Amazon SageMaker’s biggest strength is that it delivers a fully managed, end-to-end machine learning (ML) environment within AWS’s secure ecosystem. It covers the whole workflow, from data preparation and model training to deployment and ongoing monitoring. For teams, this can save a huge amount of time by abstracting away infrastructure complexity, while still providing powerful distributed training capabilities, including features like HyperPod. Review collected by and hosted on G2.com.

Opaque pricing. Costs can escalate quickly, especially with long-running jobs or larger deployments. I’ve also seen “month-end shock” because the billing model isn’t clear or predictable. Review collected by and hosted on G2.com.

Being a Senior Data Scientist in a medium sized company, we wanted an approach that would help us get notebook-based models into production without building too much infrastructure. I love the amazingly wide-ranging ecosystem, it's an end to end platform. We use it a lot across our predictive customer behavior models and NLP pipelines and smoothly jumping between pushing a test job on SageMaker Studio and being able to start heavy distributed training jobs is awesome. No longer do I need to rely on our over bounded dev-ops team to provision certain GPU instances for me. I can just specify the amount of hardware in my code and AWS will spin them up and teard them down for me. The managed endpoints during the deployment process are also a huge timesaver, meaning we'll be able to generate our model predictions through a strong API that supports auto-scaling out of the box. Review collected by and hosted on G2.com.

The number one obstacle is its starting complexity. SageMaker is a tool that requires some configuration, and the documentation, although comprehensive, can be confusing and could be viewed as a disjointed set of tutorials. There's lots of time involved to get to the "AWS way" of doing stuff, and it's a frequent pitfall for team members coming into AWS for the first time—configuring IAM roles and VPCs and permissions. Secondly, it can be quite harsh on your wallet if you're not penny-pinching. Users can easily forget to turn off their SageMaker Studio instance or an experimental endpoint over the weekend, ending up with a large bill. Last but not least, the mesh of features of the Studio interface sometimes comes at the expense of somewhat slow and cumbersome performance in comparison to a lightweight, on-node, local Jupyter server leverage my own machine. Review collected by and hosted on G2.com.

I like that Amazon SageMaker gives you one managed path from experimentation to deployment without forcing you to assemble every piece yourself.

It covers the whole ML workflow: notebooks, training jobs, tuning, pipelines, model registry, endpoints, batch inference.

It reduces infrastructure work: you can focus more on models and less on provisioning GPUs, containers, scaling, and endpoint ops.

It works well for teams: reproducible training jobs, managed pipelines, and deployment/versioning help when multiple people touch the same system.

It scales from simple to serious: you can start with a notebook and later move to distributed training or production endpoints in the same ecosystem.

It integrates with AWS well: IAM, S3, CloudWatch, ECR, Lambda, and EventBridge make it easier if the rest of your stack is already on AWS.

If I had to pick one thing: the biggest advantage is the operational glue—it makes moving from “model works in a notebook” to “model runs reliably in production” much less painful. Review collected by and hosted on G2.com.

The biggest downside is complexity: SageMaker is powerful, but it often feels like a toolbox of AWS services rather than one clean, opinionated ML platform.

The learning curve is steep; you end up needing to understand SageMaker itself plus IAM, S3, VPCs, ECR, CloudWatch, and AWS networking.

Costs can get slippery; notebook instances, endpoints, training jobs, storage, and data transfer can keep running unless you manage them carefully.

The UX can feel fragmented; some tasks are easier in the SDK, some in Studio, some in raw AWS configuration.

Debugging can be frustrating; failures are often caused by permissions, container setup, networking, or obscure configuration mismatches rather than model code.

Vendor lock-in is real; once your pipelines, deployment flow, and monitoring are built around SageMaker/AWS primitives, moving away takes work.

The abstractions can be leaky; “managed” does not always mean simple, and you still may need to think like an infra engineer.

If I had to sum it up: SageMaker is very capable, but it’s not especially elegant. It rewards teams that already operate comfortably in AWS, and can feel heavy for smaller teams or faster-moving experimentation. Review collected by and hosted on G2.com.

Working locally is great—until your dataset outgrows your RAM, or you realize you need a multi-GPU cluster to train a model in hours rather than days. SageMaker addresses that exact friction point, and it does so brilliantly. Review collected by and hosted on G2.com.

I end up spending a massive amount of time digging through CloudWatch logs just to discover that a library version was mismatched or that an S3 file path was slightly off. It really drags out the debugging process and drastically slows down the inner loop. Review collected by and hosted on G2.com.

Best is how it helps in end to end working -model building, scaling, monitoring so that everything is done in one tool. Review collected by and hosted on G2.com.

Everything is perfect, just needs to be informed to people more about it Review collected by and hosted on G2.com.

AI training without any bottlenecks or dependencies on the underlying infrastructure. Review collected by and hosted on G2.com.

The UI should be more polished and intuitive, easy to use model registry Review collected by and hosted on G2.com.

I use Amazon SageMaker for building a deep learning model, specifically an object detection model. It's a really great experience for me, especially because my laptop doesn't have advanced GPU support, and training a model would take around 7-8 hours. With Amazon SageMaker's virtual machine, training my deep learning model only takes 3-4 minutes. This platform is great, and even someone who has never used it before can adapt to it the first time and easily understand all the functionality given on SageMaker. I think the virtual machine of Amazon SageMaker is more advanced than the Microsoft Azure platform. It is more effective and less time-consuming. The ease of use is brilliant; I can easily adapt to this platform compared to Microsoft. The initial setup is very easy, and with single authentication, I have access to the resources I need for my work. In my view, I give it 10 out of 10. Review collected by and hosted on G2.com.

This is great platform. I don't dislike this. Review collected by and hosted on G2.com.

It provides a complete platform to build, train, and deploy ML models. It also makes the whole process easier and faster. Review collected by and hosted on G2.com.

I don’t see any real disadvantages, but it does take a bit of initial training for developers to get comfortable using the platform. Review collected by and hosted on G2.com.

I like how easy it is to train ML models on Amazon SageMaker and conduct fast experiments. I can easily prototype and make changes to my ML models, and the training process is straightforward. All the logs are accessible, which helps in checking the training status and testing models. This makes experimenting and changing parameters directly in SageMaker efficient. Review collected by and hosted on G2.com.

Better cost transparency can be there. also, there is a learning curve with initial setup. Review collected by and hosted on G2.com.

The login process is straightforward, and setting up the software is not complicated. The user interface is also very user-friendly. Review collected by and hosted on G2.com.

The portal could use some additional finishing touches to appear more presentable. Review collected by and hosted on G2.com.