On this page
- Which GPUs can you rent in Canada?
- How much does a GPU cost per hour in Canada?
- Can you get an L40S or RTX PRO 6000 in Canada?
- Does AWS Calgary (ca-west-1) have GPUs?
- How do you get GPU quota in Canada?
- Which GPU fits which model size?
- What this read does not tell you
- Can you keep AI inference in Canada on rented GPUs?
- FAQ
- Sources
Most NVIDIA data-centre GPUs are listed in Canada, but the big three clouds don't rent the L40S or the RTX PRO 6000 there. On 11 November 2026, AWS Montreal offered the L4, A100, H100 and H200. Azure in Toronto offered the A100 and H100, and Google the L4 and H100. H100 and H200 capacity usually needs a one- to three-year commitment.
Key facts, read on 11 November 2026
- Listed in Canada by AWS, Azure or Google: L4, A10G, A100, H100, H200. The H100 and H200 usually come only with a one- to three-year commitment.
- Not rentable in Canada from any of the three: L40S, RTX PRO 6000.
- Price premium: AWS Montreal runs 11% (small cards) to 15% (eight-GPU servers) above US East. Azure Canada Central runs 20 to 23% above East US.
- AWS Calgary: its price list carries no GPU instances at all.
- Quota: a new AWS account starts at zero GPU quota in Montreal. You ask for it before you can launch anything.
- Canadian-owned options we could verify: ISAIC in Alberta rents the L40S, A100 and H100. TELUS sells reserved H200 capacity.
We got these numbers from the providers themselves. We queried AWS's instance and pricing APIs and Azure's size and price APIs for their Canadian regions. For Google, we read its GPU locations page. A catalogue listing says a GPU exists in a region. It doesn't promise you'll get one today.
Which GPUs can you rent in Canada?
The L4, A100 and H100 are rentable in Canada from at least one of the three big clouds, and the H200 from AWS. Here is each card against each Canadian region. GPU memory, the memory on the card that holds the model while it runs, is the first number to check, because a model that doesn't fit doesn't run.
| GPU (GPU memory) | AWS Montreal (ca-central-1) | Azure Canada Central (Toronto) | Google Montreal / Toronto |
|---|---|---|---|
| L4 (24 GB) | Yes, g6 | No | Yes, G2, both cities |
| A10G / A10 (24 GB) | Yes, g5 | Yes, NVads A10 v5 | No |
| L40S (48 GB) | No | No | No |
| RTX PRO 6000 Blackwell (96 GB) | No | No | No |
| A100 | Yes, p4d (8 × 40 GB) | Yes, NC A100 v4 (80 GB) | No |
| H100 | Yes, p5 (8 × 80 GB) | Yes, NC H100 v5 (94 GB) and ND H100 v5 (8 × 80 GB) | Yes, A3 Edge, Toronto only |
| H200 (141 GB) | Yes, p5en (8 GPUs) | No | No |
The AWS and Azure columns come from their own APIs, read on 11 November 2026. The Google column comes from Google's GPU regions and zones page, read the same day. Card memory comes from each provider's own size pages (AWS G6e, AWS G7e, Azure NC A100 v4, Azure NCads H100 v5).
Two details matter when you read the table. On AWS Montreal, the A100, H100 and H200 come only as whole eight-GPU servers. AWS sells a one-H100 size, the p5.4xlarge ($6.88 an hour in US East), but doesn't offer it in Montreal. That makes Azure the one place in Canada to rent a single A100 or H100. Microsoft's A100 size page adds a caveat. It is "only deploying net new capacity" for the H100 NVL sizes, so A100 capacity in Canada Central won't grow.
How much does a GPU cost per hour in Canada?
