SkyPilot
Category: AI Infrastructure
SkyPilot is a unified AI compute platform that enables frontier AI teams to run, manage, and scale workloads across any cloud, Kubernetes, or Slurm infrastructure through a single control plane. SkyPilot was founded in 2024. The company is led by Zongheng Yang. Based in San Francisco, California, United States. Team size: 11-50. Total funding raised: $20M. Latest round: Seed. Key investors include Lux Capital, Coatue Management, Amplify Partners, Foundation Capital, Race Capital, The House Fund.
- Founded
- 2024
- Headquarters
- San Francisco, California, United States
- Team size
- 11-50
- Total funding
- $20M
Value proposition
Unified AI compute platform that turns fragmented clouds, clusters, and accelerators into one supercomputer — enabling frontier AI teams to run workloads across any infrastructure without vendor lock-in, while improving GPU utilization by 10%+.
Products and solutions
SkyPilot Platform (managed enterprise AI compute platform), SkyPilot Open Source (free, community edition), Supports development, pre-training, post-training/RL, batch inference, sandboxes, and endpoints across all compute environments
Unique value
Neutral, open-source-based GPU orchestration layer that unifies all AI compute (20+ clouds, Kubernetes, Slurm) under one platform — delivering 10%+ GPU utilization gains and eliminating cloud vendor lock-in.
Target customer
Frontier AI teams, foundation model developers, enterprise AI/ML organizations, and research labs operating large-scale GPU fleets across fragmented compute environments
Industries served
AI/ML infrastructure, Financial services (Nubank), Healthcare AI (Abridge, Hippocratic AI), Foundation model development (H Company), E-commerce (Shopify), AI research (Meta FAIR)
Technology advantage
Vendor-neutral GPU orchestration across 20+ clouds, Kubernetes, and Slurm. Born from UC Berkeley Sky Computing Lab research (NSDI 2023 paper). Intelligent scheduling with spot instance optimization, GPU health monitoring, auto-remediation, multi-cluster support. SOC 2 compliant enterprise platform.
How they differentiate
Neutral, cloud-agnostic GPU orchestration layer — does not answer to any single hardware or cloud vendor. Built on a mature open-source project (8.8k GitHub stars, 14M+ downloads) with deep UC Berkeley academic roots. Supports 20+ cloud providers, Kubernetes, and Slurm under one unified interface. Claims 10%+ GPU utilization improvement for customers spending $100M+/year on GPUs.
Main competitors
Run:ai (acquired by NVIDIA, ~$700M, 2024, open-sourced), Kubernetes-native GPU schedulers (Volcano, KAI Scheduler), YottaLabs
Key partnerships
CoreWeave (integration partner), Nebius (integration partner), AWS, GCP, Azure (cloud provider integrations), Kubernetes, Slurm (infrastructure integrations), 20+ cloud/neocloud providers supported
Notable customers
Nubank, Abridge, Applied Compute, H Company, Hippocratic AI, Shopify, Meta FAIR
Major milestones
July 2026: Launched from stealth with $20M seed funding, Open-source project surpassed 14M downloads and 280+ contributors, Platform managing 10,000+ GPUs for private preview customers, Up to 20x performance improvements over open-source version
Growth metrics
14M+ open-source downloads; 8.8k GitHub stars; 280+ contributors; 16 employees (as of Mar 2026 per Tracxn)
Market positioning
Positioned as the "Switzerland of AI compute" — a neutral orchestration layer independent of any cloud or hardware vendor. Competing in the AI orchestration market projected to grow from ~$14B (2026) to $60B+ (2034). Backed by top-tier VCs and prominent AI industry angels.
Geographic focus
Global (US-headquartered, serving frontier AI teams worldwide)
About Zongheng Yang
Ex-Google Brain (TensorFlow research); UC Berkeley PhD in Computer Science (2022, Jim Gray Dissertation Award Honorable Mention); Databricks intern (when ~10 people); Postdoctoral researcher at UC Berkeley Sky Computing Lab advised by Ion Stoica
Latest news about SkyPilot
More AI Infrastructure companies
Official website: https://skypilot.ai