Find self-service SLURM-as-a-Service options for AI workloads.

For “self-service Slurm-as-a-Service” for AI/ML workloads, I’d separate true managed offerings from cloud templates that still leave you operating Slurm.

• CoreWeave SUNK — strongest match for true self-service. SUNK runs Slurm on CoreWeave Kubernetes Service and now offers a guided self-service deployment model from the CoreWeave console. CoreWeave manages deployment workflows, upgrades, lifecycle controls, hardware management, and observability. It is explicitly positioned for large-scale distributed AI training. (docs.coreweave.com)

coreweave.com

• Lambda Managed Slurm + 1-Click Clusters — another strong match. Lambda provides managed Slurm on dedicated AI GPU clusters (including H100/B200 and newer systems), handling Slurm patches, monitoring, node failures, job history, and SchedMD-backed support. Importantly, GPU cluster reservations can be initiated self-service through its dashboard. (lambda.ai)

lambda.ai

• AWS SageMaker HyperPod with Slurm — managed AI-cluster infrastructure with native Slurm orchestration. HyperPod is specifically intended for model training/fine-tuning, and users can create clusters through AWS UI/API and submit normal Slurm workloads. It's more “managed infrastructure + Slurm” than fully outsourced Slurm administration. (docs.aws.amazon.com)

aws.amazon.com

• Azure CycleCloud Workspace for Slurm — highly automated self-service deployment rather than SaaS. The Marketplace solution creates/configures the infrastructure and Slurm cluster and includes Pyxis/Enroot for containerized AI jobs. Microsoft explicitly says it is not PaaS, so your team retains more operational responsibility. (learn.microsoft.com)

learn.microsoft.com

• Google Cloud Cluster Toolkit for Slurm — similarly, self-service provisioning rather than managed Slurm SaaS. It provides AI-optimized Slurm blueprints for GPU systems including A3/H100 and A4, with dynamic/autoscaling nodes available. (docs.cloud.google.com)

cloud.google.com

For a buyer specifically seeking “researchers use sbatch, vendor operates everything underneath,” I would shortlist CoreWeave SUNK and Lambda Managed Slurm first. AWS is the strongest hyperscaler alternative; Azure/GCP are better characterized as deployable Slurm platforms than Slurm-as-a-Service.