According to sources, ChatGPT’s parent firm, OpenAI, is considering creating its own AI chips and investigating a potential acquisition target.
According to Reuters, internal conversations indicate the corporation has not chosen to proceed. According to sources, OpenAI has considered ways to address its scarcity of pricey AI processors since last year.
Building its own AI processor, collaborating more closely with chipmakers like Nvidia, and diversifying its suppliers are among these ideas. OpenAI rejected the comment.
CEO Sam Altman prioritizes buying more AI processors. He has openly complained about the shortage of graphics processing units, which Nvidia controls over 80% of the worldwide market for AI-optimized CPUs. Altman cites a lack of powerful processors that power OpenAI’s software and the “eye-watering” expenses of running its hardware to support its initiatives and goods as reasons to obtain additional chips.
Microsoft, one of OpenAI’s biggest investors, has built a gigantic supercomputer using 10,000 Nvidia GPUs to build its generative artificial intelligence capabilities from 2020.
Running ChatGPT costs the firm a lot. Bernstein analyst Stacy Rasgon estimates each query costs 4 cents. ChatGPT would need $48.1 billion in GPUs and $16 billion in processors yearly to operate at a tenth of Google’s search volume.
Personalized chips
OpenAI would join a select group of huge tech companies like Alphabet’s Google (GOOGL.O) and Amazon.com (AMZN.O) that have pushed to build their CPUs. It is unclear if OpenAI will design a proprietary chip. Industry veterans say it would be a big strategic undertaking and investment that may cost hundreds of millions annually. Investing resources would not ensure success for OpenAI.
A chip firm purchase might speed up OpenAI’s chip development, as it did for Amazon.com and Annapurna Labs in 2015. One source said OpenAI examined due diligence on a prospective purchase candidate. The purchasing business OpenAI studied was unknown.
Suppose OpenAI pursues a proprietary processor, including an acquisition. In that case, it may take years and leave it dependent on Nvidia and AMD.O. Big IT corporations have built their CPUs for years with limited success. Reuters said that Meta’s (META.O) custom chip effort has been plagued, causing the company to abandon several AI processors. The Facebook owner is developing a new processor for all AI tasks.
The Information reports that OpenAI’s major supporter, Microsoft (MSFT.O), is building and testing a proprietary AI processor. Plans may indicate increasing distance between the firms.
Specialized AI chips have been more popular since ChatGPT launched last year. The latest generative AI technology requires AI accelerator chips to learn and run. One of the few chipmakers that makes effective AI processors, Nvidia leads the industry.