From a high level, they are two tools that manage deploying ML models. However, there are very different:

- Triton is open-source, so it is free.

- With Triton, you can ask for features and even contribute to them.

- You can run Triton on your laptop for quick testing or development (if the network is down for example). In contrast, you can't with SageMaker.

- If your are selling a solution and using sage maker, then you can only offer a cloud solution. While Triton can be deployed on any in-house server. This is true for users with very sensitive data that can't use the cloud (unless of course, they have AWS on-prem solution)

- With Triton you are vendor locked. You can deploy Triton on premise, AWS, Azure, etc. So, if another vendor is offering a better product or price, then with Sage maker you need to convert everything to what ever the vendor's solution. While with Triton it is just the deployment config that needs to change, and if you are using Kubernetes, then it is just a URL and credential change.

That being said, if your team does not have a DevOps group, then I would go with Sage maker because with Triton, you will need to handle the deployment, monitoring, and maintenance of the infrastructure.