Microsoft confirmed on June 16, 2026 that GitHub is no longer running on Azure alone. Facing a wave of reliability problems tied to AI coding agents, including a roughly 2.4-hour outage on November 28, 2025 that left GitHub Copilot’s Claude Sonnet 4.5 model returning HTTP 400 errors, and, more recently, a 7-hour-47-minute outage on August 17, 2026 that GitHub itself said began when a new traffic peak overran a Central US component, disrupting github.com, Actions, APIs, pull requests, Issues, and Copilot, with web and API error rates peaking near 20% and archive and raw-content downloads seeing roughly 50% failure rates, the company has been renting spare capacity from Amazon Web Services, its biggest cloud rival, just to keep the world’s largest code-hosting platform online. The admission lands two months after a separate shareholder lawsuit accused Microsoft of downplaying how much Azure’s growth had slowed. Together, the two stories, now compounded by the August outage, turn a quiet capacity problem into a genuine competitive embarrassment, and a case study in how AI traffic is outrunning cloud planning across the industry.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Microsoft Confirms It Is Renting AWS Capacity to Keep GitHub Online
The story broke on June 16, 2026, when Microsoft acknowledged that GitHub now runs across more than one cloud provider rather than solely on Azure, as first detailed by Cloud Computing News and later covered by TechRadar Pro. A company spokesperson pointed to a sharp rise in what Microsoft calls “agentic development,” meaning AI agents that write, test, and commit code largely on their own, as the trigger. The surge, which the company says began in late 2025, pushed traffic past what GitHub’s existing Azure footprint could absorb on the original timeline.
Microsoft frames the AWS arrangement as deliberate strategy rather than an emergency patch. The company says it is simultaneously accelerating GitHub’s move to Azure and keeping a multi-cloud option open to guarantee elasticity during traffic spikes. In practice, that means Amazon’s data centers are now handling overflow load for a developer platform Microsoft has owned since 2018.
For a company that markets Azure as the natural home for GitHub, Copilot, and the rest of its developer stack, sending traffic to AWS is a hard fact to spin. Reporting from Tech Times and Gigazine frames the arrangement bluntly: Azure could not absorb GitHub’s growth fast enough while its underlying architecture was still being rebuilt for the migration. AWS became the release valve.
Nine Outages in a Month: How Bad Did GitHub’s Reliability Get
The numbers behind Microsoft’s decision are stark. Between May 2025 and April 2026, GitHub racked up 257 separate outages, including 48 classified as major incidents, for a combined 112 hours and 18 minutes of downtime, according to a May 2026 analysis by TalonX, a tally an independent outage-history analysis from IncidentHub corroborated for that same 12-month window. A separate window tracked by monitoring service Vantaj found the pattern hadn’t let up: from May 27 to June 26, 2026, GitHub logged 25 incidents totaling roughly 500 minutes of issues, working out to about 98.8% availability. By June, the platform’s official availability had slipped below 99%, a threshold that matters enormously for a service hundreds of millions of developers treat as always-on infrastructure.
Ninety-nine percent uptime sounds close to reliable until it gets converted into hours, and independent reconstructions suggest the real picture was worse, even though GitHub’s own official status disclosures are far more conservative: the company logged just 10 incidents in April 2026 and 9 incidents in May 2026 that it classified as degrading service. GitHub’s own incident history, as tallied by The Register, shows 26 incidents in April 2026, 23 in May 2026, and 12 more by mid-June 2026, a pace that helps explain why reconstructed uptime fell below 85% in April 2026, with some months dipping under 80%. An unofficial tracker called Missing Status Page counted 12 incidents in May 2026 alone and put GitHub’s 90-day uptime at 87.26% by June 2026, with its April-only figure at just 78.33%. Even at the official below-99% mark, that works out to more than seven hours of downtime or degraded service per month, spread across pull requests, Actions runs, package registries, and API calls that modern CI/CD pipelines hit every few minutes.
