💾 Software · Infrastructure Crisis

GitHub Is Breaking, Copilot Ate the Infrastructure

48 major outages in 12 months, a CTO memo demanding 30X scale, and the Silicon Valley legend who just walked away

The reliability crisis at Microsoft’s most strategic developer platform is not a rumor anymore. Here are the numbers, the internal admissions, and the exit that made the industry stop pretending.

📅 Updated July 4, 2026 ⏱ 9-min read
GitHub Reliability, May 2025 to April 2026 Total Incidents 257 Tracked by IncidentHub Major Outages 48 Frequency rising since Dec 2025 GitHub Actions Alone 57 Most affected service CTO Vlad Fedorov’s Admission 30X scale target Started 10X plan Oct 2025, revised Feb 2026 Mitchell Hashimoto Left Ghostty 52K+ stars April 2026, after 18 years on GitHub

In April 2026, HashiCorp co-founder Mitchell Hashimoto, one of the most respected engineers in Silicon Valley, published a blog post announcing that his terminal emulator project Ghostty, which has more than 52,000 GitHub stars, was leaving the platform. His line was surgical: “This is no longer a place for serious work, if it blocks you out for hours per day, every day.” After 18 years on GitHub, he was done.

GitHub’s Chief Operating Officer responded on X within hours. “I’m sorry. The team is going to keep working to make GitHub something you can come back to with real proof, not words.” Freelance developer David Bushell had already published a post titled “GitHub is Sinking.” The Register ran an article in February concluding that GitHub was struggling to hit three-nines availability, which is enterprise speak for less than 99.9% uptime. The status page itself had quietly stopped displaying aggregate uptime numbers.

By the numbers, from IncidentHub’s monitoring data between May 2025 and April 2026, GitHub logged 257 total incidents and 48 major outages. GitHub Actions alone recorded 57 outages. Copilot recorded 44 outages, 9 major, with 10 in the single month of February 2026. And here is the irony everyone in Silicon Valley is now saying out loud: the AI product Microsoft has been aggressively pushing across every surface of its business is a major reason its most strategic developer platform is falling apart. Copilot ate the infrastructure. This is how it happened.

This article walks through the actual outage data, why GitHub’s own CTO Vlad Fedorov admitted in April 2026 that the platform needs to scale by a factor of 30, how Microsoft’s decision to fold GitHub into its CoreAI organization changed the incentive structure, and what specific steps engineering teams are already taking to protect themselves from further degradation. If you ship code for a living, this is the reliability story of 2026, and it is not being covered honestly enough anywhere else.

📊 The Numbers Behind the Crisis
Data 1

48 Major Outages in 12 Months

IncidentHub tracked 257 total incidents from May 2025 to April 2026, with 48 classified as major. Frequency has been climbing since December 2025

Data 2

Copilot Broke 44 Times

44 Copilot-specific outages including 9 major incidents. February 2026 alone saw 10 Copilot outages. Recent Cloud Agent degradation lasted 2.5 days in June 2026

Data 3

GitHub Actions: 57 Outages

The single most affected GitHub service. Actions is the CI/CD backbone that thousands of enterprise workflows depend on, and it has failed more than any other component

Data 4

CTO Says They Need 30X Scale

Vlad Fedorov’s April 2026 blog post: started a 10X capacity plan in October 2025, but by February 2026 it was clear they needed to design for 30X current load

The Timeline That Explains Everything
Event Date What Happened Impact
Aug 2025 CEO Steps Down Thomas Dohmke leaves, no replacement named GitHub folded into Microsoft CoreAI
Oct 2025 10X Plan Starts CTO announces capacity buildout target Multiple Actions outages that month
Dec 2025 Agent Coding Surge Agentic development workflows accelerate sharply Repo creation, PR activity, API usage spike
Feb 2026 Worst Month 37 incidents, plan revised from 10X to 30X 10 Copilot outages, 7 major
Apr 2026 Hashimoto Leaves Ghostty exits GitHub after 2h 13m Copilot Cloud Agent outage Public credibility rupture
Jun 2026 Still Ongoing Copilot Cloud Agent degraded 2.5 days, 8% avg error rate 27% peak error on Completions

You can watch the failure curve bend upward from December 2025 forward. That is the same window in which agentic AI development, meaning coding agents that spawn pull requests, run tests, and modify repositories autonomously, went from a niche pattern to a mainstream workflow. GitHub’s own CTO named that shift as the primary driver in his April 2026 blog post. Which is not a small admission. It is GitHub’s own leadership saying that the way Microsoft’s flagship Copilot product is being used is what is breaking GitHub.

