TurboML
TurboML · Montréal, QC
- Canadian owned
- Hosted in Canada
TurboML operates its own GPU facility in Montréal, renting bare-metal GPU servers by the hour and hosting open-source models by the token. It bills in Canadian dollars by default and states that customer workloads, prompts, outputs and stored files stay in Canada unless a customer instructs otherwise in writing. For Canadian teams currently renting H100s from an American hyperscaler, it is the closest thing in this directory to a like-for-like swap that keeps both the data and the invoice inside the country.
What they offer
Bare-metal GPU servers
Dedicated nodes of 8 GPUs — RTX PRO 6000 Blackwell available now, H100 and H200 procurement-backed.
Hosted model inference
27 open-source models across LLM, embedding, image and vision, billed per million tokens.
Reserved capacity
Discounted 1-month and 12-month terms, quoted against current GPU inventory.
Canadian data residency
Customer content is not replicated or backed up outside Canada unless instructed in writing or legally required.
What it replaces
Where the fit is imperfect, we say so. A substitution that quietly drops a capability you depend on isn’t a saving.
Amazon EC2 GPU instances
Amazon Web Services
The same class of hardware rented by the hour, from a Québec facility and invoiced in CAD. AWS has Canadian regions, so the difference here is corporate jurisdiction, currency and egress terms rather than data location alone. AWS wins on breadth — if your workload leans on the surrounding services, you are not swapping one rental for another.
All Canadian alternatives to Amazon EC2 GPU instancesMicrosoft Azure
Microsoft
Comparable for straightforward GPU training and inference. Azure carries the wider platform, compliance catalogue and enterprise agreements; TurboML is a single operator with one facility, which is the trade in both directions.
All Canadian alternatives to Microsoft AzureGoogle Cloud
Google
Overlaps on rented accelerators and hosted open models. Google Cloud offers managed training pipelines and TPUs that have no equivalent here.
All Canadian alternatives to Google CloudTogether AI
Together AI
Both host open-source models behind an API billed per token. TurboML's catalogue is smaller; the difference is where inference runs and under whose law.
All Canadian alternatives to Together AI
Cost comparison
From published list pricePricing model: Per GPU·hour for compute; per million tokens for hosted models; billed in CAD
Prices compared: September 22, 2026
TurboML · Canadian
Amazon Web Services · American
TurboML states on its own pricing page that public rates are being finalised, so treat these as current list rates rather than settled ones. H100 is CA$3.95 per GPU·hour on demand and CA$2.95 reserved, and is procurement-backed rather than in stock; RTX PRO 6000 Blackwell is available now at CA$2.50. H200 is quoted on request. There is no egress charge within Canada, and cross-border egress is billed at transit rates disclosed before data leaves the region. No American comparator is published here: hyperscaler H100 capacity is sold in 8-GPU instances with committed-use discounts, and dividing that into a per-GPU hourly figure would produce a number that looks precise and isn't.
Questions
- Is TurboML a Canadian company?
- TurboML is listed as Canadian owned and Canadian hosted, based in Montréal, QC. This profile is compiled from public product information.
- Can TurboML replace Amazon EC2 GPU instances?
- The same class of hardware rented by the hour, from a Québec facility and invoiced in CAD. AWS has Canadian regions, so the difference here is corporate jurisdiction, currency and egress terms rather than data location alone. AWS wins on breadth — if your workload leans on the surrounding services, you are not swapping one rental for another.
- What does TurboML cost?
- We don't publish pricing for TurboML yet. A figure appears here only with the page it came from and the date we read it, so this profile shows pricing as not yet verified rather than as an estimate.