The Uptime Numbers Behind the Headlines
GitHub’s own status history, published at githubstatus.com, has become required reading for engineering teams trying to time deployments around instability, and the record already showed warning signs well before 2026’s crunch: GitHub’s January 2025 availability report disclosed that a bad deployment pushed error rates as high as 6.85% for 30 minutes, and a separate caching hardware failure on January 30, 2025 drove web request error rates up to 44% for 26 minutes. February 2025 brought more of the same: GitHub logged two incidents that month, including a 2.3-hour delay in incident notifications on February 25, 2025, the same day a separate 51-minute incident pushed failure rates for GitHub Packages uploads and downloads to 14%, while a February 3, 2025 migration issue had already caused a full, 100% outage for all repository migrations for roughly 30 minutes. A year later, the pattern had not eased: GitHub’s own February 2026 status report disclosed six incidents that month, including one on February 9, 2026 that simultaneously degraded github.com, the API, Actions, and Copilot. Teams that once treated GitHub like a utility, similar to electricity or water, are now building in retries, fallbacks, and mirrored repositories, because one degraded incident can stall an entire release pipeline. The table below lays out the core numbers behind the capacity crunch.
| Metric | Figure | Period |
|---|---|---|
| Service-degrading incidents | 9 | May 2026 |
| Platform availability | Below 99% | June 2026 |
| Weekly commit volume (reported) | ~275 million | 2026 |
| Projected annual commits (reported) | ~14 billion | Full-year 2026 |
| GitHub Actions usage | ~1 billion minutes/week | 2025, up from ~500 million in 2023 |
| Monolith traffic migrated to Azure | 40% | May 2026 |
| Azure migration target | 50% | July 2026 |
The Root Cause: AI Coding Agents Overwhelmed GitHub’s Infrastructure
GitHub’s traffic problem is not generic growth. It is a specific byproduct of the AI coding boom reshaping how software gets written in 2026, and the strain shows up in concrete incidents: GitHub’s own March 2026 status disclosures counted just four incidents that month causing degraded performance across core services, but independent research firm ByteIota counted eight major outages across February and March 2026 alone, missing GitHub’s own 99.9% availability target, including a stretch on March 19-20, 2026 when core Git operations were down for a full eight hours. That was compounded by a March 3, 2026 incident in which GitHub’s own report logged GitHub.com failure rates reaching 40% and Copilot error rates hitting 21% between 18:46 and 20:09 UTC. The strain was not evenly spread: TalonX’s analysis of the May 2025-April 2026 window found GitHub Actions alone responsible for 57 outages, 16 of them major, with Copilot logging 44 separate disruptions over the same stretch. GitHub’s own leadership has reportedly told press that weekly commit volume has climbed to roughly 275 million, putting the platform on pace for close to 14 billion commits this year, up sharply from an estimated 1 billion in 2025, according to Gigazine’s coverage of earlier Business Insider reporting. GitHub Actions usage tells a similar story: weekly minutes consumed on the platform’s automation service roughly doubled from about 500 million in 2023 to 1 billion by 2025.
Autonomous coding agents do not behave like human developers. They commit more often, trigger more automated test runs, and call GitHub’s API on a tighter loop, often around the clock rather than during one time zone’s working hours. The Copilot Coding Agent itself has felt that strain directly: GitHub’s March 2026 report disclosed two separate outages that month in which datastore authentication issues drove Copilot Coding Agent error rates to 100%. That pattern turns what used to be predictable, business-hours traffic into a sustained, always-on load. Multiply that by every team that adopted an agentic coding tool since late 2025, and the traffic curve stops looking like organic growth. It starts looking like a step change capacity planners never modeled.
Why Azure Could Not Absorb the Overflow
Microsoft has spent years moving GitHub off the patchwork of infrastructure it inherited at acquisition, shifting toward Azure as the permanent home. By May 2026, that migration had reached 40% of GitHub’s core “monolith” traffic, with Microsoft targeting 50% by July 2026, per the Cloud Computing News report.
The trouble is timing. Migrations of this size are usually sequenced around predictable growth curves, not sudden demand spikes. GitHub’s architecture is being rebuilt for Azure at the same time agentic traffic is climbing far faster than the original schedule assumed. That combination left a gap: Azure’s GitHub-specific capacity was sized for a migration timeline set years earlier, not for a 2026 traffic profile reshaped by autonomous coding agents.
Renting AWS capacity solved the immediate problem without forcing Microsoft to slow the migration or overbuild Azure capacity for a demand curve nobody is certain has plateaued yet. It is a stopgap, but one that buys time while the harder architectural work continues underneath it.
Inside Microsoft’s Multi-Cloud Statement
Microsoft has not tried to hide the arrangement, though it has also stopped short of confirming AWS by name in every version of its public comments. According to the company’s statements, relayed by Cloud Computing News and TechRadar Pro, Microsoft describes the move as part of accelerating its shift to Azure while continuing to explore a multi-cloud approach that adds future capacity, compute elasticity, and horizontal scale.
