OpenAI is currently exploring options to develop its own AI chips to reduce their dependency on NVIDIA’s GPUs for its AI models. Sam Altman, CEO of OpenAI, is in talks with the semiconductor company “Broadcom” to potentially manufacture these independent chips.
This move aims not only to secure a supply of components but also to bolster infrastructure, as well as power and data centres. OpenAI has also hired ex-Google employees who specialise in working with AI hardware, reflecting its broader ambition to establish a network of semiconductor factories, potentially requiring billions of dollars in funding.
Broadcom’s proficiency in custom ASIC solutions is also a strong fit for OpenAI’s requirement to tailor their accelerators. The company can provide essential silicon solutions for their data centres, encompassing networking components, PCIe controllers, as well as SSD controllers.
The overall product lineup enables OpenAI to potentially fulfil its data centre requirements efficiently. Additionally, Broadcom’s expertise in inter-system and system-to-system communication technologies could enhance OpenAI’s overall AI infrastructure capabilities as a whole.
OpenAI has also opened an office in Japan to explore new revenue opportunities and foster collaborations with local businesses, governments, and even research institutions. To that end, they’ve already collaborated with organisations such as Carnegie Mellon University and Khan Academy to create personalised learning experiences using AI.
However, the possibility of developing their own AI chip that can rival NVIDIA could take years of research and development to achieve such significant results. Altman has plans to raise billions of dollars to set up a chain of factories to manufacture AI semiconductors with the likes of Intel, Taiwan Semiconductor Manufacturing Company and even Samsung Electronics as potential partners.
(Source: TechSpot,TrendForce)
Haikal Nazri contributed to this article.
