SkyPilot

The AI Compute Platform for frontier teams.

We turn fragmented AI compute into one AI supercomputer, so frontier AI teams build custom intelligence faster.

[SkyPilot @skypilot_org]

SkyPilot is out of stealth today.

We're announcing our $20M seed funding and SkyPilot Platform — the AI compute platform that turns fragmented compute into one AI supercomputer.

Frontier teams like @appliedcompute, @AbridgeHQ, @hippocraticai, @hcompany_ai, and @nubank already run on SkyPilot, seeing 10x faster time-to-intelligence and double-digit increase in GPU utilization on their own clouds.

Huge thanks to our customers, investors, partners, and founding Pilots.

[Zongheng Yang @zongheng_yang]

Today, SkyPilot is out of stealth.

Building custom intelligence is now existential. We help frontier AI teams build intelligence faster by removing their biggest bottleneck: AI compute fragmentation.

Frontier teams like @appliedcompute, @AbridgeHQ, @hippocraticai, @hcompany_ai, and @nubank already run on SkyPilot, with 10x faster time-to-intelligence and double-digit increase in GPU utilization.

AI teams today get compute anywhere they can. They then firefight compute fragmentation across providers. Researchers burn time on workload setup. Infra gets paged when GPUs go down. Frontier teams build slowly even on the fastest compute.

@skypilot_org turns your fragmented compute into one AI supercomputer, so you run frontier workloads faster. Many users manage 10,000+ GPUs across providers with SkyPilot. GPU hours consumption has grown 6x in the last 6 months.

1/ We're launching SkyPilot Platform — the AI compute platform for frontier AI teams to manage large GPU fleets and accelerate building custom intelligence.

Optimized for fleet management, team governance, and frontier workloads — pretraining, post-training, multi-cluster serving, and sandboxes. SkyPilot open source users can switch to the platform with a server URL change.

2/ We've raised over $20M led by @Lux_Capital (@breeves08), with participation from @AmplifyPartners (@dauber, @lennypruss), @coatuemgmt, @FoundationCap (@ashugarg , @JayaGup10), @RaceCapital, @thehousefund, and top operators like @alighodsi (CEO, Databricks), @JeffDean (Chief Scientist, Google), @rauchg (CEO, Vercel), @amasad (CEO, Replit), @ClemDelangue (CEO, @huggingface) and more.

We're hiring across Engineering and GTM to deliver the platform for the next decade of AI.

Above all, I'm excited to be building with the incredible team we've assembled, along with my cofounders Zhanghao @Michaelvll1, Romil @bromil101, Scott, and Ion @istoica05.

If you firefight AI compute, let's build.

[SkyPilot @skypilot_org]

Check out the blog: https://t.co/KpdgJxM2mL

[SkyPilot @skypilot_org]

RL is Everything, Everywhere, All at Once.

A single RL run spans training GPUs, inference engines, agent loops, and sandboxes, all feeding each other.

We put the whole stack on an everything bagel, and wrote an illustrated tour of it, its tradeoffs and failure modes.

[SkyPilot @skypilot_org]

Read the blog to learn more: https://t.co/gx1UeOnUr0

[SkyPilot @skypilot_org]

Introducing SkyPilot Agent Sessions.

Start an agent, close your laptop. It keeps working on your own K8s cluster, and the dashboard tells you the moment it needs you: from the CLI, the browser, or your phone.

[SkyPilot @skypilot_org]

Congrats to the @qdrant_engine team on Qdrant-FineWeb-10B release!

They ran the whole pipeline on SkyPilot: embedding generation, brute-force search, and loading, spread across AWS, GCP, Azure, Kubernetes, and Slurm.

One unified AI compute platform for all their infra. That's what SkyPilot is for.

[Qdrant @qdrant_engine]

1/ Production vector search: billions of vectors, thousands of RPS, tight tail latency. Public benchmarks don't come close. So we built what was missing and gave it away.

Write-up: https://t.co/6Ia5Y00sva
Dataset on @huggingface: https://t.co/mZacCaYxpf
Supernova Code: https://t.co/wvEnfkn2QP

[SkyPilot @skypilot_org]

A consistent theme at Ray Summit: AI compute is fragmented across clusters and clouds.

AI teams need a simpler way to run their workloads worrying about the infrastructure, and infra teams need management and governance across their fleet.

That's the problem SkyPilot solves: one unified platform for all your AI compute.

🔗 Book a demo: https://t.co/XGmDBy0l4j

[SkyPilot @skypilot_org]

The SkyPilot team will be at Ray Summit!

Join us at our booth to discuss pre-training, RL, serving, and running AI workloads across multiple Kubernetes and Slurm clusters, and score some SkyPilot swag!

Book a slot: https://t.co/Q4CGEDbJlH

[SkyPilot @skypilot_org]

Last night's SkyPilot AI Infra Meetup with @VAST_Data was a full house!

