Anthropic in talks with Samsung for new custom chip
▼ Summary
– Anthropic is exploring a collaboration with Samsung to develop its own AI chips, following earlier reports of interest in addressing chip shortages.
– The chip’s specific purpose, server integration, and performance specifications have not yet been determined by Anthropic.
– Anthropic stated that its compute strategy will continue to rely on a diverse hardware stack from Google, Amazon, and Nvidia.
– The move follows a trend of AI companies creating custom chips to reduce dependence on Nvidia, with competitor OpenAI recently announcing its own “Jalapeño” inference processor.
– Samsung, already a major Nvidia partner producing chips for AI training, is also discussing chip-making collaborations with Google.
Back in April, Reuters reported that Anthropic was exploring the possibility of developing its own AI chips to address ongoing chip shortages. Now, it appears the company is moving forward with that ambition. According to a report from The Information on Thursday, Anthropic has been in discussions with Samsung to explore a potential partnership around a new custom chip. However, the company has yet to finalize key details such as the chip’s intended use, its role within server architecture, or its processing power.
When asked for comment, Anthropic told TechCrunch that a diversified hardware strategy incorporating chips from Google, Amazon, and Nvidia will remain central to its computing approach. Regarding the potential Samsung collaboration, the company declined to provide further specifics.
Several AI firms are pursuing custom chip development, both to create specialized hardware for particular computing tasks and to reduce reliance on Nvidia, which still dominates the chip market. Anthropic’s move may also be a direct response to last week’s announcement from its main rival, OpenAI, which partnered with Broadcom to unveil its own custom-built inference processor called “Jalapeño.” OpenAI claims the chip offers superior performance-per-watt compared to competitors. Meanwhile, both Amazon and Google already offer custom-built TPUs as part of their cloud services.
Samsung is already deeply embedded in the AI sector. The company is a major partner of Nvidia, producing chips essential for training and running AI models. In return, Samsung uses Nvidia’s software to manufacture its own chips. The two companies are jointly developing an AI chip factory in South Korea. Samsung has also held discussions with Google about collaborating on chip production.
(Source: TechCrunch)
