Lifestyle
AWS: SageMaker Studio Can Now Manage HyperPod Spaces Directly, No Command Line Needed for Everyday Tasks
AWS said on October 6, 2026 that data scientists can now create, start and stop Spaces (interactive coding environments) on SageMaker HyperPod clusters from the SageMaker Studio web interface instead of typing commands. Here is what changed, what admins must set up first, how long startup takes and what it costs.
About 7 min read
What AWS announced
According to an AWS blog post published on October 6, 2026, the company recently launched a new capability. Users can now create and manage SageMaker Spaces on SageMaker HyperPod EKS clusters directly from the Amazon SageMaker Studio interface. Spaces are interactive development environments that run directly on the cluster. Examples are JupyterLab, a browser-based notebook for writing and running code step by step, and Code Editor, a code editor that runs in the browser. AWS says data scientists can now create, configure, start, stop and open Spaces in Studio. A new "IDE and Notebooks" tab on the HyperPod cluster details page provides a full management interface.
AWS describes SageMaker HyperPod as infrastructure built for large-scale training and inference of foundation models, which are large general-purpose AI models. It is coordinated by Amazon EKS (Amazon Elastic Kubernetes Service). Kubernetes is software that schedules workloads across many machines. With it, HyperPod can run distributed training across hundreds of accelerators, with built-in resilience and automatic fault recovery. AWS also noted that earlier this year it launched the SageMaker Spaces for HyperPod add-on. That add-on lets interactive development share the same infrastructure as training jobs and model deployments. It also supports fractional GPU allocation, meaning only part of a GPU is assigned to a given environment.
- According to AWS, a guided form lets you configure compute, namespace, storage and image settings. The form also covers HyperPod Task Governance, which manages compute quotas.
- A searchable table shows each Space's name, application type, status, access type and storage. It also shows its GPU and vCPU (virtual processor) allocation.
- Spaces can be started and stopped to free up compute resources when they are not in use.
- JupyterLab or Code Editor can be opened in the browser. You can also connect through a remote IDE such as VS Code, a development tool installed on your own computer.
What it means for users
AWS says that creating and managing Spaces previously relied mainly on command-line tools such as the HyperPod CLI or kubectl. That approach gives infrastructure admins fine-grained control. According to AWS, data scientists who prefer a visual interface can now use this Studio feature to skip command-line tools and focus on model development. In its conclusion, AWS states that teams can now go from getting cluster access to a running JupyterLab or Code Editor environment in minutes, without learning CLI tools or Kubernetes concepts.
For teams sharing a HyperPod cluster, opening a development environment day to day may feel closer to using an ordinary website. AWS also explains that work is stored on a mounted Amazon EBS volume, a type of attached storage disk. This means stopping and restarting a Space does not lose progress. For remote connections, AWS says "Open in VS Code" uses an SSH-over-SSM tunnel under the hood. This is a secure connection routed through AWS Systems Manager. As a result, there is no need to manage SSH keys or open port 22.
Setup admins must complete first
This feature does not work out of the box. According to AWS, setup is split between two roles: admins prepare the cluster, and data scientists create and open Spaces. Admins must complete the following one-time setup.
- Install the SageMaker Spaces add-on: AWS says you can choose Quick install or Custom install. Custom install is required to enable web browser access.
- Configure EKS access entries, which control who can use the cluster: AWS says three managed policies must be attached to the IAM (AWS Identity and Access Management) roles used by data scientists. The three policies are AmazonSagemakerHyperpodSpacePolicy, AmazonSagemakerHyperpodUserClusterPolicy and AmazonSagemakerHyperpodSpaceTemplatePolicy.
- Enable per-user identity propagation: AWS says this setting must be enabled if the Studio domain was created before this integration launched. It ensures each user's actions on the cluster are attributed to their own user profile in EKS access entries and AWS CloudTrail. It also enforces Space ownership, meaning a record of who created each Space and whether it is private or shared.
