# Compute - NVIDIA B300, GLM 5.3, sovereign Canadian GPUs - SPUR Innovation

Source: https://spuric.com/en/compute.html (Markdown twin, generated 2026-10-08). Machine-readable offers: https://spuric.com/offers.json

AVAILABLE NOW - WATERLOO, CANADA

# NVIDIA B300.Sovereign Canadiancompute.

Leading-edge Blackwell GPUs racked, cooled, and serving real workloads today. Skip the AWS waitlist. Talk to a real engineer. Pay by the hour or by the month - no quote-walls.

[Try GLM 5.3 ->](https://ai.spuric.com/chat)
[Browse compute](https://ai.spuric.com)
[Request compute](https://spuric.com/en/compute-request.html)

288 GB HBM3e per GPU
2.3 TB HBM per node
8 GPUs per node
NVLink-5 Native FP4

B300 fleet

## Blackwell B300 SXM6, racked in Waterloo.

SPUR Compute is fielding NVIDIA B300 SXM6 chassis in our Waterloo data centre - the most powerful production GPU NVIDIA ships, with 288 GB HBM3e per GPU and 2.3 TB per node. Native FP4, NVLink-5, InfiniBand-ready. Hardware is racked and serving paying customers; no waitlist, no months-long queue.

- NVIDIA Blackwell B300 SXM6 nodes, 8 GPUs each

- NVLink-5 mesh, InfiniBand-ready cluster fabric

- SSH, Jupyter, VS Code, ComfyUI, Ollama, vLLM templates

- B200, B100, RTX 5090, RTX 4090 capacity available on request

[Browse compute](https://ai.spuric.com)
[Request a node](https://spuric.com/en/compute-request.html)

GLM 5.3 - live today

## Frontier open model, 1M context. OpenAI-compatible API.

GLM 5.3 (MoE, 1M context) is hosted in Canada and serving inference today via an OpenAI-compatible API. Frontier-class reasoning and native tool-calling without paying frontier-cloud markup, and without sending your prompts across the border. Point your existing SDK at ai.spuric.com and switch.

- OpenAI-compatible REST and streaming API

- Generous context window backed by 288 GB HBM3e per GPU

- No request logging, no training on your prompts

- Workspace, billing, and keys at ai.spuric.com

[Try GLM 5.3 ->](https://ai.spuric.com/chat)
[Model details](https://spuric.com/en/ai-models.html)

Workloads

## Training, inference, HPC. One fleet.

B300 capacity fits the workloads that used to require a hyperscaler contract: large-model training and fine-tuning, high-throughput production inference, scientific simulation, and frontier research. SPUR engineers help you size the cluster, pick the framework, and ship.

- LLM training and fine-tuning at 7B to 200B+ parameters

- Production inference at sub-50ms latency for hot models

- HPC simulation - molecular dynamics, climate, CFD, finance

- R&D environments for universities, labs, and corporate research teams

[Talk to an engineer ->](https://spuric.com/en/compute-request.html)

Why sovereign matters

## Your data never leaves Canada.

SPUR Compute is operated end-to-end inside Canadian jurisdiction. No US CLOUD Act exposure. No cross-border transfers. PIPEDA-aligned by default and engineered to SOC 2, PCI-DSS, and ISO 27001 controls. Hardware you can walk up and touch, run by engineers who answer the phone.

- Canadian-operated infrastructure, PIPEDA-aligned

- Operationally aligned to SOC 2 / PCI-DSS / ISO 27001

- Redundant power, redundant cooling, generator-backed

- Dedicated GPU options for regulated workloads (no multi-tenancy)

[Standards posture ->](https://spuric.com/en/standards.html)
[AI solutions](https://spuric.com/en/solutions.html)

## Ready for B300 capacity?

Try GLM 5.3 in your browser, or talk to a SPUR engineer about a dedicated allocation - hourly, monthly, or annual contract.

[Try GLM 5.3 ->](https://ai.spuric.com/chat)
[Request compute](https://spuric.com/en/compute-request.html)
[Talk to a human](https://apply.spuric.com/contact)