A single L4 costs $0.89 an hour in AWS Montreal, and a single A100 costs $4.41 an hour on Azure Canada Central. An eight-H100 server costs $63.30 an hour on AWS. All prices are US dollars, on-demand Linux, with no discounts, read from AWS's price list and Azure's retail price API on 11 November 2026.
| Instance | GPUs | Canada | US East | Premium |
|---|---|---|---|---|
| AWS g6.xlarge | 1 × L4 | $0.894/h | $0.805/h | 11% |
| AWS g5.xlarge | 1 × A10G | $1.117/h | $1.006/h | 11% |
| AWS p4d.24xlarge | 8 × A100 40 GB | $25.25/h | $21.96/h | 15% |
| AWS p5.48xlarge | 8 × H100 | $63.30/h | $55.04/h | 15% |
| AWS p5en.48xlarge | 8 × H200 | $72.79/h | $63.30/h | 15% |
| Azure NC24ads A100 v4 | 1 × A100 80 GB | $4.41/h | $3.67/h | 20% |
| Azure NC40ads H100 v5 | 1 × H100 NVL 94 GB | $8.59/h | $6.98/h | 23% |
| Azure ND96isr H100 v5 | 8 × H100 | $117.98/h | $98.32/h | 20% |
Per card, AWS's eight-GPU servers work out to about $3.16 an hour per A100, $7.91 per H100 and $9.10 per H200 in Montreal. Google's G2 and A3 Edge prices for Toronto weren't part of this read; check Google's GPU pricing page for them.
The premium is the price of residency, and it's a known number. A team that budgets from a US price calculator will come up 11 to 23% short in Canada. Commitments lower the bill. Our hosting-model comparison shows a one-year AWS savings plan cutting a four-L40S server by about 37% in US East.
Can you get an L40S or RTX PRO 6000 in Canada?
Not from AWS, Azure or Google, as of 11 November 2026. AWS sells the L40S as g6e and the RTX PRO 6000 Blackwell as g7e, and neither family is offered in Montreal or priced in Calgary. Azure and Google have no size with either card in their Canadian regions.
We used both missing cards in our own work. Our decision-model benchmark ran on one RTX PRO 6000, an AWS g7e.2xlarge at $3.36 an hour in US East. Our decision-model training run used one L40S for 4.71 hours. Neither run could have happened in a Canadian region of AWS, Azure or Google.
The nearest cards that are rentable in Canada:
| If you wanted | Nearest in Canada | GPU memory | Price, 11 Nov 2026 |
|---|---|---|---|
| L40S, 48 GB (US East $1.86/h on g6e.xlarge) | L4 on AWS Montreal | 24 GB, half | $0.89/h |
| L40S, 48 GB | A100 on Azure Canada Central | 80 GB | $4.41/h |
| RTX PRO 6000, 96 GB (US East $3.36/h on g7e.2xlarge) | H100 NVL on Azure Canada Central | 94 GB | $8.59/h |
So a single 96 GB card in Canada costs about 2.6 times what the RTX PRO 6000 costs in Virginia. The H100 NVL is a faster card with faster memory, so the gap per answer will be smaller than the gap per hour, but you pay for the hour. For a deeper side-by-side, see L40S vs A100 and RTX PRO 6000 vs H100.
Canadian-owned GPU clouds we could verify
We checked the providers' own pages on 11 November 2026. Three have something concrete:
- ISAIC (Edmonton). A University of Alberta initiative that calls itself "Canada's national secure AI development environment". Its pricing page lists a dedicated one-L40S virtual machine at $858 CAD a month. The A100 and H100 versions cost $1,207 and $2,093 a month. Hourly rental is on a waiting list. This is the one Canada-hosted L40S we could confirm on a provider's own page.
- TELUS Sovereign AI Factory. TELUS's 4 February 2026 release describes NVIDIA H200 GPUs, sold as reserved capacity and "keeping everything under Canadian control and jurisdiction". It publishes no hourly price.
- Denvr Dataworks (Calgary). Its data-centre docs list a Canada West (Calgary) region with A100 (40 and 80 GB), GH200 and A40 servers. Its H100s sit in Houston, not Calgary.
