A single GitHub load balancer error knocked eight services offline for 180 million developers

Development teams worldwide had built their entire workflow around GitHub — their pipelines, AI tools, and version control. When a misrouted load balancer rule took eight services offline on March 3, 2026, none of that workflow had an independent fallback.

Because when you concentrate critical operations on a single vendor, their outage becomes your outage.

This article covers what broke, why it kept happening, the four threats it exposes, and how software-dependent businesses can use software escrow to build resilience before their vendor's next incident becomes their problem.

TL;DR

• What happened: A faulty routing configuration took down eight GitHub services simultaneously, halting code access, automated pipelines, and AI tools for approximately 70 minutes.

• Why it matters: This was one of eight major GitHub outages in two months on a platform that 90% of Fortune 100 companies depend on. Most affected teams had concentrated their entire workflow there with no independent fallback.

• What to do: Reduce single-vendor exposure by backing up your pipelines and dependencies outside the platform, and protecting your code with software escrow.

A load balancer misconfiguration took down 8 GitHub services in under five minutes

At 18:59 UTC on March 3, 2026 — peak business hours on the US East Coast — an internal networking error began throttling traffic to the server cluster handling GitHub Copilot's AI code suggestions. GitHub's architecture lacked isolation between its core services, so the fault spread from that cluster across the platform.

Within minutes, the cascade was underway. Copilot went dark first. Then GitHub Actions — the automation layer that runs build, test, and deployment jobs — stalled as its server fleet lost the instructions telling it what to run.

API endpoints began returning errors, taking down third-party integrations, monitoring dashboards, and deployment automation with them. Pull requests timed out. Git push and pull operations failed.

Webhooks stopped firing entirely, cutting the automated signals that trigger downstream tools like deployment pipelines and ticketing systems. Codespaces failed to start. GitHub's SRE team identified the root cause within 18 minutes. Full restoration took 70.

The platform hosts 180 million developers, and 90% of Fortune 100 companies depend on it for their software operations. Oxford Economics and Splunk put the average cost of downtime at $540,000 per hour across Global 2000 companies.

That cost lands the same way whether the failure was yours or your vendor's. On March 3, thousands of companies paid for a configuration change they didn't make and couldn't see.

This is what downstream dependency risk looks like in practice

The outage was not a one-off occurrence for GitHub. Between February 2 and March 20, 2026, GitHub logged eight major outages spanning entirely different root causes: storage failures, misconfigured services, cloud infrastructure errors, routing faults.

On January 13, 2025, a faulty load balancer configuration change caused 49 minutes of Git downtime. After that incident, GitHub publicly committed to improving its monitoring and deployment practices to prevent a recurrence. Fourteen months later, the same class of failure took down eight services simultaneously. The fix didn't hold, and that shifts the story from operational failure to something more structural.

GitHub promises Enterprise Cloud customers 99.9% availability. Independent monitoring put their rolling 90-day uptime as low as 90% — the equivalent of 36 days offline per year against an SLA that permits fewer than nine hours.

On March 11, GitHub's CTO Vlad Fedorov wrote that the platform "has not met its own availability standards," attributing the failures to three structural problems: rapid load growth, architectural coupling that let local faults cascade, and no way to shed load from misbehaving clients. Engineers called it an architecture without circuit breakers, the safeguards that stop one failure from tripping every connected system.

Compounding the architectural problems is a migration that is still actively underway. GitHub has been moving its infrastructure to Microsoft Azure since October 2025. At the time of Fedorov's post, 12.5% of traffic had made that move, with a target of 50% by July 2026. Until it completes, the old and new systems stay dependent on each other.

If your team has built critical workflows on GitHub, that's not GitHub's exposure. It's yours.

Update, August 2026: On August 17, GitHub went down for 7 hours and 47 minutes after a capacity failure in its Central US data center. Authentication, Actions, APIs, pull requests, issues, and Copilot went with it. That one outage burned almost a full year's downtime allowance under a 99.9% SLA. Fedorov's postmortem was his third reliability commitment since March.

