GitHub added 4 million CPU cores, 160 petabytes of storage and 50 terabits per second of networking capacity this year to absorb a flood of AI-generated code that has rewritten the platform’s traffic profile. The infrastructure spend is the clearest signal yet that agentic coding is not a feature but a new baseline for software production.
Commits are growing exponentially
Raw repository activity accelerated on a curve that defies linear planning. Code commits grew sixfold year over year and merged pull requests quadrupled. The monthly run rate tells the sharper story: 1.4 billion commits in April, 2.9 billion in August and 7.4 billion in September. GitHub recorded 17.9 billion code pushes in the trailing 12 months, a total that exceeds the combined volume of 2024 and 2025.
Agents now open the majority of pull requests
By September, more than 40 percent of pull requests were initiated by an agent, up from 17 percent in March. Repositories that start with AI generate roughly eight times the activity of handcrafted projects. The shift means every merge triggers a cascade of linting, testing and integration checks that now runs at a scale the original architecture was not built to sustain.
CI/CD pipelines absorb the real cost
Each commit and merge fires off continuous-integration workloads that check for leftover to-do items, enforce style consistency and run test suites against dependent services. The compute and I/O demand from that loop forced GitHub to reengineer its backend, decoupling storage from compute to keep latency in check. With the current trajectory, the next capacity upgrade is already visible on the horizon.