Bell AI Fabric, with BUZZ HPC, describes Ampere, Hopper and Blackwell clusters in Canadian-owned facilities for Bell's government and enterprise customers, without naming card models or prices. We found no Canadian provider page offering the RTX PRO 6000. If you need that card in Canada today, the dependable route is buying it and running it in your own office or a Canadian colocation site.
Does AWS Calgary (ca-west-1) have GPUs?
No GPU instances appear in AWS's price list for Calgary, read on 11 November 2026. The list covers 35 instance families in ca-west-1. All are general-purpose, compute, memory or storage types, such as c7i, m8i and r8g. None is from the G or P GPU families.
We couldn't query Calgary's live offerings directly. ca-west-1 is an opt-in region, and it isn't enabled on our account, so we read its price list instead. A family with no price can't be launched on demand, so for GPU work AWS Canada means Montreal. Our residency guide covers what Calgary is still useful for.
How do you get GPU quota in Canada?
You ask for it, and you should ask before you need it. AWS has two GPU quotas: one for its G instances (L4, A10G) and one for its P instances (A100, H100, H200). Its Service Quotas API, read on 11 November 2026, gives both a default of zero in Montreal. Azure's GPU families in Canada Central also start at zero.
AWS counts quota in vCPUs, not GPUs, so size the request to the instance:
- Pick the instance first. A g6.xlarge needs 4 vCPUs of G quota. A p5.48xlarge needs 192 vCPUs of P quota.
- In the AWS console, open Service Quotas, choose EC2, select the Canada (Central) region and request the G or P quota.
- Say what the workload is and how long it runs. A 192-vCPU P request is a bigger ask than a 4-vCPU G request, so allow time.
- On Azure, request the vCPU quota for the specific family (for example NCads H100 v5) in Canada Central from the subscription's Quotas page.
Quota is permission, not capacity. An approved request still fails to launch if the zone has no free cards that hour. For a steady production workload, an AWS capacity reservation or an Azure reservation is the way to hold the cards.
Listed is not the same as available. In our experience deploying for Canadian clients, H100 and H200 capacity in Canadian regions is hard to get on demand, and the providers usually ask for a one- to three-year commitment (AWS Savings Plans and Azure reservations both run one or three years) before they allocate it.
Which GPU fits which model size?
Start from GPU memory. At 8-bit precision a model's weights take about 1 GB per billion parameters, and the model needs working room on top for the text it is reading. A small model fits one 24 GB L4. A 27-billion-parameter model fits one 80 GB A100 or 94 GB H100 NVL. Frontier-class open models need a whole eight-GPU server.
Our own runs give two reference points. Perplexity's card for its 27B decider asks for about 52 GiB of free GPU memory for weights alone, which we covered in our pplx-decider post: in Canada that means an Azure A100 80 GB or H100 NVL, not an L4. Our 4B decision model trained on one 48 GB L40S, a job an A100 can also do.
The largest open-weight models, like Kimi K3.1 Pro, need an eight-H100 or eight-H200 server, which AWS Montreal has. The full per-model arithmetic is in how much GPU memory you need to run open models, and the speed side in H100 vs A100 for LLM inference.
What this read does not tell you
- Listings are not stock. We read catalogues and price lists. We didn't launch instances, so we can't say how often a p5 is actually free in Montreal.
- Google's prices for Toronto weren't read for this post.
- Calgary was read from AWS's price list, not its live offerings API.
- Azure's price list and SKU list differ slightly. The price list for Canada Central also carries older V100 and AMD MI300X sizes. We kept to the NVIDIA cards in the SKU list.
- Specialist Canadian providers change fast. Smaller hosts and GPU marketplaces advertise L40S servers in Toronto and Montreal through aggregator sites. We list only those we could check on the provider's own page.
Can you keep AI inference in Canada on rented GPUs?
Yes. Rent the GPU in a Canadian region of your own cloud account, run an open-weight model on it, and every prompt and answer is processed on hardware in Canada. That makes the GPU choice the residency choice. An H100 or A100 in Montreal or Toronto keeps inference in Canada today. An L40S or RTX PRO 6000 does so only on hardware you own or from a specialist provider like ISAIC.
