Packet.ai + SkyPilot: Run ML Workloads with One Command (Alpha)

packet.ai now has a native SkyPilot cloud provider, letting you run ML workloads with sky launch --cloud packet using the same YAML files you already use for AWS, GCP, or Lambda Labs.

SkyPilot’s power comes from having options. The more clouds in the catalog, the better its optimizer works — more availability, better pricing, less lock-in. The packet.ai integration brings competitive pricing and instant availability from datacenter infrastructure powered by hosted·ai.

Getting started

# Install with packet.ai support
pip install "skypilot[packet]"

# Set API key
export PACKET_API_KEY=your_api_key

# Verify connection
sky check packet

Launch a GPU cluster

# train.yaml
resources:
  cloud: packet
  accelerators: H100:1

setup: |
  pip install torch transformers datasets

run: |
  python train.py --model meta-llama/Llama-3.1-8B

# Launch on packet.ai specifically
sky launch -c my-cluster train.yaml

# Or with explicit GPU type
sky launch --cloud packet --gpus H100:4 train.yaml

# Multi-cloud fallback (packet first, then AWS)
sky launch --cloud packet,aws --gpus H100:1 train.yaml

What’s supported

Pricing through SkyPilot

SkyPilot’s cost optimizer selects the cheapest available option matching your GPU specification. With packet.ai H100 from $0.65/hr, packet.ai will typically be selected over Lambda ($3.29/hr for H100) and AWS ($4.59–$8.90/hr) in any multi-cloud configuration that includes all three.

Combined with SkyPilot’s auto-stop (which terminates clusters when jobs complete), the packet.ai integration reduces both the per-GPU-hour cost and the risk of paying for idle time.

Frequently asked questions

Last reviewed: 10 June 2026. For alpha access: help@packet.ai or browse clusters directly on packet.ai →