OpenAI's Codex tool is hammering solid-state drives with needless write operations due to a clumsy logging implementation, according to a report from The Register. The bug causes the system to squirrel away data without regard for the cost of SSD wear. The issue has already resulted in millions of dollars in extra hardware costs.
The problem stems from how Codex handles logging—writing far more data than necessary, which accelerates the degradation of SSDs. Unlike traditional hard drives, SSDs have a finite number of write cycles before they fail. The unnecessary writes mean drives must be replaced more frequently, inflating operational expenses.
The Register, citing sources familiar with the matter, did not provide specific figures beyond noting the costs run into the millions. No official statement from OpenAI has been released yet. The report does not indicate whether the bug also affects performance or user experience.
The impact is likely concentrated on the infrastructure side, affecting cloud providers or internal server costs rather than individual users. If unaddressed, the expense could compound over time as Codex scales. The inefficiency may also raise questions about broader software optimization practices at OpenAI.
Critics might argue that the cost represents a small fraction of OpenAI's overall spending, and that the issue is relatively minor compared to model training expenses. A fix could reduce costs significantly with a relatively simple code change.