OpenAI reduces ChatGPT operating costs by half
Engineers at OpenAI, a leading company in artificial intelligence technologies, have developed a method to significantly reduce system operating costs without purchasing new equipment. This was reported by Zamin.uz.
According to data from influential publications, the company has succeeded in reducing the computational power required to process ChatGPT user queries by more than half. This achievement not only ensures financial savings but also provides a strategic advantage at a time when computing resources are scarce worldwide.
The innovation relates to improving the inference process—the stage where a trained model directly responds to user queries, i.e., generating answers. For companies developing generative AI systems today, this stage demands the highest computational costs.
While model training occurs within a specific timeframe, interacting with users consumes separate resources and power for each query. As emphasized by sources, OpenAI’s newly implemented efficiency system primarily targets the layer of users who access ChatGPT without registration or for free.
As a result, the number of graphics processors required to serve these users over a given period has dropped by several hundred. For a service of this scale, such a reduction is exceptionally low and remarkable.
So far, OpenAI has not officially disclosed the specific technological methods used to achieve this result. However, experts speculate that the gain was not achieved by adding extra hardware, but through smarter utilization of existing server infrastructure, improved memory management, or optimization of batch-processing algorithms.
This development could fundamentally reshape the economic balance in the AI market. At a time when powerful chips are in short supply globally and billions of dollars are being spent building data centers, reducing costs through software optimization is the right path forward.
It opens the opportunity for OpenAI to offer its services to a much broader audience. If this technology is widely adopted, the company could expand its free offerings, lower prices for corporate clients, and serve more users with existing infrastructure.
Currently, it is unclear whether this innovation applies to paying subscribers or the company’s most complex logical models. Nevertheless, such achievements in software optimization demonstrate that in the AI race, success depends not only on the number of technical devices but also on the skill of using them effectively.
