TABLE OF CONTENTS
NVIDIA H100 SXM On-Demand
Hyperstack Monthly Update
It’s time for our January monthly update.
A new month, a stronger Hyperstack. January was all about new features, updates and improvements that make your experience even better. Let’s take a quick look at what shipped and what’s coming up next.
New on Hyperstack
Check out what's new on Hyperstack this month:
Kubernetes CSI Driver RBAC Policy
We have added a new policy, KubernetesCSIDriver to grant users the required permissions for proper CSI driver functionality within Kubernetes clusters.
Contracted VMs Isolated from On-Demand Credit Enforcement
Contracted virtual machines no longer consume on-demand credits. They are exempt from auto-hibernation due to low balance and can be launched or restored even when the account balance is zero. Learn more about Contract Billing and Resource Behaviour here.
Firewall Warnings for Kubernetes VMs
A warning banner now appears in the Firewalls tab for any VM that’s part of a Kubernetes cluster. This helps users avoid firewall changes that could disrupt cluster operations or controller reconciliation.
Image–Flavour Compatibility Support
Hyperstack now detects compatibility between selected images and VM flavours to help ensure successful deployments. While no images currently define restrictions, this enables future warnings for suboptimal combinations and enforcement of incompatible configurations.
New on AI Studio
Check out what's new on AI Studio this month:
Third-Party Hosted Models
AI Studio now supports selected third-party hosted models for inference-only use. These models are available directly from the Playground. Check out the available third-party models here.
Persistent Playground Chat
Playground chat sessions are now persistent, retaining conversation history, selected models, prompts and parameter settings across page refreshes and logins.
Latest Fixes and Improvements
Check out all the things we fixed and improved this month on our platform:
- Kubernetes Ingress Capacity Upgrade: Increased the maximum HAProxy connections on Kubernetes ingress load balancer nodes from 250 to 2,000, significantly improving ingress concurrency and stability for high-load workloads.
- Cluster and Node Naming Updates: New cluster names are now limited to 20 characters. Node VM names no longer include the
kube-prefix and now use a shorter, clearer format based on cluster name, node role, and count. - Image Size Display Update: The
display_sizefield in the List Images API now uses IEC units (e.g., GiB) instead of SI units (GB) for more accurate byte conversion. - Improved Billing Clarity for Postpaid Accounts: The Billing Overview page now shows billing data exclusively for on-demand usage. Contract-related details are available in the Resource Activity and Contracts tabs.
- Insufficient Balance Policies Applied to Object Storage: Object storage access is now subject to the same account balance rules as other Hyperstack resources.
- Reliable VM Creation with Bootable Volumes via API: VM creation
create_bootable_volume = truenow waits for the volume to become available, improving provisioning reliability. - Firewall Management Fixes: Fixed issues with firewall assignment and detachment workflows. The VM list now correctly scopes to the selected environment, and firewall removal actions function as expected.
- Improved Redirect After Session Expiry: Users are now redirected back to their previous page after re-authentication, even if the access token expired during a page refresh.
- TLS SAN Mismatch Fix for Standard Kubernetes Deployments: The public IP of the master node (acting as both bastion and API endpoint) is now included in the Kubernetes API server certificate SANs, resolving TLS handshake failures.
New on the Blog
Check out exciting blogs on Hyperstack this month:
Best Object Storage Solutions for Cloud Data
What’s the Difference?
If you’re working with large volumes of cloud data, object storage is not an option anymore. No matter if you’re managing AI training datasets, media libraries or long-term archives, choosing the right object storage solution affects performance. In this blog, you’ll learn what object storage is, why you need it and which solutions stand out as the best object storage solutions for cloud data in 2026.
Learn more in our latest blog.
What is a GPU Cluster
How it Powers Modern AI Workloads
Modern AI workloads push computing infrastructure far beyond what a single server can handle. Training LLMs, running deep learning experiments or processing large datasets requires more parallelism and high-throughput computation. This is why enterprises and large-scale organisations choose GPU clusters for modern AI workloads. In this blog, you’ll learn what a GPU cluster is, how it works, why it’s essential for modern AI workloads and real-world use cases.
Learn more in our latest blog.
How to Run and Deploy Qwen3-Coder-Next:
A Complete Guide
Qwen3-Coder-Next is the latest open-weight language model from the Qwen Team, built specifically for coding agents and local development. Unlike traditional dense models, it utilises a Mixture-of-Experts (MoE) architecture. While it boasts a massive 80B total parameter count, it only activates 3B parameters per token. The tutorial walks through deploying Qwen3-Coder-Next on Hyperstack using GPU-powered virtual machines
Learn more in our latest blog.
Your Ideas Power Hyperstack
You know your workflow better than anyone. If there’s anything you wish Hyperstack did differently or better, now’s your chance to tell us.
Maybe it’s a feature you’ve been thinking about, a tool that could speed up your workflow, or a simple improvement that would make your project easier. Whatever it is, we’re listening.
For any questions or suggestions, feel free to reach out at support@hyperstack.cloud. Stay tuned for even more updates and exciting tools next month.
Subscribe to Hyperstack!
Enter your email to get updates to your inbox every week
