing.

Get a quote Talk to our team