GitHub now processes 2.9 billion commits every month, more than double the 1.4 billion it handled in April, according to an August analysis published on The New Stack by Signadot. The surge overwhelmed the platform's infrastructure on August 17, forcing a nearly eight-hour outage after a core component in GitHub's Central US data center couldn't scale with the traffic. The report argues that the real problem isn't GitHub's capacity crisis but the widening gap between how fast code is now generated and how slowly it's verified.
Alongside the commit spike, GitHub reported roughly 130 million merged pull requests and 24 million new repositories each month, the analysis notes. GitHub attributed the explosion to AI-generated code, with developers now running multiple coding agent sessions in parallel, producing commits far faster than any hiring forecast anticipated. To handle the load, GitHub deployed more than 3 million additional CPU cores, 120 petabytes of high-speed storage, and accelerated its migration to Azure, which now serves 58% of platform traffic. The commit growth curve shifted from tracking human developers to tracking machine output, breaking years of steady, predictable expansion.
"Code generation has become machine-paced, and its volume curve is exponential," the report states. "Verification is still human-paced, and its capacity curve is close to flat." According to the analysis, every commit carries an implicit claim that the change works, but almost nothing in the system that produced and merged those commits checks that claim against a running system. The report warns that the distance between generation speed and verification capacity is "the defining infrastructure problem of the next three years." AI code review tools and faster CI pipelines have improved triage and static analysis, but the step that verifies behavior against a live system still funnels through shared staging environments or expensive full-stack duplicates that can't scale at the same rate.
The report explains that verification infrastructure was designed when code arrived at the pace humans type, making staging contention and review queues manageable. Agents didn't create the bottleneck—they multiplied it past the point where old strategies work and brought it to smaller teams. Throughput that once strained 500-engineer platforms now appears on 50-engineer teams running agents in parallel. Teams respond by speeding up reviews, throttling agents to protect the verification queue, or merging unverified changes and absorbing downstream failures. The analysis identifies a structural fix: making verification parallel and per-change by deploying only the services a change touched in a shared environment, so each change gets isolation where it matters while sharing the rest of the infrastructure.
The report concludes that generation is now effectively free and unlimited, so it's no longer the source of advantage. "The teams pulling ahead are not the ones producing the most commits," it states. "They are the ones whose verification capacity rises with their generation capacity." Organizations should expect commit volume to keep doubling, turning more generated code into shipped code rather than longer review queues, deeper staging backlogs, and bigger incident bills. The analysis frames the gap between generation and verification curves as urgent: teams need to close it before the curve doubles again. The shift from human-paced to machine-paced development demands infrastructure that scales verification as aggressively as it scales generation, or the productivity gains from AI coding agents will be lost to bottlenecks that were never designed for this volume.