🌟 Highlights:
• Connor Pedersen (@nvidia) showed how they tame the complexity of GB200/GB300 with SkyPilot's topology-aware scheduling
• Zhanghao Wu @Michaelvll1 (@skypilot_org) unveiled the SkyPilot Platform and have a sneak peek at an offering that pays you on your idle GPUs
• Anat Heilper (@VAST_Data) demoed the VAST + SkyPilot integration for instant data access
• Xiaoyu Yao & Jing Zhao (@togethercompute) broke down their shared data plane for inference

Thanks to @VAST_Data for being amazing partners, and to everyone who joined. Excited to keep building this AI infra community. More events are coming soon.

[SkyPilot @skypilot_org]

Today in SF: the SkyPilot AI Infra Meetup with @VAST_Data.

📆 Tue, Aug 18, 5:30 PM
📍 San Francisco
🎙️ Hear from @skypilot_org, @VAST_Data, @nvidia, and @togethercompute on RL infra, agent sandboxes, and AI storage systems
🍕 Food, drinks, and networking

Register now: https://t.co/pzyYBW3Zje

[SkyPilot @skypilot_org]

SkyPilot AI Infra Meetup is next Tuesday in SF!

Join us and @VAST_Data as we discuss optimizing RL infra, scaling sandboxes for agents, and the storage and compute behind training and inference.

Speakers:
• Zhanghao Wu @Michaelvll1, CTO & Co-founder, SkyPilot
​• Anat Heilper, Director of AI Architecture, @VAST_Data
• ​Connor Pedersen, Software Engineer, @nvidia

Limited spots. Register: https://t.co/jPwqF68Tjm

[SkyPilot @skypilot_org]

Managing multiple Slurm clusters is painful.

Each cluster has different login nodes, partition names, and environments. Managing them manually is time-consuming, error-prone, and a bottleneck as your team scales.

SkyPilot provides you with:

• One unified interface to access all your Slurm clusters
• Automatic cluster selection and GPU observability across your fleet
• One YAML, works on any Slurm cluster and extends to K8s too

Read the blog: https://t.co/PPMcNeppDQ

[SkyPilot @skypilot_org]

👋 The SkyPilot AI Infra Meetup is back, this time with @VAST_Data.

📆 Tue, Aug 18 · 5:30–8:30 PM
📍 San Francisco

Tech talks on optimizing RL infra, scaling sandboxes for agents, and the compute and storage systems behind training and inference. Come connect with fellow builders. 🚀

https://t.co/jPwqF68ltO

[SkyPilot @skypilot_org]

RL trainer is holding expensive GPUs waiting on your inference engine.

We trained a Qwen3-14B coding agent, and identified the bottleneck in the training process to be the rollout step: the trainer spent most of every step waiting; rollout generation dominated the step time.

With SkyPilot Job Groups, we were able to scale @sgl_project rollout engines from 1 to 3 with a one line change, and nearly doubled end-to-end speed.

Full benchmarks and setup in the blog: https://t.co/cCh4y1x8dg

[SkyPilot @skypilot_org]

Announcing 𝗦𝗸𝘆𝗣𝗶𝗹𝗼𝘁 𝘃𝟬.𝟭𝟯.𝟬! 🚀

• @huggingface Storage: mount Hub repos and Buckets with one hf:// URL
• Batch Inference Abstractions: large-scale batch inference across clusters
• Lifecycle Hooks: run custom scripts on autostop, preemption, and down events
• Governance & Robustness on API Server: RBAC viewer role, workspace permission checks, and graceful upgrades

Get it now with:
uv pip install "skypilot[all]==0.13.0"

🔗Release notes: https://t.co/wfBWp1c4Ga

[SkyPilot @skypilot_org]

SkyPilot has proudly co-signed the Open Weights letter.

We believe the frontier should be open to everyone building on it.

Open weights give you the model. SkyPilot handles the compute. Together, that’s custom intelligence you own.

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[SkyPilot @skypilot_org]

Kimi K3 is out. 2.8T parameters. 1.4 TB of weights, frontier-level performance.

Serving it on your own infra is the hard part - you need multi-node inference spread across every cluster you can grab capacity from. And then autoscaling, cache-aware routing, PD disaggregation - an entire serving stack around the model.

SkyPilot Endpoints ships a complete stack on day 0 for @Kimi_Moonshot's Kimi K3:

• Runs on GPUs you own - hyperscalers, neoclouds or on-prem.
• Production-ready out of the box - autoscaling, KV-aware routing, PD disaggregation, observability
• Fault-tolerant by design - endpoint stays up even when GPUs fail

Frontier open models, on GPUs you own.

🔗 https://t.co/iPTQnQ4OhT

[SkyPilot @skypilot_org]

🌟Star our Github repo: https://t.co/MS0GN2MJAp

[SkyPilot @skypilot_org]

Fresh out of stealth, and our first announcement belongs to the community: SkyPilot open source has hit 10,000 GitHub stars ⭐

We are now powering AI infra for frontier teams from neolabs to Fortune 500 enterprises, across 25+ clouds, Kubernetes, and on-prem.

Every star, issue, and PR got us here. This milestone belongs to the community. Onwards to accelerate GPU fleets!

[SkyPilot @skypilot_org]

@JensenHuang @nvidia 🫡