Read closely, that language does two things at once. It reassures investors that Azure remains the strategic destination, and it quietly concedes that Azure alone was not enough to absorb 2026’s traffic curve. Companies rarely volunteer that their flagship cloud platform needed a rival’s help. Microsoft framing it as strategy, rather than shortfall, says a lot about how sensitive the optics are heading into a summer already crowded with cloud-spending scrutiny.
The Irony: Microsoft’s Biggest Cloud Rival Becomes Its Safety Net
AWS and Azure have spent a decade positioning themselves as opposites in enterprise pitches, competing for the same workloads, the same government contracts, and the same AI-training deals. Coverage of this story from Koda Community put it in blunt terms: Microsoft is effectively paying its biggest rival to keep its own developer platform online.
That framing matters beyond its headline value. GitHub is not just another SaaS product Microsoft happens to own. It is the default home for a huge share of the world’s open-source code, the distribution point for GitHub Copilot, and increasingly the execution layer for autonomous coding agents built by Microsoft and its competitors alike. Every hour of degraded GitHub service ripples into delayed releases at companies that have nothing to do with the Microsoft-AWS rivalry. The dependency runs both directions too: AWS now earns revenue for keeping a rival’s flagship product running, a detail unlikely to fade quietly from either company’s competitive messaging.
A Second Front: The Azure Shareholder Lawsuit
The capacity story did not emerge in isolation. Reuters reported on June 15, 2026 that Microsoft shareholders had filed a class action alleging the company misled investors about Azure’s growth trajectory and the scale of AI infrastructure spending required to sustain it. The proposed class period runs from May 1, 2025 to January 28, 2026.
The core allegation is straightforward. Plaintiffs say Microsoft did not adequately disclose that Azure’s growth was decelerating at the same time capital spending on AI infrastructure was climbing sharply, and that the combination artificially inflated the stock before the market caught up. The case remains at an early stage, and Microsoft has not been found liable for anything. But the timing is awkward. A lawsuit about whether Microsoft was transparent about Azure’s capacity to support growth lands the same month the company confirms one of its own core products needed help from a rival cloud just to stay online.
Neither story proves the other. A shareholder disclosure dispute is a legal and financial question, while GitHub’s outages are an operational one. Both point to the same underlying tension, though: Azure’s growth story and Azure’s operational capacity have not always moved in lockstep, and 2026 is the year that gap became visible from multiple directions at once.
The UK’s Separate $2.8 Billion Cloud Licensing Case
Microsoft is also fighting a second, unrelated legal battle that adds to the pile. Reuters reported on April 21, 2026 that Microsoft must face a $2.8 billion lawsuit in the UK over software licensing charges applied to customers running Microsoft software on rival cloud platforms, including AWS, Google Cloud, and Alibaba Cloud.
That case predates the GitHub capacity story by nearly two months and rests on a different legal theory entirely, centered on licensing terms rather than disclosure obligations. It reinforces the same pattern from another angle: Microsoft’s relationship with rival clouds is under scrutiny on several fronts at once, from regulators, from customers, and now from its own shareholders. The table below separates the two cases, since they are easy to conflate.
| Attribute | Shareholder Class Action | UK Licensing Case |
|---|---|---|
| Reported by | Reuters, June 15, 2026 | Reuters, April 21, 2026 |
| Core allegation | Misleading disclosure on Azure growth deceleration and AI capex needs | Software licensing charges applied on rival cloud platforms |
| Named rival clouds involved | Not applicable | AWS, Google Cloud, Alibaba Cloud |
| Amount at stake | Not yet quantified | $2.8 billion |
| Class period / scope | May 1, 2025 to January 28, 2026 | Not a class action |
| Status as of June 2026 | Early stage | Microsoft ordered to face the claim |
AWS vs Azure vs Google Cloud: Who Actually Has the Capacity Edge
Strip away the GitHub-specific drama and the underlying capacity question applies to the entire hyperscaler market. AWS remains the largest cloud infrastructure provider by share, with Azure in second place and Google Cloud a growing third, according to 2026 market-share estimates. Combined, the three companies account for roughly two-thirds of global cloud infrastructure spending, a concentration that leaves enterprise customers with limited real alternatives when their primary provider hits a wall.