Five Angles That Explain the Whole Story
01

The Numbers Nobody at Microsoft Wanted You to See

The Data

The most damning single artifact in this whole story is the disappearance of aggregate uptime numbers from GitHub’s own status page. If you had checked the platform’s status page a year ago you could see rolling uptime figures. Now you cannot. An independent developer, Marek Šuppa, built a mirror called The Missing GitHub Status Page that rebuilds those numbers from archived incident data. It is not flattering.

IncidentHub, an independent status aggregator, filled the reporting vacuum with a full audit from May 2025 through April 2026. Their headline numbers: 257 total incidents, 48 major outages, and a clear upward trend since December 2025. GitHub Actions is the most bruised component with 57 individual outages. Copilot is right behind with 44 outages and nine of them major, ten of which came in February 2026 alone. When The Register looked at the same data in February, its conclusion was blunt: GitHub was struggling to hit three-nines availability, which for a platform Microsoft bought for $7.5 billion in 2018 is embarrassing.

What makes the situation harder to spin is that the pattern is not random. February 2026 alone recorded 37 incidents, the worst single month in the tracked period. Both February and April 2026 tied for the worst months by major incident count with seven each. That is not a series of unrelated bad breaks. It is a consistent slope, month after month, in the direction of more failures, longer failures, and failures that cascade across more services. Any engineering leader looking at that pattern reaches the same conclusion GitHub’s own CTO reached in his April memo: this is a scaling problem, and it is going to take a lot more capacity to solve.

💡 What “three-nines” means. 99.9% uptime, or roughly 8.76 hours of allowed downtime per year. GitHub is losing more than that in individual months right now. Enterprise SLAs typically demand higher.
02

Mitchell Hashimoto’s Exit Was Not About One Outage

The Rupture

Hashimoto is not a random developer. He co-founded HashiCorp, the infrastructure company that IBM acquired in 2025 for roughly $6.4 billion, and he is one of the more recognizable technical voices in modern DevOps. When someone at his level publicly says a platform “is no longer a place for serious work” and moves a project with 52,000 stars off it, the industry pays attention. His stated trigger was a specific two-hour and thirteen-minute Copilot Cloud Agent outage on April 27, 2026, but his real complaint was cumulative: every single day for months he had lost hours to some GitHub subsystem failure.

The most damaging line in his post was not the exit announcement. It was the phrase, “I want to ship software, and it doesn’t want me to ship software.” That is a specific accusation. It says the platform’s product priorities are actively getting in the way of the workflow it exists to serve. GitHub COO Kyle Daigle apologized directly on X. But an apology from the COO does not fix the underlying architecture, and everyone in the industry knows it. Hashimoto’s exit reframed the story. Before that post, GitHub outages were a technical inconvenience. After it, they became a strategic risk that competitors could exploit.

💡 Why Ghostty leaving matters. Terminal emulators are the tool developers stare at all day. Ghostty’s exit signals to every project maintainer that leaving GitHub is now a legitimate option, not a fringe stance.
03

The Copilot Paradox, AI Broke Its Own Home

The Irony

Here is the part that is impossible to overstate. In his April 2026 blog post, CTO Vlad Fedorov named the driver of the load surge in plain terms: “The main driver is a rapid change in how software is being built. Since the second half of December 2025, agentic development workflows have accelerated sharply.” Agentic workflows means AI coding agents, and the flagship agentic product in the ecosystem is Copilot, and Copilot is owned by GitHub, and GitHub is owned by Microsoft.

Repository creation, pull request activity, API usage, automation, and large-repository workloads are all growing quickly, per Fedorov’s own numbers. That is happening because Copilot Cloud Agent and similar agentic tools are creating pull requests and running Actions workflows around the clock at a rate no human team could match. GitHub’s platform was designed for humans clicking merge buttons. It is now being asked to serve autonomous agents making thousands of concurrent operations. Microsoft acquired GitHub in October 2018 for $7.5 billion. Copilot launched in June 2021. The commercial pressure to grow Copilot revenue, particularly after Microsoft folded GitHub into its CoreAI organization in August 2025, competes directly with the engineering attention reliability now demands. That is the whole loop.

The recent June 2026 incident record makes the pattern painfully concrete. Between June 16 and June 28, GitHub logged Copilot degradation on at least five separate days, including a 2.5-day Copilot Cloud Agent incident where up to 26% of agent tool calls failed silently. Silent failures are the worst kind, because agent jobs appear to succeed while actually doing nothing. That is exactly the failure mode that erodes trust fastest, because developers stop believing the CI signal even after the platform is back up.

