RL Is Bottlenecked by Inference. Scale It Independently.
Agentic RL spends most of every step on inference, not training. Scale inference in RL with SkyPilot Job Groups and slime.
SkyPilot Blog
Guides, posts, and resources from SkyPilot.
Blog
Serving Kimi K3 on Your Own GPUs with SkyPilot
Kimi K3's weights are open. Here's how to serve the 2.8T-parameter model across your own GPU clusters with a single YAML.
Accelerating custom intelligence: Announcing SkyPilot Platform and our $20M funding
SkyPilot Endpoints: Production-Ready Inference on Every Cluster You Own
One YAML, one endpoint to run and manage inference on your own GPU fleet.
SkyPilot Sandboxes: Run Agent Code on Your Own Kubernetes, at Scale
SkyPilot runs untrusted, LLM-generated code in sandboxes on the Kubernetes clusters you already own. A single cluster sustains 50,000+ sandboxes, with multi-cluster support to go further. Individual launches in less than a second. All at up to one-tenth the cost of hosted alternatives, with your code and data never leaving your cloud.
RL Doesn't Work on Slurm
Online reinforcement learning for LLMs breaks Slurm's batch scheduling model. We'll discuss why, and what can be done about it.
Testimonials and case studies
Trusted by top AI teams
“SkyPilot provides a flexible orchestration layer that adapts to Nubank's infrastructure and operational requirements, rather than constraining us to a rigid platform.
Equally valuable is the strength of the SkyPilot community and the responsiveness of its support. Together, these have made SkyPilot a foundational part of how we run AI workloads at scale.”
Abhishek Shivanna
Machine Learning Senior Manager, Nubank
“We originally spent some time investing in our own home-grown GPU scheduling solution, but SkyPilot gave us all of those features and more out of the box - gang scheduling, multi-node jobs, SSH access - in a unified interface across all our clouds.
Scaling to new GPU clusters now takes minutes instead of weeks, and our researchers launch 1000s of jobs across all our clouds in seconds.”
Linden Li
Co-founder, Applied Compute
“Some really cool (programming) tech that I think are underrated right now and that have proven very useful to me this year: SkyPilot!”
Tobi Lütke
CEO, Shopify
“We have been using SkyPilot at Unconventional AI to scale very quickly. It has been amazing experience. So much pain removed, quick and easy ways to launch large jobs quickly.
SkyPilot has also been so responsive to us, their platform gets better daily.
Strongly recommend!”
Xavier Gonzalez
AI Researcher, Unconventional AI
“SkyPilot is one of the best training orchestrators I’ve used. Easy to setup, light weight to use and configurable in just the right places.”
Sisil Mehta
ML Platform Lead, Abridge
“Moving from SLURM to SkyPilot was a strong win for us. A unified next-gen platform to manage all our clusters means we can scale GPUs exactly when we need them without lock-in.”
Debajyoti Datta
Co-founder, Hippocratic AI
“SkyPilot is now our standard AI infrastructure layer, powering all our training and enabling us to scale online RL to 2,000+ GPUs on K8s, previously impossible on Slurm.”
Tony Wu
Core Researcher Engineer, H Company
“SkyPilot powers all of our Physical AI workloads — from robotics data processing to research projects — across our entire GPU fleet. With SkyPilot, our researchers move fast without being bogged down by infrastructure.”
Philipp Wu
CEO & Co-founder, XDOF
“SkyPilot simplifies job submission to kubernetes, reducing yaml burden for users to contend with. Open source, growing fast in adoption.
We expect the market to consolidate around a few approaches [...] and SkyPilot for those running in multi-cloud scenarios.”
SemiAnalysis
AI & Semiconductor Research Firm
“We were fortunate to work with SkyPilot at Eigen AI and were incredibly impressed—not only by how much easier the product made it to run AI workloads across fragmented compute, but also by the technical depth, speed, and customer focus of the team behind it.”
Ryan Wang
Co-founder and CEO, Eigen
“As we scaled, we needed to onboard new GPU vendors like Nebius for additional GPU capacity. SkyPilot Platform made it super easy for our AI team to start training foundation models on new infrastructure from day one, with built-in GPU health monitoring, no Kubernetes expertise needed.”
Rui Zhang
Head of Core Engineering, HeyGen
“SkyPilot is one of my favorite tools. I've been using it since 2023 both at work and for my side projects.”
Isaac Vidas
Machine Learning Engineering and Platform, Shopify
“SkyPilot enables us to accelerate and scale our experimentation by seamlessly managing compute resources across clouds and clusters.
It abstracts away low-level infrastructure operations, such as provisioning and scheduling, allowing researchers to focus on model development and evaluation.”
Srijith Rajamohan
Head of AI Research, Redis
“Skypilot makes prototyping ML workloads incredibly low friction! I can just throw this header at the top of my script, and it'll run automatically.”
Kyle Corbitt
CEO, OpenPipe
“SkyPilot's approach to a unified interface across clouds is the kind of thoughtful abstraction the AI community needs.
”
Jeremy Howard
Founder, Answer.AI
“Infra behind RFM-1: Our Robotics Foundation Model is trained on several clouds; this gives us more GPUs for high experimentation velocity.
”
Rocky Duan
CTO, Covariant
“SkyPilot has been a great tool for saving costs and scaling easily on neoclouds.