Where the Capacity Gap Really Comes From
Synergy Research Group has tracked the industry’s combined annual cloud infrastructure revenue run rate crossing the half-trillion-dollar mark in early 2026, driven heavily by AI-related workloads rather than traditional enterprise migration. That growth is exactly what is straining capacity planning across the board, not just at Microsoft. AWS has its own history of regional capacity constraints during demand surges, and Google Cloud has spent the past two years building out regions specifically to compete on availability rather than price alone. GitHub’s overflow to AWS is a visible, named example of a dynamic every hyperscaler is quietly managing: promising customers elastic, always-available capacity while AI workloads scale faster than any single company’s build-out schedule.
| Provider | Est. Global Cloud Market Share (2026) | Recent Signal | Capacity Note |
|---|---|---|---|
| Amazon Web Services | ~31% | Now hosting overflow GitHub traffic for rival Microsoft | Longest-established hyperscale footprint of the three |
| Microsoft Azure | ~25% | 40% of GitHub monolith traffic migrated as of May 2026 | Migration schedule outpaced by AI-agent traffic growth |
| Google Cloud | ~11% | Continues aggressive region build-out | Smallest of the big three, among the fastest-growing by percentage |
| Oracle, Alibaba and others | ~33% combined | Picking up sovereign-cloud and overflow demand | Beneficiaries when big-three customers hedge their capacity risk |
| Combined big three | ~66-67% | Concentration limits real alternatives for enterprise buyers | All three report AI-driven capacity pressure in 2026 |
From a $7.5 Billion Acquisition to an Overflow Tenant: GitHub’s Azure Timeline
Microsoft bought GitHub in 2018 for $7.5 billion in stock, a deal that at the time raised real questions about whether the notoriously platform-agnostic developer community would tolerate Microsoft ownership. The company spent years insisting GitHub would keep running independently of Azure’s commercial pressure, all while quietly working toward deeper technical integration.
The long-term plan always pointed toward Azure. Reporting on the current capacity crunch notes that Microsoft had previously targeted having GitHub run almost entirely on Azure by 2027, a target now expected to slip given how much faster demand grew than planned. Eight years after the acquisition closed in October 2018, GitHub is simultaneously Microsoft’s flagship proof that the deal worked, its most visible AI distribution channel for Copilot, and, this month, a tenant renting spare room from Amazon. That is not the arc Microsoft envisioned at closing, and it is a reminder that owning a platform is not the same as controlling its infrastructure destiny on a fixed schedule.
What This Means for Enterprise Cloud Buyers
For engineering leaders who do not care about the Microsoft-AWS rivalry, the practical takeaway is simpler: even the biggest platforms are not immune to capacity shocks, and single-cloud dependency carries real operational risk. Teams that build CI/CD pipelines, release schedules, and incident response entirely around one provider’s assumed reliability are exposed when that provider hits an unplanned wall.
The GitHub situation also validates a trend enterprise procurement teams have pushed for two years: negotiating multi-cloud terms even when a company has a clear primary vendor. If Microsoft, with all its Azure investment and internal alignment, still needed AWS as a release valve, smaller companies with far less negotiating leverage have even less room to assume their primary cloud will always absorb a demand spike gracefully.
There is a cost angle too. Renting capacity from a direct competitor is rarely the cheapest option available, which suggests Microsoft judged reliability risk to GitHub’s reputation as more expensive than whatever premium AWS charged for the overflow. For FinOps teams already stretched by AI-driven cloud bills, that trade-off, paying more to guarantee uptime rather than risk a public reliability failure, is worth studying rather than dismissing as a one-off.
The Bigger Pattern: AI Demand Is Breaking Capacity Planning Industry-Wide
GitHub is the most visible example this month, but it will not be the last. Autonomous coding agents, AI search assistants, and agentic workflows all share a trait: they generate machine-speed traffic patterns that do not resemble the human usage curves most infrastructure was originally sized around. A support-ticket system built for humans typing at human speed behaves very differently once an AI agent starts filing, reading, and resolving tickets in a tight automated loop.
Flexera’s 2025 State of the Cloud Report already found that 94% of enterprises run workloads in the cloud and that security, not capacity, was the top concern cited by 84% of respondents. That ranking may not hold for long. As agentic tools multiply across coding, customer support, and data pipelines through the rest of 2026, capacity planning is likely to climb back up every CIO’s risk list. The GitHub episode shows that even a company with Microsoft’s balance sheet and infrastructure expertise can get caught behind the curve, and that broader concentration in the cloud computing market means fewer places to hide when one provider stumbles.
