AI Potluck
Back to Gap Map Infrastructure / ML orchestration

dstack

dstack
open source / Overall score: 3.2

Control plane for GPU provisioning and orchestration across any GPU cloud, Kubernetes cluster or on-premise fleet. A run declares the resources it needs - including NVIDIA, AMD, Google TPU or Tenstorrent accelerators - and dstack provisions a fleet to satisfy it, places development environments, training tasks and inference services onto it, and tears it down after.

dstack Sky is a hosted offering beside the open control plane. Verified 2026-09-15 via the repository, its recursive tree and the PyPI download API.

Openness

5 high confidence
5.0
license
MPL-2.0(OSI)
source
public(dstackai/dstack is the control plane you run)
core-gated
ungated(a 2,640-entry recursive tree carries a single LICENSE.md and no path matching enterprise, ee, commercial or proprietary)

Mozilla Public License 2.0, read from the LICENSE.md body, and MPL-2.0 sits in the OSI tier beside Apache and MIT. The 2,640-entry recursive tree carries that single license file and no enterprise, ee or commercial path, and dstack's own deployment guide names the self-hosted control plane and the commercial offerings as separate products.

  • https://github.com/dstackai/dstack recorded 2026-09-15

    Repository page for dstackai/dstack: MPL-2.0 license, public and unarchived, described as 'Vendor-agnostic orchestration for training, inference and agentic workloads across NVIDIA, AMD, TPU, and Tenstorrent'. Establishes the license and that the source is public.

  • https://api.github.com/repos/dstackai/dstack/git/trees/master?recursive=1 recorded 2026-09-15

    Full untruncated recursive tree of the default branch, 2,640 entries. One LICENSE.md. No path matches enterprise, ee, commercial or proprietary. A tree lists paths; it is cited for that and for nothing about the vendor's offerings.

  • https://dstack.ai/docs/guides/server-deployment/ recorded 2026-09-16

    dstack's server-deployment guide, naming the products separately: 'Self-hosted dstack - the open-source control plane for AI-native orchestration', beside 'dstack Factory' and the hosted 'dstack Sky'. Establishes commercial offerings beside the open control plane rather than inside it.

Adoption

2 high confidence
2.0

The PyPI download API reports 27,109 downloads of dstack in the trailing 30 days, inside the 10K-100K band for level 2. PyPI is the product's declared distribution channel.

Capability

5 medium confidence
5.0

Admitted on the accelerator limb: the resource request is dstack's own schema and it provisions against it. Band 5 because dstack creates the fleet and places a multi-node task across those nodes, so it both chooses the machines and holds the set together. Banded level with skypilot, which does the same across a different set of backends.

  • https://dstack.ai/docs/concepts/fleets/ recorded 2026-09-15

    dstack's fleets documentation, describing fleets of instances provisioned by dstack across clouds, Kubernetes and on-premise, and used to place runs.

  • https://github.com/dstackai/dstack recorded 2026-09-15

    Repository page and README for dstackai/dstack describing the control plane and its accelerator support.

Verified 2026-09-15