GitHub's Commits Doubled in 4 Months. Its Servers Couldn't Keep Up.

GitHub went dark for large parts of eight hours on August 17, and the company says the reason is simple: too much traffic, not enough infrastructure to handle it.

The outage ran from 13:28 to 21:15 UTC, according to GitHub's own incident timeline, cited by cyberinsider and explainx.ai. That's 7 hours and 47 minutes. During the worst of it, web and API traffic saw roughly 20% error rates, while downloads of archives and raw repository content failed around 50% of the time, per GitHub's status page as reported by BigGo Finance and gigazine.

The list of broken services reads like most of what GitHub does: github.com itself, GitHub Actions, Pull Requests, Issues, Pages, Webhooks, SAML and OIDC authentication, SCIM, Team Sync, and Copilot. Downdetector logged nearly 3,000 user reports at the peak, according to cyberinsider.

What actually broke

GitHub published its root cause analysis on August 20, three days after the outage. The company's own explanation, quoted by Engadget: "traffic reached a new peak, and a critical infrastructure component in our Central US data center failed to scale with it."

The mechanical detail, laid out by BigGo Finance and gigazine, points to a single blind spot. GitHub uses Istio sidecar proxies to manage communication between internal services. Those sidecars hit their processing limit under record traffic. But GitHub's auto-scaling system only monitored the main service, not the sidecar's ceiling, so it never triggered the expansion that was needed.

From there it cascaded. The capacity shortage spread to GitHub's HAProxy load balancers, four of which hit their own processing limits. That created delays in GitHub's internal authentication system, and because nearly every GitHub service depends on authentication, the failure spread wide.

Retry logic made it worse. When services fail, systems automatically retry the connection. GitHub told users the flood of retries piled more load onto an already strained load balancer. explainx.ai reports a specific bug in Visual Studio Code compounded this. When GitHub Copilot authentication failed, VS Code triggered a surge of retry attempts that pushed traffic to the Copilot token service from a normal 7,000-9,000 requests per second up to 70,000-100,000 requests per second, roughly 10 times normal load. That's why Copilot took longer to recover than everything else, with GitHub's timeline showing Copilot's token service didn't fully recover until 21:02-21:15 UTC, near the very end of the incident window.

The AI traffic surge behind it

The headline number GitHub disclosed: monthly commits grew from 1.4 billion in April to 2.9 billion in August, more than doubling in four months, according to GitHub's own blog post as cited by Engadget, BigGo Finance, and explainx.ai. GitHub attributes that growth to pressure from AI-assisted coding tools generating far more commits, pull requests, and automated workflows than human developers alone ever did.

cyberinsider adds important context that some other coverage skipped. This wasn't a one-off. GitHub reported 26 separate incidents in July 2026 alone, and the August 17 outage was the platform's second major incident that month, following an Actions failure on August 6, per explainx.ai's timeline. GitHub CTO Vladimir Fedorov told cyberinsider the company had already been working to scale infrastructure 10-fold last year, then determined it actually needed a 30-fold increase to keep pace with AI-driven workload growth.

What GitHub says it's doing next

GitHub's stated fixes, per explainx.ai's account of the postmortem, include adding roughly 3 million CPU cores and 120 petabytes of storage, shifting more load onto Microsoft Azure infrastructure (now handling an estimated 58% of load), and accelerating the broader migration to Azure, according to BigGo Finance. The company also says it's implementing consistent retry limits, retry budgets, and variable timeouts to stop future "retry storms" from cascading, in its own words as quoted by Engadget.

None of that addresses the underlying driver. If AI tools keep generating more commits, more pull requests, and more automated Actions runs every month, GitHub's infrastructure will keep getting tested. GitHub itself says the growth in traffic doesn't excuse the outages, but the company's own numbers suggest the traffic isn't slowing down.

What's unresolved

GitHub hasn't said what specific commit or usage volume it's now engineering toward, or whether the 30-fold capacity target Fedorov described to cyberinsider has been met. With 26 incidents in July and two major outages in August, the open question is whether GitHub's fixes are ahead of the growth curve or still chasing it. The next high-traffic spike, whenever it comes, will be the real test.