If you’re planning to use SageMaker in the ca-central-1 region, you might run into some limitations with the available instances. Specifically, the g6e and g7e instance types are not supported in this region at the moment. This is because these families are only available in select regions, such as ap-northeast-1, ap-northeast-2, eu-central-1, eu-north-1, eu-south-2, us-east-1, us-east-2, and us-west-2. Since ca-central-1 isn’t on this list, these options aren’t accessible there.
Additionally, the g7e family, which was announced in April 2026 in U.S. regions, has so far only been expanded to places like Seoul, London, and Tokyo, with no confirmation yet that it will arrive in ca-central-1. Because SageMaker instances generally follow the same regional availability as EC2, it’s safe to say that these newer GPUs aren’t available for real-time inference in ca-central-1 either. This means it’s too early to consider them as alternatives for running large language models that need four GPUs.
When it comes to other suitable options in ca-central-1, only a few GPU instance families are supported. These include g4ad, g4dn, g5, g5g, g6, g6f, and gr6. Of these, the g6f family is designed for fractional GPUs, meaning it provides parts of a GPU (like 1/8 or 1/2), which aren’t suitable for projects like large language model inference that require a full four-GPU setup. Currently, the best options for 4-GPU configurations in ca-central-1 are the g6.12xlarge and g5.12xlarge instances you already identified. Unfortunately, there aren’t any new or better options available right now that match these in price and performance.
It’s also good to know that smaller regions like ca-central-1 typically face tighter capacity limits on high-performance GPU instances such as g4, g5, and g6 families. Reports frequently show shortages in these regions, not only in ca-central-1 but also in eu-central-1, which suggests a common trend among smaller or more localized data centers. If your data residency requirements are flexible, consider using larger regions such as us-east-1 or us-west-2, where the availability is often better.
Recently, AWS introduced a new feature in May 2026 called “Capacity-aware inference.” This feature automatically helps you manage instance availability by following a prioritized list of instance types you specify. Whenever there’s capacity constraint during the setup or scaling of your SageMaker endpoints, the service will automatically fall back to the next available instance type in your list, making it easier to maintain operations without manual intervention. This feature is a significant help, especially in regions with limited capacity, ensuring your applications stay up and running smoothly.