SkyPilot

SkyPilot's July 21, 2026 announcement includes a named Abridge ML platform lead account of launching ten experiments in the time previously needed to set up one, documenting a concrete experimentation workflow outcome rather than GPU fleet size alone.

SkyPilot reports use by hundreds of companies, including frontier AI teams; some customers manage 10K+ GPUs and H Company uses it for online RL across 2,000+ GPUs.

SkyPilot reports use by hundreds of companies, including frontier AI teams; some customers manage 10K+ GPUs and H Company uses it for online RL across 2,000+ GPUs.

SkyPilot unifies Kubernetes, Slurm, VMs, on-prem and 20+ clouds into one AI compute layer for training, inference, sandboxes and agent workloads.

SkyPilot unifies Kubernetes, Slurm, VMs, on-prem and 20+ clouds into one AI compute layer for training, inference, sandboxes and agent workloads.

SkyPilot introduced a centralized multi-user client-server control plane supporting asynchronous launches, remote reconnection, shared clusters, multi-tenancy and fault-tolerant cloud-native orchestration, extending its earlier single-machine design.

SkyPilot released version 0.10 in July 2025 with SSO, team-isolated workspaces and external PostgreSQL state persistence across API server restarts, providing concrete delivered organization-oriented infrastructure capabilities.

SkyPilot evolved from Berkeley research into an open-source platform and 2026 commercial company with $20M funding, enterprise controls and major cloud integrations.

SkyPilot evolved from Berkeley research into an open-source platform and 2026 commercial company with $20M funding, enterprise controls and major cloud integrations.

SkyPilot unifies AI compute across Kubernetes, Slurm, VMs, on-premises infrastructure and 20+ clouds, with GPU scheduling, multi-cluster management, quotas, preemption, monitoring and the full AI lifecycle, supporting broad portability across cloud, HPC and ML-infrastructure roles.

Kubernetes, Slurm, GPU scheduling, multi-cloud and AI infrastructure skills transfer exceptionally broadly across frontier labs and cloud infrastructure companies.

Kubernetes, Slurm, GPU scheduling, multi-cloud and AI infrastructure skills transfer exceptionally broadly across frontier labs and cloud infrastructure companies.