dstack
An open-source orchestration layer that provisions GPU fleets and runs training jobs, dev environments and inference services on GPU clouds, Kubernetes, VMs or bare-metal servers.
About dstack
Teams using GPUs from several clouds or their own servers have to build and maintain separate provisioning, scheduling and serving setups for each environment.
Who it’s for
- ML teams running training and inference across GPU clouds and on-prem servers
- Data-centre operators and GPU cloud tenants who need a shared control plane with tenant isolation
When to consider it
Consider dstack when you want a vendor-neutral, self-hosted alternative to Kubernetes or Slurm for provisioning GPUs and running AI jobs and services.
Tradeoffs & limitations
- The open-source server is MPL-2.0 licensed and self-hosted; it provisions fleets in your own cloud accounts using credentials you configure, so compute is billed by those providers.
- The Slurm backend is marked experimental. A hosted version, dstack Sky, offers access to GPU clouds with one billing, but its prices are not listed on the public site.
SOTA overview · Documentation-based assessment · Sources & review method
Updates
No updates shared yet.
Discussion
Newest firstAsk a question or share how you use dstack.
Keep it helpful. Community rules
Loading discussion…
Sources & review method
Documentation-based assessment · Nov 1, 2026 · Prepared with AI assistance; not a hands-on benchmark.
Checked by SOTA · AI-assisted documentation review. Selection advice is our assessment; verify current requirements for your deployment.
Import history & original evidence
dstack official website
An open-source orchestration layer that provisions GPU fleets and runs training jobs, dev environments and inference services on GPU clouds, Kubernetes, VMs or bare-metal servers.
Based on official pages and announcements checked on 2026-11-01. No hands-on test, performance benchmark or popularity ranking is claimed.