CNBC: GitHub strains under AI-coding pressure

CNBC published a detailed report Friday on how Microsoft’s GitHub has lost momentum in the AI-coding race. The core point is blunt: GitHub had distribution, Copilot moved early, and Microsoft owns both the developer platform and Azure. Still, outages, internal changes and fast-moving competitors such as Cursor and Anthropic’s Claude Code have weakened that advantage.

This is not just a Silicon Valley story about developer tools. For companies using AI agents in software engineering, GitHub has become part of the production line. When repositories, pull requests, build flows, security checks and agent suggestions stall, delivery slows as well.

CNBC says GitHub has suffered more than a dozen incidents lasting over an hour since March, citing GitHub’s status page. GitHub technology chief Vlad Fedorov wrote in March that the company had not met its own availability standards. At the same time, GitHub has been working through a long migration from dedicated infrastructure to Microsoft Azure. According to CNBC, that migration has constrained capacity and frustrated users.

The timing is bad. AI-assisted development shifts load from humans writing code at a slower rhythm to agents and assistants that can propose, revise and generate code much faster. The platforms underneath see more requests, more changes, more queues and more security checkpoints. Once AI coding becomes a standard workflow, the developer platform becomes critical infrastructure.

GitHub had home-field advantage

Microsoft bought GitHub for $7.5 billion in 2018. Since then, GitHub has grown sharply. CNBC notes that GitHub has six times more developers than when Microsoft acquired it, and that Satya Nadella said in October that GitHub was adding one developer every second and had reached 180 million developers.

Copilot gave Microsoft an early and visible grip on AI coding. But the market has moved quickly. CNBC points to Ramp data showing Cursor overtook GitHub Copilot in market share among its customers, and to a Jellyfish survey of 636 software professionals where Copilot was less widely used than Claude Code and Google Gemini Code Assist.

That does not mean Copilot is fading away. Nadella said in January that GitHub Copilot had 4.7 million paid subscribers, up 75% from a year earlier. But growth is not the same as control. When developers test Cursor, Claude Code and Gemini Code Assist alongside Copilot, GitHub becomes less automatic as the centre of AI development.

For CIOs, this is a procurement and governance issue. AI coding is no longer only about which model writes better code. It is about who controls the repository, who logs agent activity, who can roll back mistakes, who keeps the service available, and who carries the risk when an agent changes code close to production.

Downtime becomes AI risk

Many companies have treated GitHub as a convenient tool. That is no longer enough. When AI agents connect to codebases, issues, pull requests, CI/CD and secrets, the platform becomes a control point for both productivity and security.

CNBC describes how large customers, including Cisco, have been affected by GitHub availability problems. Cisco told CNBC it has fail-safes and runs enterprise versions of GitHub on its own servers. That detail matters. Large organisations are beginning to treat developer platforms the way they treat payment systems, ERP and identity: core operations, not IT support.

GitHub also disclosed a security incident this week. The company said an employee device was compromised and that the attacker obtained about 3,800 of GitHub’s own code repositories. Hogby.ai covered the security incident earlier. CNBC’s new report puts it into a wider context: GitHub is being pressured on availability, capacity, leadership and competition at the same time.

That makes the issue more serious. A single breach can be handled as incident response. Repeated instability in a platform carrying AI-assisted development has to be managed as supplier and operating risk.

What leaders should do now

The first step is to stop treating AI coding as a developer-only initiative. If Copilot, Claude Code, Cursor or Gemini Code Assist can access repositories, tickets, documentation or deployment workflows, they need to sit inside the same governance regime as other privileged systems.

The second step is to require an operating model. Which parts of the software-delivery flow stop if GitHub is down for two hours? Which teams can keep working locally? Which controls prevent agent-generated code from being merged without human review? Which logs show who, or what, proposed a change?

The third step is to test the exit plan before it is needed. CNBC writes that some customers are looking at alternatives such as GitLab, Bitbucket and offerings from Amazon and Atlassian. That does not mean every company should switch. It means boards and CIOs should know how hard a switch would be, which contracts lock them in, and which repositories, secrets and workflows would have to move if GitHub failed to perform.

The fourth step is to separate AI productivity from AI dependency. A coding agent that gives a team 20% faster development may still be expensive if it increases platform risk, security work and vendor lock-in. The real gain must be measured after operating cost: licence, controls, security, downtime, training, review quality and rollback.

Microsoft’s problem is also the customer’s problem

Microsoft is trying to organise its AI story around Copilot, Azure and agentic workflows. GitHub should be one of its strongest cards. When GitHub struggles with capacity and Copilot is pressured by newer competitors, customers get a clear signal: AI platforms that promise speed must also be assessed as mature operating suppliers.

For European and Norwegian organisations, the lesson is practical. AI agents in the codebase should not be rolled out without policies for access, logging, human review, continuity planning and supplier exit. Agents can write fast. That is not enough if the infrastructure around them cannot handle the pace.

Sources and media

• Primary source: CNBC, "Microsoft’s GitHub was positioned to win the AI coding race. Outages got in the way", published May 22, 2026 at 12:00 UTC: https://www.cnbc.com/2026/05/22/microsoft-was-positioned-to-win-in-ai-coding-outages-got-in-the-way.html
• Background: GitHub, "An update on GitHub availability": https://github.blog/news-insights/company-news/an-update-on-github-availability/
• Background: GitHub, "Addressing GitHub’s recent availability issues": https://github.blog/news-insights/company-news/addressing-githubs-recent-availability-issues-2/
• Background: GitHub Security, "Investigating unauthorized access to GitHub’s internal repositories": https://github.blog/security/investigating-unauthorized-access-to-githubs-internal-repositories/
• Thumbnail: OpenAI Image 2 / hogby.ai.