Blog
Insights on GPU computing, infrastructure, and building efficient AI workloads.
GPU Utilization: The Lie Your Dashboard Tells You
Your GPU shows 90% utilization but your training is crawling. We explain why GPU_UTIL is fundamentally broken and introduce SM Activity metrics that reveal what your GPU is actually doing.
Packet.ai + SkyPilot: Run ML Workloads with One Command (Alpha)
We've built a native SkyPilot integration for Packet.ai. Launch GPU clusters, run distributed training, and deploy models using the same familiar SkyPilot interface—now with access to H100s, RTX 6000 Pros, and more. Alpha testers get $100 in free compute.
Why We Built Token Factory
The story behind Token Factory—why we decided to build an OpenAI-compatible inference API, the problems we wanted to solve, and our vision for democratizing access to AI.
Token Factory: How We Built a 98% Cheaper OpenAI Alternative
A deep technical dive into Token Factory—our OpenAI-compatible inference API with real-time chat, batch processing, LoRA fine-tuning, and pricing at $0.10-0.15 per million tokens.
How Persistent Workspaces Actually Work (A Deep Dive)
GPU pods are ephemeral by default. Here's exactly how we make your files survive restarts — down to the bash scripts, init markers, and volume mounts.
One-Click GPU Environments: VS Code, Jupyter, and More
Launch a GPU with your development environment pre-configured. No SSH setup, no manual installs—just click and code.
Welcome to Packet.ai
Packet.ai was built for teams that need serious GPU compute - without the usual friction.
How Dynamic GPU Placement Enables Lower Prices
A deep dive into how packet.ai's scheduling technology differs from slicing and oversubscription—and why it means better pricing for you.
Why We Only Use Infrastructure Powered by hosted.ai
We don't just resell GPUs. We partner exclusively with infrastructure we understand deeply—and can stand behind completely.
GPU Cloud Pricing: Dedicated vs Spot vs packet.ai
A clear comparison of different GPU pricing models, their trade-offs, and how packet.ai fits into the landscape.