dstack Inc.
dstack builds an open-source, vendor-agnostic orchestration layer/control plane for heterogeneous AI compute. It standardizes provisioning and management of GPU workloads across GPU clouds, Kubernetes, VMs, bare-metal and on-premises clusters for development, training and inference.
At a Glance
- AI/ML teams
- AI labs
- data-center operators
- GPU cloud tenants
- +2 more
AI Tools by dstack Inc.
(1)
dstack
Open Source AI Workload Orchestration
Discussions
No discussions yet
Be the first to start a discussion about dstack Inc.
Latest News
Presets: a toolkit for agent-based inference optimization
Deploying NVIDIA Dynamo PD disaggregation with dstack
Deploying inference endpoints with PD disaggregation on AMD GPUs
Deploying SGLang with PD disaggregation via Shepherd Model Gateway
Products & Services
Mozilla Public License 2.0 orchestration stack with CLI, API and server for provisioning and orchestrating AI workloads across clouds, Kubernetes, Slurm, VMs, bare-metal and on-prem clusters.
Hosted GPU orchestration service with GPU Marketplace and Bring Your Own Cloud modes; provides provider access, provisioning and unified workload management.
Commercial multi-tenant orchestration stack for AI labs, data centers and AI token factories; formerly called dstack Enterprise.
Preview open-source toolkit for agent-driven inference optimization with a portable preset format.
Market Position
dstack positions itself as an AI-native, open-source and vendor-agnostic alternative to using Kubernetes or Slurm directly for containerized AI workloads. Its differentiators are first-class GPU provisioning, training and inference primitives, one interface across heterogeneous accelerators and cloud/on-prem environments, and portability that reduces infrastructure and vendor lock-in.
Leadership
Founders
Andrey Cheptsov
Founder and CEO; former Product Manager for PyCharm and DataSpell at JetBrains and previously a product-marketing manager. He studied at Saratov State Technical University and the Technical University of Munich.
Vitaly Khudobakhshov
Co-founder; worked at JetBrains, including creating tools for data engineers, and was Director of Engineering at Product Science Inc.
Executive Team
Andrey Cheptsov
Founder and CEO
Former JetBrains Product Manager for PyCharm and DataSpell; open-source and AI infrastructure practitioner.
Founding Story
dstack was started to give AI/ML teams a simpler, open-source alternative to building and operating bespoke infrastructure on top of Kubernetes, Slurm, or a single GPU vendor. Its initial vision was a portable command-line/API-driven way to provision compute and run ML development and training workflows across clouds.
Business Model
Revenue Model
dstack OSS is open source. dstack Sky monetizes hosted GPU Marketplace usage through prepaid credits and resource charges; its BYOC mode currently has no charge under the terms. dstack Factory is a separate commercial deployment governed by written agreements.
Pricing Tiers
Customers connect and manage their own AWS, GCP, Azure or other supported cloud accounts.
Usage is deducted from prepaid credits; pricing is displayed in the console before provisioning.
Commercial deployments are governed by a separate written agreement.
Target Markets
- AI/ML teams
- AI labs
- data-center operators
- GPU cloud tenants
- enterprise platform teams
- AI token factories
- AI/ML development environments
- Model training and fine-tuning
- Batch and distributed workloads
- Production LLM and generative-AI inference
- Multi-cloud GPU cost and capacity optimization
- On-premises/data-center AI cluster operations
- Electronic Arts
- Graphsignal
- Toffee
- Metamorphic