AI coding boom puts pressure on infrastructure as platforms expand capacity

AI coding agents are driving a surge in code commits, pushing GitHub to expand computing, storage and network capacity to handle growing software development workloads

Artificial intelligence (AI) coding agents are increasing the volume of code being written, tested and reviewed, creating fresh infrastructure demands for software development platforms. As developers use these tools to write code, fix bugs and handle multiple programming tasks simultaneously, platforms need more computing power, storage and network capacity to manage the activity.

 

GitHub, the Microsoft-owned software development platform, in September, said that the platform recorded 7.38 billion code commits, more than five times the number reported a year earlier. This increase is linked to the growing use of AI agents to write code and fix bugs. As more code is generated, GitHub needs additional computing power and storage to store it, test it and check for errors. 

 

 

GitHub said that it added 4 million CPU cores, 160 petabytes (PB) of high-speed storage and 50 terabits per second (Tb/s) of network capacity in 2026 to handle the growing volume of code.

 

AI agents are increasing the volume of code changes

 

The rise in code commits is closely linked to the way developers are using AI agents. Traditionally, developers wrote code, tested it and submitted changes for review. AI coding assistants initially helped with smaller tasks, such as suggesting lines of code, completing functions or identifying errors. The newer generation of AI agents can take instructions, work across multiple files, make changes, run tests and submit code for review with less human intervention.

 

This allows developers to assign different tasks to multiple agents at the same time. For instance, one agent can work on fixing a software bug while another develops a new feature and a third checks existing code. While this can reduce the time required to complete individual tasks, it also increases the number of changes being submitted to a shared software repository.

 

GitHub's September figures explain this change. According to its engineering update, code pushes increased 4.9 times year-on-year to 3.35 billion a month. Pull-request merges, through which developers combine proposed code changes with the main project, also rose to nearly four times their year-earlier volume. These figures indicate that the increase is not limited to code being generated. More changes are also moving through the processes used to integrate software. 

 

Why more code requires more computing power and storage

 

The increase in software development activity creates additional work for the infrastructure supporting platforms such as GitHub. Every time a developer or an AI agent submits code, the platform must store the changes, make them available to other users and ensure that different versions of a project remain consistent.

 

A code push also triggers several activities that require computing resources. These can include running automated tests, checking whether the new code affects existing features, scanning for security vulnerabilities and verifying whether the changes can be integrated with the rest of the project.

 

GitHub said each push can generate thousands of additional reads by continuous integration (CI) and code-scanning systems. CI refers to the process through which software changes are automatically built and tested before they are accepted into a project. These systems repeatedly access the submitted code to check whether it works as expected.

 

The scale of this activity is reflected in GitHub Actions, the platform's service for automating software development workflows. It ran 3.26 billion times in September, more than four times the number recorded a year earlier.

 

This means the infrastructure has to handle both the initial code submission and the additional activity triggered by it. When thousands of AI agents work simultaneously, the number of requests made to a repository can rise sharply. The platform must ensure that these requests do not slow down other developers working on the same project.

 

The challenge is not only storing more code. It is also making sure that the code can be accessed, tested and updated quickly, even when several agents and automated systems are working on it at the same time. 

 

Why GitHub is rebuilding its infrastructure

 

The increase in activity has also exposed limitations in the way traditional software repositories are designed to handle large volumes of simultaneous changes.

 

In its October 2026 engineering update, GitHub said its existing architecture was designed to manage repository data through multiple copies stored across different file servers. This approach helps keep data available and allows several users to access the same repository.

 

However, as the number of code submissions increases, maintaining these copies can become more complicated. GitHub explained that every additional replica involved in a write operation can add to the time required to complete it. In other words, adding more capacity to handle incoming requests can also create additional work when new code has to be stored and made available consistently.

 

This becomes a problem when thousands of agents are making changes to the same repository. Unlike a human developer, who may submit a few changes during a working session, an AI agent can repeatedly update and save its progress while completing a task.

 

GitHub said it is rebuilding its Git infrastructure to address these demands. Its approach includes reducing unnecessary coordination between systems, separating storage from computing resources and allowing more operations to take place simultaneously.

 

The company is also working to move heavy maintenance tasks, such as repository data clean-up and compaction, away from the systems that handle live requests. This is intended to prevent background maintenance work from slowing down code submissions and downloads.

 

Code review and security checks face a larger workload

 

The increase in code generation also has implications for software testing and security. More code submissions mean more changes that developers and automated systems have to examine before software is released. AI-generated code can contain errors, introduce compatibility problems or create security weaknesses. Even when an AI agent completes a task successfully, its output still needs to be checked against the requirements of the software project.

 

This makes code review an important part of the changing development process. Developers have to assess whether a proposed change works, whether it affects other parts of the application and whether it introduces vulnerabilities.

 

The challenge is particularly relevant when several agents are working on different parts of the same project. Changes made by one agent may affect the work being carried out by another. Developers therefore need to ensure that these changes can be combined without creating conflicts or breaking existing features.

 

GitHub's October 28 announcement also acknowledged the growing demand for code reviews. The company said that more code means more pull requests and a greater need for review. It introduced Copilot code review Max, a more detailed review option designed to identify significant bugs that developers should examine before changes are accepted.

 

However, automated review tools do not remove the need for human oversight. Developers remain responsible for understanding, reviewing and approving the code that eventually becomes part of a software product.  

 

Infrastructure investment becomes part of the AI coding shift

 

GitHub's infrastructure expansion shows how AI coding tools are increasing demand for computing power, storage and network capacity. The company is also moving towards a multi-region deployment on Microsoft Azure, with the migration expected to be completed in early 2027. This is intended to improve service availability and help manage growing workloads.

 

The company said its investments in 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity are aimed at supporting the rising volume of software development activity. As developers increasingly delegate tasks to AI agents, more code is being generated and submitted for testing and review.

 

While AI agents can help developers complete tasks faster and work on multiple projects, the increase in code also adds to the workload of testing, security reviews and maintenance. Software platforms will need to expand their infrastructure while ensuring that faster code generation does not affect software quality and reliability.