According to reports from Reuters, the company is actively exploring the idea of creating its own artificial intelligence chips and has even considered the possibility of acquiring chip-related assets. The motivation behind this move stems from OpenAI CEO Sam Altman’s acknowledgment of GPU shortages impacting the speed and reliability of the company’s API. To address this issue, OpenAI is prioritizing the acquisition of more AI chips.
Beyond mitigating GPU shortages, utilizing proprietary chips could help OpenAI manage the costs associated with running its AI products more efficiently. An analysis conducted by Stacy Rasgon of Bernstein Research revealed that each ChatGPT query currently costs the company approximately 4 cents. With the service rapidly gaining popularity and reaching 100 million monthly users within its initial two months, it handles millions of queries daily. However, there was a temporary decline in users in July. Rasgon estimates that if ChatGPT’s query volume were to reach a fraction of Google’s, OpenAI would initially require $48.1 billion worth of GPUs and an ongoing annual expenditure of $16 billion on chips.
Currently, NVIDIA dominates the market for AI-focused chips. For example, the Microsoft supercomputer used by OpenAI for its technology development relies on 10,000 NVIDIA GPUs. As a result, several major tech players have begun developing their own custom AI chips. Microsoft, OpenAI’s primary supporter, has been working on its AI chip since 2019, known by the codename Athena. OpenAI is said to have been testing this technology.
OpenAI has not yet made a final decision on whether to proceed with the development of its own chips, as per Reuters. Additionally, even if they opt to move forward, it may take several years before OpenAI can effectively implement its custom chips to power its AI products.