The trade-offs are real and manageable. You pay 11 to 23% more than in the US, you pay for the card while it idles, you request quota before launch, and you own the serving stack. Our comparison of on-premise, client cloud and managed cloud sets out who carries each of those costs.
XY Space deploys open-weight models on dedicated GPUs in the client's own Canadian cloud account or on-premises, sizes the hardware to the model, and builds the workflow on top. Your team runs the work and the model takes part. See how we deploy and the Canadian residency guide for the provider-by-provider detail.
If you need a model running with inference kept in Canada, bring one job to a free 30-minute call.
FAQ
Can I rent an H100 in Canada?
Yes. AWS Montreal offers the H100 only as an eight-card server, $63.30 an hour on demand on 11 November 2026. Azure Canada Central rents a single H100 NVL (94 GB) for $8.59 an hour, and an eight-card server too. Google lists H100 (A3 Edge) in one Toronto zone. All three start you at zero GPU quota, so request it first. In our experience the H100 is hard to get on demand in Canada, and providers usually ask for a one- to three-year commitment.
Is GPU cloud more expensive in Canada?
Yes, by a predictable margin. On 11 November 2026, AWS Montreal ran 11% above US East for its single-GPU instances. Its eight-GPU servers ran 15% above. Azure Canada Central ran 20 to 23% above East US for its A100 and H100 sizes. Budget from the Canadian price list, not the US one.
Where can I rent an L40S in Canada?
Not from AWS, Azure or Google, which don't list the L40S in any Canadian region as of 11 November 2026. ISAIC, a University of Alberta initiative in Edmonton, lists a dedicated L40S VM at $858 CAD a month on its own pricing page. The other routes are a different card (an A100 on Azure Canada Central) or buying L40S hardware and hosting it in Canada.
Does AWS have GPUs in Calgary?
Not on demand. AWS's price list for Canada West (Calgary), read on 11 November 2026, covers 35 instance families and none of them are GPU families. For GPU instances on AWS in Canada, use Montreal (ca-central-1), which offers the L4, A10G, A100, H100 and H200.
Does renting a GPU in Canada keep my data in Canada?
Renting in Canada keeps processing in Canada if the GPU, the storage and the logs all sit in a Canadian region of your account and nothing calls an outside API. Residency is a property of the whole system, not of one server. Our residency guide covers the logging, abuse-monitoring and legal questions that sit on top of the hardware.
Sources
- AWS, Savings Plans; Microsoft, Azure reservations: one- and three-year commitment terms
- AWS, EC2
describe-instance-type-offeringsand Price List API for ca-central-1, ca-west-1 and us-east-1, and Service Quotas default values for ca-central-1, read 11 November 2026; public pages: On-Demand pricing, G6, G6e, G7e, P4, P5 - Microsoft Azure,
az vm list-skusfor canadacentral and the Azure Retail Prices API, read 11 November 2026; NC A100 v4 sizes (updated 25 September 2026), NCads H100 v5 sizes (updated 28 July 2026) - Google Cloud, GPU regions and zones, read 11 November 2026
- ISAIC, pricing, read 11 November 2026
- TELUS, TELUS opens Canada's first fully sovereign AI Factory to startups and small businesses, 4 February 2026
- Denvr Dataworks, Data centers, read 11 November 2026
- Bell, Bell AI Fabric adds advanced NVIDIA GPU infrastructure with BUZZ HPC, 19 August 2025
Written by
Cho Yin Yong
Principal AI Solutions Engineer, XY Space
Principal AI Solutions Engineer at XY Space. University of Toronto lecturer for five years, co-author of two patents, winner of two competitive AI awards, and nine years of regulated engineering leadership.
More from Cho Yin YongShare this article
Work with us
We build the systems these posts describe, and we'll tell you in the first call whether yours is worth building.
Start a project