Due to a shortage of processors needed to run OpenAI’s intense software as well as the exorbitant costs associated with operating it, the makers of ChatGPT are reportedly considering making their own AI chips.
OpenAI, the company behind ChatGPT, has been relying on Nvidia for its chips ever since it came into existence. With more than 80% of the global market for AI chips captured, Nvidia has been enjoying the pseudo-monopoly that it holds in this field. However, OpenAI CEO Sam Altman has recently commented on the scarcity of these chips. On Friday, sources that are familiar with the company’s plans told Reuters that Altman has been exploring the possibility of his company either getting into the chip business or acquiring a company that already makes them. He has also considered diversifying suppliers by working with companies other than Nvidia.
OpenAI uses a supercomputer by Microsoft which makes use of 10,000 Nvidia GPUs (Graphics Processing Units). According to Bernstein analyst Stacy Rasgon, each ChatGPT query costs the company around 4 cents which could end up being an unimaginably huge amount as ChatGPT continues to grow.
Of course, the exorbitant costs of getting into the chip business represent a significant hurdle. Building its own AI chips will easily cost the company hundreds of millions of dollars and a couple of years. And that’s on top of reports that OpenAI is in discussions with former Apple designer Jony Ive regarding the possibility of building the “iPhone of AI”—a moonshot that would also surely require billions of dollars in investment.
When it comes to producing its own chips, the acquisition of a chip company does sound more doable for OpenAI. Both Google and Amazon are in control of designing the chips that run their businesses. Amazon accomplished that in 2015 by acquiring Annapurna Labs. Meta also explored this option and even came up with its own chip, but following several issues with the technology, the company has been working on a redesign.
In the meantime, OpenAI will remain dependent on incumbent giants of GPUs.