GitHub, the world's largest source code hosting platform, said it had to add four million CPU cores and 160 PB of storage in just one year as the amount of AI-generated code skyrocketed.

The AI ​​boom is reshaping the way the world writes software. According to a new report from GitHub, the amount of code generated by AI has increased dramatically, forcing the source code hosting platform to add a huge amount of infrastructure resources to meet "unprecedented processing needs".

Specifically, to be able to store the incoming source code, GitHub is equipped with 4 million CPU cores, 160 petabytes of storage capacity and 50 Tb/s of network bandwidth. Activity on the repository recorded exponential growth year over year.

In particular, the amount of source code added or deleted (committed) increased 6 times, reaching a record of 7.4 billion commits per month in September, exceeding the 1.4 billion commits in April. The total number of requests to complete features or fix errors (pull requests) increased four times. The total number of code pushes in the past 12 months on GitHub reached 17.9 billion, surpassing the total number of 2024 and 2025 combined.

The cause is identified as the proliferation of AI agents. As of September, more than 40% of pull requests were generated by AI, up from 17% in March.

According to Tom's Hardware, the source code explosion also puts extreme pressure on CI/CD (continuous integration and delivery) systems. Every time a commit or pull request is executed, a series of automated tasks are triggered such as checking for errors, formatting the source code, and running integration tests to ensure the system is conflict-free.

Automation processes require huge amounts of computing power and data import and export speed. The pressure is so great that GitHub said it must redesign the entire backend system to separate the storage part from the calculation part.

With the current rate of development, experts predict that the platform will soon have to continue upgrading its infrastructure in the near future.

AI agents are considered to disrupt the software industry. In early October, Google announced the suspension of the OSS VRP open source software bug bounty program until at least 2027, after the system was overloaded with thousands of false reports generated by AI.

Google's team of engineers and open source project administrators were overwhelmed when they had to process thousands of reports. Most contain false information or result from AI illusions that cannot be exploited in reality. Spending too much time verifying error codes leaves the technical team with insufficient resources to focus on truly serious security issues.

In May, Linus Torvalds, the father of Linux, said the wave of duplicate vulnerability reports caused the internal security email list to become an "almost unmanageable" landfill.

A series of independent researchers run the same AI tool on a piece of source code, detect identical errors, and then massively send reports back. Because the list is kept private, the researcher cannot know whether the error has been reported by anyone or not. As a result, the team of engineers maintaining the Linux project spend a lot of time sorting emails and guiding senders to bug fixes that were updated weeks ago.

Huy Duc summary