Industries
Trusted Partner for AI Tech Providers
-
151 Front
Street West -
100% Canadian
Owned & Operated -
Affordable
Solutions -
Mix & Match
Infrastructure
Location
151 Front Street West, TOR2
Connectivity
TORIX + 400+ peers
Bandwidth
1Gbps to 100Gbps unmetered
Compliance
SOC 2 Type II, PIPEDA, PHIPA
What AI providers needs
Standard colocation was not designed for GPU workloads.
A single NVIDIA H100 SXM draws ~700W. An 8-GPU server draws ~5kW to 6kW. A rack of four such servers draws 20kW to 24kW. Standard racks cannot handle this the power or the heat.
Up to 25kW per rack. Provisioned to your specs 208V and three-phase available.
In-row cooling units at rack level. Hot/cold aisle containment across the suite.
Lake Ontario-sourced cooling. High thermal capacity for sustained full-load operation.
Training pipelines, inference serving, checkpoint transfers no bandwidth caps.
Cross-connects to AWS, Azure, Google Cloud for hybrid AI workflows.
Predictable monthly cost regardless of GPU utilisation.
Challenge vs solution
What AI infrastructure requires.
Each row maps a requirement to what Amanah provides at 151 Front Street West.
CHALLENGE
WHAT AMANAH PROVIDES
Standard racks cannot support GPU power draw above 5kW
TOR2 supports up to 25kW per rack with engineered power distribution
Air cooling causes thermal throttling at full GPU load
In-row cooling + aisle containment + Enwave manages sustained thermal load
Cloud GPU cost is unpredictable at scale
Fixed monthly colocation cost regardless of utilisation
Training pipelines need high-throughput bandwidth
1Gbps to 100Gbps unmetered no caps on training or inference traffic
Hybrid workflows need private cloud connectivity
Cross-connects to AWS, Azure, Google Cloud from your rack
Canadian data residency required for regulated AI data
All infrastructure in Ontario, Canada SOC 2, PIPEDA, PHIPA
TOR2 infrastructure for AI
Three layers of cooling. 25kW per rack. 100Gbps.
TOR2 is the only suite at 151 Front Street West designed for sustained high-density compute workloads.
In-row cooling
Cooling units positioned in the row alongside your GPU racks. Heat managed at rack level before it spreads.
Hot/cold aisle containment
Cold supply air delivered to intake faces. Hot exhaust physically contained. No hot spots, maximum cooling efficiency.
Enwave deep lake cooling
Lake Ontario-sourced. High thermal capacity. Enables sustained 25kW per rack without mechanical chillers.
Up to 25kW 208V and three-phase
Circuits provisioned to your specs. 20A 110V standard. 208V, three-phase, and B feeds available on request.
100Gbps unmetered
No caps on training data, inference traffic, or checkpoint transfers. First provider at 151 Front to offer 100Gbps unmetered.
Private cloud connectivity
Cross-connects to AWS, Azure, Google Cloud for hybrid AI pipelines. No public internet for data movement.
Use Cases
How AI companies use Amanah.
Model training
LLM and foundation model training
Dense GPU clusters at full power for days or weeks. Fixed colocation cost vs unpredictable cloud GPU billing.
Inference
AI inference at scale
GPU inference endpoints at Canada's primary internet exchange. TORIX peering minimises latency to Canadian API users.
HPC
High-performance compute clusters
Scientific computing, genomics, and climate simulation at 10kW to 20kW per rack. Sustained load, no throttling.
Rendering
GPU rendering and visual compute
Dense GPU rendering farms with in-row cooling. 100Gbps for large asset transfers.
Research
University and research computing
PIPEDA compliant infrastructure in Ontario for research data. SOC 2 for institutional procurement.
MLOps
Hybrid MLOps pipelines
Train on colocated GPUs, store on cloud. Private cross-connects move datasets and checkpoints without public internet.
Certifications
Independently audited. Documentation on request.
SOC 2 Type II
All five AICPA Trust Services Criteria. Audit period, not point-in-time.
PIPEDA
Canada's federal privacy law. All data in Ontario, Canada. No cross-border transfers.
PHIPA
Ontario's Personal Health Information Protection Act. Suitable for regulated health workloads.