GitHub powering up four million extra cores to stop its service going down

GitHub recorded 7.4 billion code commits in September 2026, five times the year-earlier level, as AI-agent traffic contributed to repeated outages; 90-day availability fell to about 84% on May 7. The company has put 4 million CPU cores and 160 petabytes of high-speed storage on its infrastructure roadmap and expects a redesigned architecture to enable 35x more throughput. GitHub also made Copilot Fusion generally available, reporting Fusion Max scored within two points of Opus 5.5 Max at half the cost and 40% less time per task, and announced additional Copilot and security tools.

Analysis

Investment read-through

This is a platform-quality and monetization test for Microsoft, not yet a material standalone earnings catalyst. Reliability failures threaten developer trust and could encourage enterprise customers to diversify workflows toward GitLab or other tools; remediation may protect retention more than create near-term growth. The infrastructure build also raises a utilization question: added capacity can improve service and support agent growth, but risks becoming costly ahead of monetization.

Copilot access expansion is a funnel strategy. Token-based billing preserves a usage revenue path, but broad access alone does not establish profitable adoption: verify paid usage, inference cost per completed task, and incremental Azure consumption. Model routing could lower cost per task and strengthen GitHub’s position, while intensifying price pressure on model providers and making raw token volume a poor proxy for economics. Security scanning and package controls may improve enterprise retention, but adoption and independently validated detection quality matter more than feature count.

Timing and risks: In the next 1–3 months, outages and Copilot usage disclosures are the actionable signals. The early-next-year multiregion migration and throughput redesign are 6–18 month execution catalysts, with delivery risk. Contrarian angle: the market may overvalue agent-driven code volume while underweighting reliability, inference costs, and whether customers will pay for outcomes. Conversely, successful routing could make usage growth cheaper than headline infrastructure needs imply. For MSFT, the effect is likely immaterial to consolidated results unless these signals change meaningfully.

Key Decisions for Investors

• No event-driven MSFT trade on the announcement alone; retain exposure only within the existing thesis, given the uncertain near-term earnings contribution.
• Set an alert for GitHub availability and migration milestones, alongside disclosed Copilot paid usage and inference economics. Consider adding MSFT exposure only if reliability improves and usage monetizes without a material cost burden.
• Treat repeated outages, migration slippage, or weak paid conversion despite rising usage as thesis-negative; these would indicate agent traffic is increasing operating burden without strengthening retention or revenue.
• Monitor competitive substitution toward GitLab and other developer platforms, plus model-provider pricing pressure. Reassess if enterprise customer behavior or Copilot unit economics show a durable change.