What Developers and DevOps Teams Should Watch For
Practically, teams relying on GitHub for mission-critical CI/CD should treat the current stabilization period as ongoing, not resolved. Recent incidents underscore why: on May 30, 2025, the Microsoft Teams-GitHub integration suffered a full, 100% outage lasting from 08:10 to 16:00 UTC, on October 29, 2025, Codespaces error rates averaged 90% and peaked at 100% during a third-party outage, and on November 17, 2025, roughly 57% of Dependabot jobs failed to meet their service-level objective because of GitHub Container Registry rate limits. A few concrete steps help now. First, monitor GitHub’s own status page rather than assume silence means stability, since incident disclosure matters more with availability sitting below 99%. Second, build retry logic and timeout tolerance into pipelines that call GitHub’s API, Actions runners, Codespaces, or Dependabot, instead of treating every call as guaranteed to succeed on the first attempt. Third, consider mirroring critical repositories to a secondary host as a fallback for release-critical windows.
A simple way to script a status check into a deploy pipeline looks like this:
curl -s https://www.githubstatus.com/api/v2/status.json | \
python3 -c "import json,sys; d=json.load(sys.stdin); print(d['status']['description'])"Wiring a check like that into a pre-deploy step costs almost nothing, and it gives a release pipeline the chance to pause automatically rather than fail mid-deploy during a degraded window.
Five Predictions for the Next Phase of This Story
- The 50% Azure migration target for July 2026 slips. Given the original goal was full migration by 2027 and progress already trailed the traffic curve, expect Microsoft to quietly push the 50% monolith-traffic milestone into the third quarter rather than hit it on schedule.
- The AWS arrangement outlasts its “stopgap” label. Unwinding a working dual-cloud setup is operationally risky and commercially awkward to do quickly. Expect Microsoft to keep some AWS overflow capacity active well into 2027, even after public messaging shifts back to an Azure-first narrative.
- More hyperscaler overflow deals surface publicly. GitHub’s arrangement became public because Microsoft confirmed it. Other AI-heavy platforms are almost certainly managing comparable overflow deals quietly. Expect at least one more major platform’s multi-cloud overflow arrangement to come to light before the end of 2026.
- Shareholder litigation over AI-capex disclosure spreads beyond Microsoft. With capital spending climbing across every major cloud provider, the legal theory behind Microsoft’s shareholder suit, that growth deceleration and AI spending needs were not clearly disclosed, is reusable against Google or Amazon if either reports a similar growth-versus-capex mismatch.
- GitHub introduces agent-specific rate limits or tiered SLAs. Since autonomous coding agents are the specific traffic driver, expect GitHub to eventually separate agent traffic from human developer traffic with its own throttling rules, rather than solving the problem purely by adding capacity.
Frequently Asked Questions
Why is Microsoft using AWS to run parts of GitHub?
Microsoft says a surge in AI-driven “agentic development” since late 2025 pushed GitHub’s traffic past what its Azure footprint could absorb on schedule, so the company is renting AWS capacity as overflow while the Azure migration continues.
How many outages has GitHub had in 2026?
Between May 2025 and April 2026, GitHub logged 257 outages, including 48 major incidents totaling 112.3 hours of downtime, according to a May 2026 TalonX analysis, and its availability had fallen below 99% by June 2026, according to reporting cited by Cloud Computing News and Tech Times.
Is GitHub moving away from Azure permanently?
No. Microsoft describes AWS as a stopgap. The Azure migration continues, with 40% of monolith traffic moved by May 2026 and a 50% target set for July 2026, on the way to Microsoft’s original goal of near-complete Azure migration by 2027.
What is the Microsoft shareholder lawsuit about?
Reuters reported on June 15, 2026 that shareholders filed a class action alleging Microsoft misled investors about Azure’s growth deceleration and the scale of AI infrastructure spending needed, covering a proposed class period from May 1, 2025 to January 28, 2026.
Is the shareholder lawsuit connected to the GitHub capacity story?
Not legally. They are separate matters, one operational and one financial, but both point to the same underlying tension between Azure’s growth narrative and its real-world operational capacity in 2026.
What is the separate $2.8 billion UK lawsuit about?
Reuters reported on April 21, 2026 that Microsoft must face a $2.8 billion UK claim over software licensing charges applied to customers running Microsoft software on rival clouds, including AWS, Google Cloud, and Alibaba Cloud. It is a different case with a different legal theory.
Will this affect everyday GitHub or Copilot users?
Mostly behind the scenes, but it explains recent reliability issues. Developers with release-critical pipelines should monitor GitHub’s status page and add retry logic rather than assume every API call will succeed on the first try.
Which cloud provider has the most capacity, AWS or Azure?
AWS remains the largest by estimated global market share at around 31%, with Azure near 25% and Google Cloud around 11%. Capacity pressure from AI workloads has hit all three providers in 2026, not just Microsoft.
