Lambda plans first-class Run:ai support on its Kubernetes Service

Lambda says it wants to make Run:ai a first-class option on Lambda Kubernetes Service (LKS), following NVIDIA's decision to open-source the GPU scheduler. The AI cloud company says it has supported Run:ai on its infrastructure since 2022.
- Lambda welcomes NVIDIA's decision to open-source Run:ai and plans to make it a first-class option on Lambda Kubernetes Service (LKS).
- Lambda says it has supported Run:ai on its infrastructure since 2022.
- Run:ai adds GPU sharing and quota-based scheduling to Kubernetes, according to Lambda.
- LKS already ships node auto-remediation that cordons, drains and revalidates faulty GPU nodes.
- Lambda is working with NVIDIA on a managed AI software library including Run:ai and Slurm; no date was given.
Why Lambda wants Run:ai on LKS
In a blog post dated October 13, 2026, Lambda said it welcomes NVIDIA's decision to open-source Run:ai and is working to bring it into LKS. According to Lambda, the goal is to make Run:ai's capabilities easier to adopt and operate, giving teams more control over how they allocate and share GPU capacity.
Lambda describes the problem as one of idle capacity. AI work places varied demands on GPUs: a debugging session may use a fraction of a GPU, an evaluation may launch many independent jobs, and distributed training may need many tightly connected GPUs at once. Allocating whole GPUs to small jobs, Lambda says, leaves researchers waiting while usable capacity sits reserved. The same friction appears between teams, when one group's experiments queue behind another group's idle allocation.
What Run:ai does
Run:ai is a GPU scheduler for Kubernetes. Lambda says it adds two things the default Kubernetes scheduler does not do:
- GPU sharing, so more than one workload can run on a single GPU.
- Quota-based scheduling, so teams get a guaranteed allocation but can borrow idle capacity from each other.
In practice, this comes down to GPU fractions and scheduling policies. Fractions let workloads request and share portions of GPU memory. The policies let projects borrow spare capacity for interruptible jobs and return it when other teams need their share. Lambda says the result is higher GPU utilization on a shared cluster without teams having to coordinate by hand.
Node auto-remediation keeps GPUs healthy
Lambda stresses that sharing and preemption assume the GPU underneath is healthy, which is not guaranteed at scale. The company lists accelerator XID errors, degraded NVLink and PCIe links, remapped memory rows and InfiniBand ports that drop below their expected rate. A scheduler cannot tell a healthy GPU from a failing one, so it may keep sending jobs to a node that drops them.
This is the part LKS owns, according to Lambda. LKS is managed Kubernetes on Lambda GPUs, and auto-remediation ships in it today. Lambda monitors every node for these faults and surfaces each as a standard Kubernetes node condition. When one fires, the node is cordoned so nothing new schedules to it and running work is drained. The fault is then either corrected automatically or escalated to hardware repair, and the node returns only after it revalidates.
An AI software library on the roadmap
Lambda says GPU scheduling is the layer it deliberately leaves open. LKS is headed toward an AI software library: schedulers that run on Kubernetes, such as Slurm and Run:ai, delivered as managed components that Lambda validates against its hardware and keeps current, rather than assembled from Helm charts by hand.
Run:ai is a first-class target for that library, and Lambda says it is working with NVIDIA to get there. The company argues that open sourcing makes this possible, because it lets Lambda package, test and maintain Run:ai as part of LKS instead of giving customers integration instructions. Lambda says it is aiming for an out-of-the-box integration, with the company owning the packaging, validation and ongoing operations.
What it means for customers
The post sets out intent rather than a release. It gives no date or availability details for Run:ai on LKS. What Lambda does describe is a direction: teams choose how to allocate and share their GPUs, while the platform keeps them healthy and available. Lambda, which says it produces open research, frames NVIDIA's move as extending the ability to inspect, adapt and improve tools to orchestration. Interested teams are pointed to Lambda's Kubernetes pages, its docs, or its machine learning engineers.
Lambda: Run:ai open source comes to Lambda Kubernetes Service


