Anthropic's Secret Weapon: Developing Custom AI Chips with Samsung

Anthropic is reportedly intensifying its efforts to produce its own custom AI chips. This strategic move was first hinted at in April following concerns over chip shortages.

Recent developments, as reported by The Information, indicate that Anthropic has engaged in discussions with Samsung to explore a potential collaboration for the development of these proprietary chips. However, the exact application, server integration and power specifications of the prospective chip remain undecided, according to the report.

When questioned about these developments, Anthropic affirmed to TechCrunch that its compute strategy continues to heavily rely on a diversified hardware stack, incorporating chips from major players like Google, Amazon and Nvidia. The company chose not to elaborate further on the specifics of a potential partnership with Samsung.

This pursuit of custom silicon by Anthropic is part of a broader trend within the AI industry. Many AI firms are endeavoring to develop their own chips, driven by a dual objective to create hardware specifically optimized for unique compute tasks and to reduce their reliance on Nvidia, which currently holds a dominant position in the AI chip market.

This move by Anthropic could also be interpreted as a response to its key competitor, OpenAI, which recently announced its collaboration with Broadcom to develop its custom-built inference processor, dubbed “Jalapeño.” OpenAI claims that its new chip offers superior efficiency and performance-per-watt compared to existing competitor chips.

Furthermore, major cloud providers such as Amazon and Google have already integrated custom-built Tensor Processing Units (TPUs) into their cloud offerings, underscoring the value of specialized hardware in the AI ecosystem.

Samsung, a global technology powerhouse, is already deeply entrenched in the AI industry. The company serves as a significant partner to Nvidia, manufacturing the chips essential for training and running Nvidia’s AI models, while also utilizing Nvidia’s software for its own chip production.

The two companies are also jointly developing an AI chip manufacturing facility in South Korea. Samsung has also held discussions with Google regarding potential partnerships in its chip-making endeavors, highlighting its extensive involvement and capabilities in the semiconductor landscape.