Key Takeaways

  • GitHub is experiencing "astronomical" 14x year-on-year growth in commits, now hitting 275 million per week, largely fueled by AI agents.
  • This rapid expansion exposed critical bottlenecks in decades-old infrastructure, including exponential CPU demand from GitHub Actions.
  • Legacy permissioning layers, specifically an internal system called "My SQL 1," are a chief source of performance degradation due to outdated database structures.
  • The rise of large mono-repos is creating unique performance problems, pushing traditional scaling limits and requiring fresh approaches.
  • To cope, GitHub is moving beyond simple vertical or horizontal scaling to a "diagonal" re-architecture, requiring a deep overhaul of 10-15 year old services and a shift to Azure cloud compute.

The AI Tsunami That Broke GitHub

Forget linear growth charts. GitHub, home to over 200 million developers, isn't just growing; it's experiencing what CEO Kyle Daigle calls "astronomical" expansion. The numbers are almost impossible to grasp: The platform now sees 275 million commits per week, putting it on track for a staggering 14 billion commits this year. This is a dramatic jump from the 1 billion commits projected for 2025 just a short while ago, as Swyx noted on the podcast. What's driving this? AI agents. They're not just helping developers; they're acting as developers, churning out code, running tests, and pushing changes at a scale human teams simply can't match.

This explosion of activity, while a triumph, has a darker side: It's breaking things. GitHub has faced recent availability problems because its infrastructure, built over years, simply wasn't designed for this new, AI-driven reality. Daigle said, “The growth is astronomical but also we're making such material progress... I'm excited once we've kind of like re-imagined the underlying foundation layer.” This isn't just about adding more servers; it's about tearing down and rebuilding core components under unprecedented load.

Core Tech Debt Exploded by Hypergrowth

When you hit 14x growth, every old assumption cracks. For GitHub, several core technical challenges have come to light. First, there's the exponential demand from GitHub Actions. “More tools, more agents, more PRs mean more builds. More builds need more CPUs,” Daigle explained. This isn't just a linear scaling problem; it's a compounding effect as AI agents generate more work for the existing CI/CD pipelines. This spike in CPU usage has pushed their systems to the limit.

Then there's the permissioning layer. Many of GitHub's access control systems run on a database internally known as "My SQL 1." Daigle admitted, “The place that we continue to have pain is in uh permissioning. And so right now many of our permissioning layers sit into a database that we like internally call my SQL 1.” Imagine building a skyscraper on a 15-year-old foundation designed for a two-story building – it's wobbly. This legacy layer, while functional for years, simply can't keep up with the volume of checks required by hundreds of millions of developers and their AI assistants.

Finally, the rise of mono-repos presents a unique challenge. Daigle noted, “Repos are bigger... we're just seeing many more big repos. big monor repos have always had like a unique performance problem.” Large, monolithic repositories stress version control systems and build pipelines in ways smaller, more distributed repos don't. GitHub is responding by moving services to Azure cloud compute and adopting what Daigle calls a "diagonal" approach to scaling. This means a deep, architectural re-imagining of 10-15 year old services, not just throwing more hardware at the problem or sharding databases.

What to Do With This

Most founders won't see 14x growth this year. But GitHub's pain offers a sharp lesson: The type of breaking points exposed by AI-driven scale are universal, even if your magnitude is smaller. Don't just plan for linear user growth; identify which parts of your core system would buckle if usage spiked 5-10x in a single month due to a sudden virality event or a new AI integration. This week, audit your application's equivalent of GitHub's "My SQL 1"—your core, legacy permissioning or data access layer. Map out how it would handle a 10x increase in simultaneous requests. Then, simulate a stress test on your most compute-intensive workflows (like GitHub Actions) and your largest data structures (your mono-repos). Prioritize re-architecture for these weakest links before hypergrowth forces your hand.