The threats that downstream dependency risk poses to you

When a platform as central as GitHub fails, how deeply it cuts depends entirely on what recovery systems and independent fallbacks each team had built beforehand — and the businesses that absorbed the most damage were the ones unprepared for third-party failure. These are the four most common threats associated with dependency failure:

• Failure: When GitHub goes dark, every team that stores, reviews, or deploys code through it loses access at the same moment. The repositories survive. Pipelines, deployment scripts, secrets, and the review record do not, and none of those live on a developer's laptop.

• Attacks: GitHub hosts 630 million repositories, making it the effective software supply chain for most of the development world. When attackers target that infrastructure, one compromised component reaches everything connected to it downstream. What you have stored before a compromise becomes your only clean path back.

• Non-compliance: When the third-party tools and cloud environments your operations depend on go offline, your own uptime commitments, SLA obligations, and regulatory requirements don't pause with them. That downtime is yours to answer for. 

• Broken: Most of GitHub's 2026 outages traced back to configuration changes that broke systems in ways nobody caught until production went down. You inherit that risk from every vendor whose release process you can't see.

GitHub isn't the poster child for this kind of dependency risk; it happens more often than you think. One faulty configuration update also caused the July 2024 CrowdStrike incident, which disabled 8.5 million Windows systems globally.

Splunk and Oxford Economics found that 43% of downtime events trace to network or IT environment problems, and another 24% to application or infrastructure failures. Two-thirds of unplanned downtime has nothing to do with an attack. A single vendor's internal error reaching every business wired into it is not an edge case. It is how concentrated infrastructure fails.

If this happened to your business

Your team is in the middle of a hotfix for a critical production bug. Your developer opens a pull request for review, and that's where your workflow stops. The pull request system is degraded. Copilot is unavailable. GitHub's status page shows eight services affected, an investigation underway, and no estimated resolution time.

For the next 70 minutes, your production bug runs live while your tools are down. Your engineers shift from fixing to communicating — updating stakeholders, fielding escalations, documenting what was attempted and when. GitHub's services eventually recover, but at different times across different services, with no single moment where you can confirm your environment is fully operational.

Restoring your access ends the outage, but it doesn't change what happened while it was open. Your audit trail has a documented gap, and if your team was navigating a DORA or NIS2 deadline that month, the disruption is now part of your compliance picture.

» If you don't have a recovery plan that works independently of your vendors, start building one here.

How software escrow solves your dependency risk

Wondering what you can do about dependency risk? You can't audit your vendor's deploy pipeline or sit in their incident review, and having a backup service to migrate to doesn't work well when all your data is lost. What you can do is hold a working copy of the external software you depend on with software escrow.

Software escrow works like an internal backup with rollback versions, except the system it backs up belongs to your supplier. Source code, configurations, deployment infrastructure, credentials, and your client and operational data sit in a vault outside their environment. When the vendor shuts down, sunsets the product, or drops support for the version you run, you deploy your own copy and keep working.

That only holds if the copy works. GitHub committed to a fix in January 2025 and found out fourteen months later that it hadn't held. Untested deposits fail the same way. Software escrow verification runs the build, deploys the application, and checks admin access, so the deposit is confirmed while there's still time to fix it.

When your vendor goes down, preparation is the only variable you control

GitHub's outage did not single anyone out. 90% of the Fortune 100 build on the same platform, so when it stopped, a large share of the industry stopped with it.

That is why concentration rarely gets fixed. Nobody loses ground to a competitor when the competitor is down too, so it gets absorbed and the dependency stays where it is. The real damage shows up later, when a smaller vendor fails and takes weeks to come back instead of 70 minutes.

Codekeeper's Software Escrow and Verification solutions give you a tested copy of that vendor's software, held outside their infrastructure, so their recovery time doesn't determine your recovery time.