# AI Infrastructure Index
# https://alpha-one-index.github.io/ai-infra-index/
# Maintained by Alpha One Index

> Comprehensive open-source reference for AI hardware specifications, cloud GPU pricing, inference benchmarks, and infrastructure data. Updated hourly via automated pipelines.

## Overview

The AI Infrastructure Index is a vendor-neutral knowledge base covering:
- Data center GPU specifications (NVIDIA, AMD, Intel)
- Cloud GPU pricing from 12 providers (updated hourly)
- AI accelerator specifications (Google TPU, AWS Trainium, Cerebras, Groq)
- Inference benchmarks (MLPerf v4.1, tokens/second)
- Model GPU sizing guide (VRAM requirements)
- Networking and interconnect specifications
- Training cost estimates
- GPU cost optimization playbook
- Buy vs rent decision framework

## Key Facts

- 12 cloud providers tracked (AWS, GCP, Azure, CoreWeave, Lambda Labs, RunPod, Vast.ai, Nebius, OCI, Cudo Compute, Fluidstack, Paperspace)
- 57+ GPU SKUs with pricing
- H100 on-demand range: $1.87-$6.15/GPU-hour (March 2026)
- Pricing auto-updated hourly via GitHub Actions
- MIT License

## Pages

- [Homepage](https://alpha-one-index.github.io/ai-infra-index/): Overview, pricing table, GPU comparison, FAQ
- [GPU Specifications](https://alpha-one-index.github.io/ai-infra-index/specs/gpu-specifications.html): H100, H200, B200, A100, MI300X, Gaudi 3 specs
- [Cloud GPU Pricing](https://alpha-one-index.github.io/ai-infra-index/specs/cloud-gpu-pricing.html): Per-GPU-hour pricing from 12 providers
- [AI Accelerators](https://alpha-one-index.github.io/ai-infra-index/specs/ai-accelerators.html): Google TPU, AWS Trainium, Cerebras WSE-3, Groq LPU
- [Inference Benchmarks](https://alpha-one-index.github.io/ai-infra-index/specs/inference-benchmarks.html): MLPerf v4.1, tokens/second, performance per dollar
- [Model GPU Sizing](https://alpha-one-index.github.io/ai-infra-index/specs/model-gpu-sizing.html): VRAM requirements for LLaMA, Mixtral, GPT-4
- [Networking & Interconnects](https://alpha-one-index.github.io/ai-infra-index/specs/networking-interconnects.html): NVLink, InfiniBand, RoCE, cluster topologies
- [Training Costs](https://alpha-one-index.github.io/ai-infra-index/specs/training-costs.html): GPT-4, LLaMA 3, DeepSeek training cost estimates
- [GPU Cost Optimization](https://alpha-one-index.github.io/ai-infra-index/specs/gpu-cost-optimization.html): Right-sizing, quantization, spot, reserved, batching
- [Buy vs Rent](https://alpha-one-index.github.io/ai-infra-index/specs/buy-vs-rent.html): Cloud vs on-premise TCO analysis

## Citation

@misc{aiinfraindex2026,
  title = {AI Infrastructure Index},
  author = {Alpha One Index},
  year = {2026},
  url = {https://github.com/alpha-one-index/ai-infra-index}
}

Last updated: 2026-03-01