OpenAI has been recruiting former members of Google's TPU division, seeking to develop AI server chips, and has been negotiating with chip design companies, including Broadcom, to develop this new AI chip, according to The Information, citing people familiar with the matter. Last year, OpenAI CEO Sam Altman decided to start a new company to develop and produce new AI chips and help build chip factories and data centers to manufacture and house them. While the plan has taken several turns, there are signs that his efforts to build new chips are taking shape, and that they are starting within OpenAI.
Broadcom is a semiconductor design company headquartered in California, United States, primarily engaged in semiconductor design, development and supply, as well as enterprise software and security solutions. Broadcom has extensive experience and expertise in the semiconductor industry, especially in the field of AI chips. According to Broadcom's second-quarter results, its revenue surged 43% year-on-year, mainly benefiting from strong demand for artificial intelligence. In addition, Broadcom has also been considered by Wall Street analysts to be likely to be the "Next Nvidia".
It is reported that OpenAI's negotiations with Broadcom are still in the early stages, but OpenAI has already "engaged with the entire industry."
Figure: OpenAI was revealed to have approached Broadcom to jointly develop AI chips
To strengthen the technical team, OpenAI also hired former Google employees who were involved in the development and production of Google's TPU to help with the design of its AI chips. In addition, OpenAI is also actively expanding its cooperation with other chipmakers, and plans to establish partnerships with top chipmakers such as TSMC to expand production capacity and accelerate the large-scale production of new AI chips.
This collaboration could have a significant impact on the existing AI chip market landscape. Currently, Nvidia controls about 80% of the AI chip market, and OpenAI's new plan could change that. Nvidia's recently launched next-generation architecture Blackwell chips have surpassed the previous generation in terms of performance and power consumption, but OpenAI's self-developed chip program will undoubtedly bring new competitiveness to the market.
In addition to accelerating the development of self-developed chips, OpenAI has also begun to enter the small model. On July 18, OpenAI launched GPT-4o mini, calling it the most powerful and cost-effective small model available. Initially, the model will only support text and image generation tasks, and will later integrate image, video, and audio functions. OpenAI says GPT-4o mini is its first AI model to use a new security policy "instruction level" with the goal of making the AI system prioritize certain instructions to make it harder for people to get tools to do things they shouldn't.
OpenAI's cooperation with Broadcom to jointly develop AI chips is an important initiative in the field of artificial intelligence. This cooperation will not only help reduce OpenAI's dependence on external suppliers such as Nvidia and strengthen the stability of its supply chain; It will also promote the innovation and development of AI chip technology and inject new vitality into the technology industry.
