dstack review
Broad GPU orchestration for teams spanning cloud, Kubernetes, Slurm, and on-premises systems.
Reviewed by iTechGuides Editors · Editorial team · Updated Oct 2026
dstack is an open-source orchestration control plane for teams provisioning and managing GPU infrastructure across GPU clouds, Kubernetes clusters, and on-premises systems. It supports interactive development environments, batch and distributed training jobs, inference services, fleets, volumes, and gateways. Workloads can be defined in YAML and applied through a CLI, HTTP API, or web interface. Support for NVIDIA, AMD, TPU, and Tenstorrent accelerators broadens its fit for teams managing more than one accelerator type. Teams that need one management layer across varied GPU environments are the clearest audience.
Its provider and backend coverage is a central strength. dstack connects AWS, Azure, Google Cloud, Lambda Cloud, Nebius, Crusoe, OCI, Runpod, and Vast.ai, while also supporting Kubernetes, Slurm, and SSH backends. That combination suits organizations combining public clouds with existing clusters or on-premises systems. GPU utilization and DCGM metrics provide operational visibility, while auto-scaling, retry, and spot-instance policies address workload placement and interruption handling. Projects, users, roles, and resource isolation add organizational controls, and volumes, gateways, secrets, and API access extend the platform beyond basic job submission.
The published product model separates the open-source and hosted options. dstack OSS is self-hosted and free, while dstack Sky is a hosted service priced through GPU marketplace usage rather than fixed monthly software plans. The platform supports both interactive and batch scheduling and offers hybrid deployment, but quota controls are not included. That makes dstack a strong fit for teams that value broad integrations, flexible deployment, and control over GPU workloads across multiple environments. Teams that require built-in quota controls or prefer a fixed software subscription should consider whether its operating model matches their requirements.
dstack pros and cons
- Where it wins
- Supports interactive, batch, distributed, and inference workloads
- Connects AWS, Azure, Google Cloud, Kubernetes, Slurm, and more
- Includes autoscaling, GPU metrics, isolation, and API access
- Where it doesn't
- Quota controls are not included
- The open-source edition is self-hosted
- dstack Sky uses GPU marketplace usage pricing, not fixed software plans
dstack fact sheet, pricing and score →
Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.
Last updated · How we research and update