dstack
Free tierOpen-source GPU orchestration control plane for any cloud, Kubernetes, or on-prem cluster
Free tier available·Technical·API available·Open source
Key strengths
Unified GPU orchestration across clouds, Kubernetes, VMs, and bare-metalSupports NVIDIA, AMD, TPU, and Tenstorrent acceleratorsSelf-hostable open-source with Mozilla Public License 2.0Supports training, inference, dev environments, and fleet managementAuto-scaling, port-forwarding, ingress handlingProgrammatic API and CLI support
Free tier + paid plans
Founded 2022
Self-hostable
Facts verified against dstack.ai on
Get alerts when dstack changes
Free. We'll email you when pricing, the free tier, API access or open-source status changes.
No ratings yet
dstack is purpose-built for the following engineering scenarios:
- Distributed GPU training: Orchestrate multi-node training jobs using frameworks like TRL, Axolotl, or Ray across heterogeneous cloud or on-prem hardware — including NVIDIA, AMD, TPU, and Tenstorrent accelerators
- Inference fleet management: Deploy and auto-scale inference services backed by vLLM, SGLang, Dynamo, NIM, or TensorRT-LLM with built-in ingress and port-forwarding
- Multi-cloud compute arbitrage: Schedule workloads across AWS, GCP, Lambda, Crusoe, and Nebius from a single API, avoiding provider lock-in
- Kubernetes + bare-metal unification: Extend Kubernetes-native workflows to include bare-metal GPU nodes or cloud VMs under one orchestration layer
- Remote dev environments: Provision GPU-backed development environments programmatically for reproducible, hardware-accurate experimentation
- AI token factories and data centers: Manage large accelerator fleets with fleet management primitives and auto-scaling policies suited for high-throughput inference workloads
Alternatives
See all alternatives Alternatives to dstack
Is dstack yours? Claim the listing & get the badge