What used to require weeks of custom setup now takes minutes, giving us seamless access to better GPU availability and pricing beyond just the hyperscalers.”
What used to require weeks of custom setup now takes minutes, giving us seamless access to better GPU availability and pricing beyond just the hyperscalers.”
Hongbo Miao
Senior Staff AI Engineer, Archer Aviation
“Awesome package that we use on a daily basis for our NIH BRAIN project. Thx Skypilot team 🙏”
Joe Ecker
Director, Genomic Analysis Lab, Salk Institute
“GenAI workloads demand new tools to handle everything from dev pods to large training runs, often on different clusters and accelerators. SkyPilot is such a tool that gives us not only these AI-native capabilities but also the flexibility of infra choices. Our large-scale model training has all moved to SkyPilot now.”
Junghwan Lim
CTO, Motif Technologies
“Great cloud management framework for AI researchers. SkyPilot makes provisioning and controlling GPU instances across cloud providers simple and cost-efficient, thereby allowing the scientists to focus on core AI work.”
Ljubomir Buturovic
VP Machine Learning, Inflammatix
“SkyPilot has become our go-to platform for AI workloads. Its intelligent scheduler saves us the overhead of manually managing GPU resources.
Our teams are more productive, and our compute costs are substantially lower.”
Our teams are more productive, and our compute costs are substantially lower.”
Srivatsan Sridhar
Head of Data Science, 314e Corporation
“By integrating Nebius with SkyPilot we are able to execute jobs across multiple GPU providers without disrupting internal processes.”
Javier Moreno
Principal Engineer, Shopify
“SkyPilot's flexibility across k8s and neoclouds lets us easily get capacity when we need it - we're hitting 90%+ utilization without any overprovisioning!”
Moin Nadeem
Co-founder & CEO, Phonic
“SkyPilot has enabled our scientists to get up and running quickly, removing the need to wrestle with Kubernetes provisioning, resource tuning, or configuration details.
It makes our complex Kubernetes cluster feel accessible and empowers AI researchers - regardless of their k8s experience - to run scalable workloads with confidence.”
Jason Chin
VP, AI Development, Pathos
“We always find ourselves migrating between clouds to find the best deal for our long-running training jobs. SkyPilot has made interoperability across clouds very ergonomic, and our entire team is now scheduling workloads via our centralized SkyPilot deployment.
This has saved us countless hours wrestling with manual config and allows us to focus on the actual ML research. Could not recommend enough!”
Eitan Borgnia
COO, Relace
“SkyPilot stands out because it treats governance and cost visibility as first-class concerns. That combination makes it much easier to run large-scale AI workloads responsibly while keeping operational controls and spend in view.”
Jayachander Kandakatla
Senior MLOps Engineer, Ford Credit
“SkyPilot's ease of use and multi-cloud flexibility make it the obvious choice for our AI infrastructure. Our researchers run workloads easily across multiple clouds and on-prem without wrestling with Kubernetes - and when requirements shift, the infra team can pivot cloud vendors in minutes, not months.”
Casper Thuis
Technical Lead, Core AI, TKH Group
Scaling AI Infrastructure at Abridge with SkyPilot
Abridge moved from SLURM to a scalable multi-cloud AI infra with SkyPilot, unlocking 10x faster development cycles.
How H Company Unlocked Online RL and Unified their AI Platform
SkyPilot enabled H Company to scale online RL to 2,000+ nodes on K8s, previously impossible on Slurm.
SkyPilot at Shopify: Multi-cloud GPUs without the pain
Shopify runs all AI training on SkyPilot, launching jobs in minutes instead of days with a unified interface across clouds.
The latest news
SkyPilot raises $20M to be the Switzerland of AI compute
Today, Stoica and Yang are publicly launching SkyPilot, backed by $20 million seed funding, Fortune learned exclusively.
The Future Of Computing: Supercloud And Sky Computing
Cloud computing, multi-cloud, and hybrid-cloud are all terms we’ve become used to hearing. Now we can add “super cloud” and “sky computing” to the list of terminology that describes the computing infrastructure of the coming decade.
From Partners
Run AI workloads on any cloud, store on Hugging Face: zero-egress storage with SkyPilot
Together with Hugging Face, we've joined the two halves: your models and datasets stay on the Hub, and SkyPilot runs the compute (dev, training, or serving) on whatever cluster has the GPUs.
Instant Data Access for SkyPilot AI Workloads
VAST lets teams achieve instant data access for SkyPilot AI workloads on any compute (cloud, neocloud, on-prem) and start working immediately. Same data, same YAML, just different compute
How to deploy ML jobs on Lambda Cloud with SkyPilot
SkyPilot is an open-source orchestration tool that automates ML job deployment on Lambda Cloud. This tutorial covers installation, configuration, and running an example job that evaluates DeepSeek-R1-Distill-Qwen-7B on multiplication tasks with automatic instance termination when the job completes.
How SkyPilot solves modern AI cloud workflow challenges
A hands-on guide to training and serving modern language models across clouds with SkyPilot and Weights & Biases, showing why this workflow is simpler and more portable than legacy solutions like SLURM
Subscribe to our monthly newsletter
Get the latest stories, tips, and insights delivered to your inbox.