# GitHub migrates 99.99999% of repos to Azure in agent-scale infrastructure shift

> **Key takeaway:** GitHub has moved nearly all repositories to a dual-hosted Azure and data center architecture, with the remaining 0.00001% blocked only by a 25ms network round-trip latency in write quorum serialization.

**Published:** 2026-10-21T16:00:00+00:00 
**Source:** GitHub Engineering Blog 
**Category:** software-engineering 
**Canonical URL:** https://fosswire.org/news/github-migrates-9999999-of-repos-to-azure-in-agent-scale-inf.html 

## Summary
GitHub's Git infrastructure now runs across its own data centers and Microsoft Azure, with the first Azure replica landing in August 2025. About half of Git traffic is already served from Azure. The migration was driven by exponential traffic growth, power capacity limits at existing facilities, and supply chain constraints. The company is now modifying how its Spokes quorum system serializes writes to eliminate the 25ms cross-region penalty that still prevents the busiest repositories from fully migrating.

## How Spokes quorum serialization blocks the final migration
- GitHub's Spokes system uses five replicas per repository for durability, availability, and scale
- The quorum requires only a majority to commit a write, but currently waits for every healthy replica to respond before proceeding
- This means the slowest healthy replica sets latency for all writes in a repository
- Cross-region replicas in Azure are ~1,000 miles from data centers, creating ~25ms round trips per required network exchange
- Each write holds a repository-wide lock until commit, so latency directly caps throughput
- Git's consistency model only requires per-reference consistency, not cross-reference serialization
- GitHub is changing write serialization scope from repository-wide to per-reference, allowing independent reference updates to proceed in parallel

## Why it matters
The architectural shift positions GitHub to eventually run stateless Git replicas that depend only on cloud storage, reducing operational complexity and potentially affecting how other large-scale Git hosts design their infrastructure.