AWS says updating the domain settings does not affect applications that are already running. The new settings apply the next time users sign in. AWS also lists several optional features: - Space templates with settings predefined by admins - Task Governance quotas and queues via Kueue - Karpenter autoscaling, which adds and removes nodes (servers) based on Space demand - Idle auto-shutdown - NVIDIA MIG fractional GPU allocation for A100/H100 hardware - EFS/FSx persistent volumes - Custom images hosted in Amazon ECR
Startup time and costs
Some clusters use Karpenter autoscaling and have scaled down to zero, meaning no standby servers are running. AWS says that on such a cluster, creating a first Space incurs a default cold-start delay of 5–7 minutes. That is the wait while everything starts from scratch. Most of it comes from launching the Amazon EC2 instance, registering the Kubernetes node and pulling the SageMaker Distribution image. AWS describes an overprovisioning approach that keeps a set of nodes powered on with images already downloaded. A low-priority placeholder Deployment puts a "placeholder Pod" on each warm node. When a user creates a Space, the scheduler lets the Space take the placeholder Pod's place, which is called preempting it. AWS says this cuts startup time to about 30–40 seconds.
| Startup path | Latency reported by AWS |
|---|---|
| Space co-located with placeholder Pod on the same node | About 14 seconds |
| Space preempts placeholder Pod | About 35 seconds |
| Cold start (no warm pool) | 5–7 minutes |
On costs, AWS says setting up the SageMaker Spaces add-on incurs no additional charge. Users do pay for the HyperPod cluster compute consumed by their Spaces. They also pay an hourly fee for the AWS Systems Manager Advanced On-Premises Instance used for SSH-over-SSM remote connections.
FAQ
How is this different from before?
AWS says creating and managing Spaces previously relied mainly on HyperPod CLI or kubectl commands. Now data scientists can create, configure, start, stop and open Spaces directly from the "IDE and Notebooks" tab in SageMaker Studio.
Can data scientists use it right away?
Not necessarily. According to AWS, admins must first install the SageMaker Spaces add-on on the HyperPod EKS cluster. They must also attach three managed policies to the relevant IAM roles. If the Studio domain was created before this integration launched, per-user identity propagation must also be enabled.
How long does it take to start a Space?
AWS says a cold start on a cluster scaled to zero takes about 5–7 minutes. With overprovisioning and warm nodes, it takes about 30–40 seconds. AWS notes these figures apply to CPU only, and GPU Spaces require separate setup.
Does using this feature cost extra?
AWS says setting up the add-on itself is free. You do pay for the HyperPod cluster compute your Spaces consume. You also pay the hourly AWS Systems Manager Advanced On-Premises Instance fee for SSH-over-SSM remote connections. If you use overprovisioning, warm nodes add further cost.
Where does this information come from?
Every detail in this article comes from AWS's official blog post. All figures, including startup latency, were published by AWS itself.
Browse the latest news in this topic
Lifestyle
AI ad generation platform Melius raises $25M: ex-Ramp engineers pivot after scrapping their first product
According to TechCrunch, New York AI startup Melius has announced $25 million in total funding, including a $20 million Series A led by CRV and a $5 million seed round led by General Catalyst. Three former Ramp engineers abandoned their original tool for managing ad spend to build a platform that generates ad campaigns, images and videos from everyday language, joining Higgsfield and others in the race for the AI ad creative market.
Lifestyle
LibreOffice says it will not add AI for the foreseeable future, to keep users' documents on their own computers
According to TechCrunch, The Document Foundation, the nonprofit behind the open-source document editor LibreOffice, says it will not add AI to the software for the foreseeable future, so that documents are not uploaded to a server for processing. This matters to people and organizations that care about data privacy. Anyone who wants AI can still install extensions that connect to local AI models.
Lifestyle
Technology Innovation Institute releases Falcon-Emirati-7B: an Arabic AI model focused on the Emirati dialect that, by its own tests, answers in dialect more often than rivals
Technology Innovation Institute (TII) released Falcon-Emirati-7B on October 6, 2026, an AI language model built to understand and use the spoken dialect of the UAE (United Arab Emirates). TII says other Arabic models often get the content right but reply in written Standard Arabic. This article summarizes the training method, evaluation figures and usage limits TII published; all data are TII's own claims.
Lifestyle
NVIDIA launches DGX Spark 64GB: on sale October 23 from $4,999, two units can be linked into 128GB
On October 2, 2026, NVIDIA announced a more affordable 64GB memory version of its DGX Spark personal AI computer, available from October 23 through six makers including Acer and ASUS. It is aimed mainly at developers and researchers who want to run AI models on their own machines. Below we summarize the specs NVIDIA published, its claims about linking two units, and what it means for general readers.