AI Coding Demand: From Nice-to-Have to Infrastructure Crisis
AI infrastructure scaling pressure describes the rapid, unexpected growth in AI-driven workloads that overwhelms existing cloud, storage, and networking capacity, forcing platforms to rethink architectures, demand forecasts, and capital plans. GitHub is now a clear example of this shift. Once a code-hosting utility, it has become a live execution surface for AI-assisted coding and agentic development. In 2025 it reportedly handled 1 billion commits over the entire year; now it receives 1.4 billion commits every month, with internal forecasts pointing to 14 billion commits in 2026. Each AI agent session multiplies background tasks: repository reads, test runs, pull requests, and code reviews queue up on the same systems that human developers depend on. The result is visible strain, with outages, degraded performance, and stop‑start feature rollouts revealing how far real‑world AI deployments are outrunning traditional capacity planning and cloud infrastructure limits.
GitHub Capacity Issues Show Forecasts Are Broken
GitHub’s own reports show how badly its planning missed the AI moment. In October 2025, the company aimed for a 10x capacity increase. By February 2026, data from AI-assisted coding forced a rethink: managers concluded they needed a 30x expansion instead to stay ahead of pull requests, commits, and new repositories. The platform has more than doubled effective capacity in four months, with 40 percent of monolith traffic and 30 percent of Git traffic now running on Azure, and repository replication at 99 percent. Yet availability remains uneven, with nine incidents degrading performance in May alone and unofficial tracking estimating uptime as low as 78.33 percent in April. For developers, these GitHub capacity issues are not theoretical. Slow clones, stuck CI pipelines, and delayed code reviews all stem from the same root cause: AI demand forecasting was far too conservative.
When One Cloud Is Not Enough: Turning to AWS
The most striking signal that AI workloads are escaping early plans is Microsoft’s decision to bring Amazon Web Services into the picture. Business Insider reported that GitHub, despite being owned by Microsoft and mid‑migration to Azure, is adding extra computing capacity through AWS after AI-driven coding activity strained its core infrastructure. According to the report, Microsoft had hoped to move GitHub fully onto Azure by 2027 but now expects to keep using multiple cloud providers where it needs more elasticity. The backdrop is a wave of outages: GitHub itself counted nine incidents in May that degraded service. The move underlines that cloud infrastructure limits are practical, not political. Azure remains the strategic favorite, but GitHub needs capacity where and when AI traffic hits, even if that means calling a rival to keep repositories, pull requests, and AI agents online.
Agentic Workflows Are Redefining Infrastructure Load
The strain is not only about more users; it is about new behavior. Early AI coding tools mostly suggested lines inside an editor. Today’s agentic workflows are different. GitHub Copilot and newer agents from providers like Anthropic and OpenAI can read whole repositories, plan tasks, modify files, run tests, and open pull requests without constant human input. Each of those actions triggers extra compute, storage, and networking demands on GitHub’s platform. This is why outages hit harder: when the site slows down, both humans and AI agents stall in the same queues. GitHub’s engineering leaders say they are isolating critical database components—such as users, authentication, and authorization—to prevent cascading failures. But as long as AI systems are actively working on code inside the platform, capacity planning has to account for autonomous multi‑step sessions, not just human commits and occasional autocomplete queries.
Economics and the Future of AI Infrastructure Scaling
GitHub’s infrastructure headaches are mirrored in its pricing changes. Copilot plans have moved to usage-based billing tied to token consumption through GitHub AI Credits, replacing a simpler request-based model. Base prices remain at USD 10 (approx. RM46) a month for Copilot Pro, USD 19 (approx. RM87) per user a month for Business, and USD 39 (approx. RM178) per user a month for Enterprise, but the company has warned that long, multi‑hour agent sessions cannot be treated like quick chat prompts. Some users have already shared screenshots of projected bills hundreds of dollars higher than before, prompting backlash and closer scrutiny of how they use AI tools. Microsoft, meanwhile, still projects USD 190 billion (approx. RM868 billion) in capital expenditures for 2026, much of it for data centers. The gap between that headline figure and GitHub’s day‑to‑day strain shows that building enough capacity, in the right form and at the right time, is the real AI infrastructure scaling challenge.
