Tired of shortages, OpenAI considers making its own AI chips

OpenAI, the creator of ChatGPT and DALL-E 3 generative AI merchandise, is exploring the opportunity of manufacturing its personal AI accelerator chips, in keeping with Reuters. Citing nameless sources, the Reuters report signifies that OpenAI is contemplating the choice attributable to a scarcity of specialised AI GPU chips and the excessive prices related to working them.

OpenAI has been evaluating numerous choices to deal with this concern, together with doubtlessly buying a chipmaking firm and dealing extra intently with different chip producers like Nvidia. At the moment, the AI agency has not made a ultimate determination, however the discussions have been ongoing since a minimum of final 12 months. Nvidia dominates the AI chip market, holding greater than 80 p.c of the worldwide share for processors greatest suited to AI functions. OpenAI CEO Sam Altman has publicly expressed his concerns over the shortage and value of those chips.

The {hardware} scenario is alleged to be a high precedence for OpenAI, as the corporate presently depends on an enormous supercomputer constructed by Microsoft, one among its largest backers. The supercomputer makes use of 10,000 Nvidia graphics processing items (GPUs), in keeping with Reuters. Working ChatGPT comes with vital prices, with every question costing roughly 4 cents, in keeping with Bernstein analyst Stacy Rasgon. If queries develop to even a tenth of the size of Google search, the preliminary funding in GPUs could be round $48.1 billion, with annual upkeep prices at about $16 billion.

Creating customized AI chips would place OpenAI within the firm of different massive tech corporations like Google and Amazon, which have additionally ventured into customized chip design. Nonetheless, manufacturing its chips could be a major strategic shift for OpenAI and signify a considerable monetary funding. An acquisition of an present chip agency might doubtlessly pace up the method, very similar to Amazon’s acquisition of Annapurna Labs in 2015 did for its customized chip tasks.

Reuters notes that the method of creating a customized chip, even with an acquisition, would seemingly take a number of years, leaving OpenAI depending on business suppliers like Nvidia and AMD within the meantime. And it is not a simple course of—Meta’s efforts to create its personal AI chips have encountered issues, in keeping with Reuters.

OpenAI’s principal backer, Microsoft, can also be stated to be working on a custom AI chip, which OpenAI is presently testing, in keeping with a report by The Info. Microsoft plans to disclose that chip subsequent month, in keeping with the report.