① According to reports, OpenAI is in touch with chip designers including Broadcom to discuss the development of new AI chips; ② In addition, the company is also recruiting former Google employees, hoping to use their experience and technology in developing Tensor processors to develop its own AI server chips.
Finance Association, July 19 (Editor Huang Junzhi) According to media reports on Thursday, quoting people familiar with the matter, ChatGPT manufacturer OpenAI is in touch with chip designers, including Broadcom (Broadcom), to discuss the development of new artificial intelligence (AI) chips.
Artificial intelligence models such as ChatGPT, GPT-4, and DALL-E3 developed by OpenAI rely on expensive graphics processing units (GPUs). To overcome this problem, it is exploring the idea of making its own artificial intelligence chips.
The report quoted three people familiar with the matter as saying that the Microsoft-supported company is still recruiting former Google employees, hoping to use its experience and technology in developing Tensor processors to develop its own AI server chips. The move is not only aimed at reducing reliance on Nvidia, but is also part of OpenAI CEO Altman's vision to enhance global semiconductor infrastructure.
“OpenAI is in ongoing dialogue with industry and government stakeholders to expand access to the infrastructure needed to ensure that the benefits of artificial intelligence are widely accessible,” an OpenAI spokesperson said.
Furthermore, the report also said that since many details still need to be finalized, if the chip is finally developed, it will not be produced until 2026 at the earliest. Affected by this news, Broadcom closed up 2.91% on Thursday, with an increase of about 48% so far this year.
In order to fully enter the AI field, OpenAI did take quite a bit of effort. According to media reports earlier this year, the company's CEO Sam Altman (Sam Altman) plans to raise billions of dollars to produce semiconductors with chipmakers Intel, TSMC, and Samsung Electronics as potential partners.
Furthermore, in order to gain more market share, OpenAI launched a “cheap and powerful” GPT-4o mini model on Thursday, starting a price war. According to the company, the GPT-4o mini is an entry-level artificial intelligence “little model”, and the price has dropped significantly.
According to reports, ChatGPT's free/paid users have been using this new model since Thursday, and enterprise users will get the update next week.
According to statistics, the GPT-4O mini has reached the lowest price among the mainstream “small models” of American AI companies. The input/output price of 1 millionToken is only 15 cents and 60 cents. As a comparison, the GPT-4O model has an input/output price of 5 USD/15 US dollars per million tokens.