💡 The revenue tension. Copilot is one of Microsoft’s most strategic products, integrated across Office, Windows, and Azure. Slowing Copilot to protect GitHub reliability is not politically feasible inside Microsoft. So both problems compound.
04

Why the 30X Scale Number Is Actually Terrifying

The Admission

The most quietly alarming line in Fedorov’s April 2026 post was the scale revision. GitHub started a 10X capacity plan in October 2025. Four months later, in February 2026, they realized they needed to design for a future that requires 30X today’s scale. That is not a normal quarterly correction. That is engineering leadership admitting the load projection they built their reliability plan around was off by a factor of three, in the direction of more load, in less than half a year.

Fedorov identified the technical causes with unusual candor: “rapid load growth, architectural coupling that allowed localized issues to cascade across critical services, and inability of the system to adequately shed load from misbehaving clients.” That is a distributed systems engineer’s way of saying that when one thing on GitHub breaks it now takes other things with it, and there is no reliable way to isolate the damage in real time. Root cause analyses on the February 2 and March 5, 2026 outages revealed single points of failure, which is a term any senior infrastructure engineer will tell you should not appear in the post-mortem of a platform hosting the world’s code.

💡 What a 30X target really costs. Building three-times-more capacity than an already-massive plan requires enormous data center, network, and engineering investment. Microsoft is choosing to fund it, but the visible reliability wins will not appear for many months.
05

The Governance Vacuum Microsoft Created in August 2025

The Structure

When GitHub CEO Thomas Dohmke stepped down in August 2025, Microsoft made a decision that a lot of ex-GitHub employees found revealing. They did not replace him. Instead, GitHub was folded directly into Microsoft’s CoreAI organization, with the remaining leadership reporting up through Microsoft executives. For a company that had spent seven years since its 2018 acquisition operating with significant independence, this was a real structural change. Freelance developer David Bushell put it starkly in his post: Microsoft has turned GitHub into an expensive liability.

The practical result is that there is no longer anyone at the very top of GitHub whose sole job is to be GitHub’s advocate against competing internal priorities. Every trade-off between reliability investment and Copilot growth now runs through Microsoft’s AI-first executive structure, and the AI-first structure has clear incentives to keep pushing agentic products at maximum velocity. This is not a conspiracy. It is just organizational math. If your reporting line goes to Microsoft’s CoreAI leadership, the CoreAI product roadmap will get more attention than the GitHub Actions capacity plan. And that is exactly what the outage data seems to be showing.

Fedorov’s April post did explicitly commit to reliability improvements, and Microsoft has both the capital and the engineering depth to deliver them. What is missing is a public commitment on when the developer community should expect visible improvement. Nine months into the crisis, that number still has not appeared in any Microsoft communication. Until it does, the working assumption inside most engineering leadership teams is that GitHub reliability is not going to snap back on any specific timeline they can plan around.

💡 Why the structure matters more than the memo. Fedorov’s April post was well-written and unusually candid. But without a dedicated GitHub CEO to defend developer-experience priorities against Copilot revenue pressure, the memo cannot translate into sustained execution.
📊 By the Numbers
💥
48
Major GitHub outages, May 2025 to April 2026
🤖
44
Copilot-specific outages, 9 of them major
📈
30X
Scale target admitted in CTO’s April 2026 memo
💰
$7.5B
Microsoft’s 2018 acquisition price for GitHub

The AI product Microsoft pushed hardest
is now what is breaking Microsoft’s most strategic developer platform.

The Copilot paradox, in one sentence
✅ What Serious Teams Are Actually Doing Right Now
  • 1. Adding a mirror on a second platform — GitLab, Gitea, and Forgejo are all seeing measurable inbound repo migrations. Even without a full exit, a mirror gives you optionality
  • 2. Caching Actions artifacts locally — Self-hosted runners with artifact caching removes the Actions single point of failure for critical CI pipelines
  • 3. Monitoring GitHub status independently — Do not trust the official status page. IncidentHub, StatusGator, and Marek Šuppa’s mirror all give a more complete picture
  • 4. Diversifying AI coding tools — Cursor, Claude Code, Windsurf, and Continue are the current alternatives seeing real professional adoption. Locking your team fully into Copilot is now a resilience risk, not just a vendor risk
  • 5. Building for graceful degradation — Assume Copilot will silently fail some percentage of tool calls. Recent Cloud Agent incidents showed 8% average and 26% peak error rates that were hard to detect
  • 6. Auditing dependency on GitHub-specific features — Anything that ties your workflow to Actions, Codespaces, or Copilot Cloud Agent is a reliability concentration risk right now
  • 7. Watching the July and August 2026 status page — The capacity buildout Fedorov described is a multi-quarter effort. Reliability should visibly improve if the plan is working. If it does not, that is signal

⚠️ What This Article Is Not Saying

1. This is not a prediction that GitHub will fail. Microsoft has the capital and the engineering depth to fix this if the executive attention holds. What it is saying is that reliability has already visibly slipped for over six months and the CTO has publicly conceded the scaling gap.

