On July 19, OpenAICEO Sam · Altman Altman) is working with companies including Broadcom Inc.), semiconductor design companies are in talks to develop new chips to reduce dependence on NVIDIA and strengthen supply chains. It is reported that the negotiations are still in the early stages, and OpenAI has already "engaged with the entire industry".
The next NVIDIA?
Headquartered in California, United States, Broadcom designs, develops, and supplies semiconductors, as well as enterprise software and security solutions. In June, Broadcom announced its results for the fiscal second quarter ended May 5, with revenue up 43% year-on-year to $12.487 billion, mainly benefiting from strong demand for artificial intelligence and virtualization software VMware business.
Among them, the revenue of the semiconductor solutions segment (responsible for network chips and custom chips) increased by about 6% to $7.2 billion, and the revenue of AI products was $3.1 billion. At the same time, Broadcom also raised its estimated sales of AI chips this year from $10 billion to $11 billion.
Previously, Broadcom was considered by Wall Street analysts to be likely to be the "next Nvidia". Broadcom had said in March that it had added a third customer to its AI chip customization business, and although the names of the customers were not explicitly disclosed, analysts generally believed that those customers included Google and Meta.
Boosted by the news, Broadcom (AVGO.O) shares closed up 2.91% at $160.52 per share (Broadcom conducted a 1-for-10 stock split on July 15), with a market capitalization of $747.2 billion.
Challenges to NVIDIA?
The development of a new server chip that can rival Nvidia is highly unlikely and will take years to materialize.
If OpenAI or the new joint venture proceeds to design new AI chips, it could offend Nvidia, OpenAI's most important chip supplier, but it could also provide potential leverage for OpenAI in future pricing negotiations with the company.
Nvidia has been focusing on AI GPUs create unprecedented margins and sales because there is no viable alternative for customers like OpenAI. The chip design company has already provided hundreds of thousands of GPUs to Microsoft so that OpenAI can use them to develop ChatGPT and other products.
The development of a new server chip that can rival Nvidia is highly unlikely and will take years to materialize.
If OpenAI or the new joint venture proceeds to design new AI chips, it could offend Nvidia, OpenAI's most important chip supplier, but it could also provide potential leverage for OpenAI in future pricing negotiations with the company.
Nvidia has been focusing on AI GPUs create unprecedented margins and sales because there is no viable alternative for customers like OpenAI. The chip design company has already provided hundreds of thousands of GPUs to Microsoft so that OpenAI can use them to develop ChatGPT and other products.
|OpenAI has also begun to enter the small model
On July 18, OpenAI launched GPT-4o The mini, which is currently the most powerful and cost-effective small model available, will initially only support text and image generation tasks, and will later integrate image, video, and audio functions into the model.
OpenAI says GPT-4o The mini is its first AI model to use a new security policy "instruction hierarchy" with the aim of making the AI system prioritize certain instructions, such as those from companies, to make it harder for people to get tools to do things they shouldn't.
Altman had previously predicted that AI would learn so fast that one day the data would not be able to keep up. "One day we may run into a data bottleneck. It may or may not break through...... But human-generated data is clearly not enough in the long run...... We need a model that can learn more with less data. He said.
OpenAI disclosed that GPT-4o The API price of mini will be more than 90% lower than GPT-4o. The model will be available to ChatGPT free users, Plus, and Team subscribers, and will be available to enterprise users next week.
