Sky Computing: Charting a New Course in AI Cloud Infrastructure
summary
Sky Computing proposes an intercloud broker model (SkyPilot) that already has over 1 million downloads and a v0.10 release, exposing a unified interface to 17 clouds plus Kubernetes and on‑prem. Claims include 3–6x cost savings from spot orchestration, 4x faster provisioning, 200 GPUs provisioned in under 90 seconds, 9.6x faster checkpointing, and zero code changes using a single YAML and one command. Case studies show Slurm shops replacing brittle pipelines with familiar SSH/Jupyter workflows; blockers remain standards, hyperscaler economics, and heavy legacy inertia.
AI-written summary · about
chapters (9)
- 0:00Introduction and Apologies
- 0:32Cloud Promises and Drawbacks
- 1:26Cloud Repatriation
- 2:58Introduction to Sky Computing
- 3:47Pillars of Sky Computing
- 7:37Sky Pilot: The Inaugural Broker Framework
- 8:47Case Studies and Real-World Impact
- 12:08Challenges and Future of Sky Computing
- 15:05Conclusion and Final Thoughts
related talks
52mSky Computing
17mSkyPilot: An Intercloud Broker for Sky Computing- 52mJoint Keynote Address
11mSkyPilot: Simple Orchestration of Workloads
13mSpace Clouds
16mSkyplane: Optimizing Transfer Cost and Throughput Using Cloud-Aware Overlays
16mThe Supercloud
18mKeynote: Mind the Gap: Bridging Cloud Native Innovation with Real-World Use