2. Copilot is not “bad software.” The point is that its usage pattern, autonomous coding agents operating continuously, generates traffic characteristics that GitHub’s platform was not architected to serve. That is an architectural fit problem, not a product-quality complaint.

3. Alternative platforms are not perfect either. GitLab has had its own reliability issues, and self-hosted Gitea requires operational overhead most teams cannot spare. Migration is not free. This is about optionality, not a wholesale exit.

4. SLA credits do not cover actual business cost. Standard Microsoft 365-style SLA credits are typically 25 to 100 percent of monthly service fees, not actual business impact. A nine-hour outage at a mid-sized SaaS company can cost hundreds of times what the credit returns.

✅ The Bottom Line

What to Take Away From the GitHub Crisis

1
The reliability decline is real and documented — 48 major outages in 12 months, per independent tracking, with an upward trend since December 2025
2
Copilot is the driver, and GitHub’s CTO said so — Agentic development workflows are generating load GitHub was not architected to serve, per the April 2026 memo
3
The 30X scale admission is the story — Not a routine capacity plan, an internal recognition that the original 10X plan was off by a factor of three
4
Mitchell Hashimoto’s Ghostty exit changed the discourse — Before it, this was inside-baseball. After it, leaving GitHub became a socially acceptable move
5
The governance structure is the deeper problem — With GitHub folded into Microsoft CoreAI and no dedicated CEO, developer-experience priorities compete against Copilot revenue at every trade-off
🔗 The Register’s February 2026 coverage (theregister.com) flagged that GitHub was struggling to hit three-nines availability before the story went mainstream, and remains a reliable primary source on the ongoing reliability data.
💬 Frequently Asked Questions
Q. Is GitHub actually going to fail?
No, that is not what the data suggests. Microsoft has both the capital and the engineering talent to work through this. What the data does suggest is that reliability has visibly declined for at least six months and the CTO has publicly conceded a large scaling gap. Failing to recover from that would be a strategic disaster for Microsoft, so the incentive to fix it is enormous. But whether the fix appears fast enough to hold the developer community’s confidence is a genuinely open question right now.
Q. Why is Copilot causing so many outages?
Not Copilot the AI model, but the usage pattern Copilot enables. Agentic development workflows generate continuous, automated pull requests, Actions runs, and API calls at volumes and cadences that human-driven workflows never approached. GitHub’s platform architecture was designed for humans clicking merge buttons, not autonomous agents making thousands of concurrent operations. GitHub’s own CTO named this shift as the primary driver of the load surge in his April 2026 blog post.
Q. What exactly did Mitchell Hashimoto say?
Two lines from his April 2026 blog post are being widely quoted. First: “This is no longer a place for serious work, if it blocks you out for hours per day, every day.” Second: “I want to ship software, and it doesn’t want me to ship software.” The second one is the more damaging framing because it suggests the platform’s own product direction is now working against the workflow it exists to serve. HashiCorp co-founder Mitchell Hashimoto pulled his Ghostty terminal project, which has more than 52,000 GitHub stars, off the platform.
Q. Should I move my repositories somewhere else?
For most teams, not a full migration, but yes to a mirror. Setting up an automated mirror on GitLab, Gitea, or a self-hosted Forgejo instance gives you a fallback if GitHub has a prolonged outage, without disrupting your day-to-day workflow. Full migrations are expensive and disruptive and usually not justified for a single service having a bad year. But mirrors are cheap insurance that a lot of engineering leaders are now quietly implementing.
Q. What does “30X scale” actually mean?
It is engineering shorthand for building the platform to handle thirty times the current traffic load. GitHub CTO Vlad Fedorov announced a 10X plan in October 2025 and revised it to 30X in February 2026 after seeing how fast agentic development workflows were accelerating. It matters because revising a scale target from 10X to 30X in four months is not a routine adjustment. It signals that GitHub’s leadership was surprised by how fast load was growing, and it implies the initial reliability plan was based on assumptions that no longer hold.
✍️
Editor’s Note. Sources for this article include GitHub CTO Vlad Fedorov’s April 2026 blog post, GitHub’s own status page and incident history archives, IncidentHub’s monitoring data for May 2025 through April 2026, StatusGator historical data, Marek Šuppa’s Missing GitHub Status Page mirror, Mitchell Hashimoto’s personal blog announcement, David Bushell’s “GitHub is Sinking” post, LeadDev’s May 2026 analysis, and The Register’s February 2026 reliability coverage. All figures are as reported by those sources and may be updated by the platforms themselves. This article does not offer investment advice and does not represent any position on Microsoft (NASDAQ: MSFT) securities.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top