Anthropic is in talks with Samsung to develop its first custom AI chip

Anthropic plans to manufacture its own customized Artificial Intelligence (AI) chip, for which is in full talks with the Korean company Samsung regarding a possible collaboration for the development of the processor with a 2 nanometer manufacturing process.

The AI ​​company is considering this option in order to face the shortage of chip production that the sector is going through, with the aim of diversify your computing strategyby continuing to collaborate with the companies with which it currently has agreements, such as Google, Amazon and Nvidia.

This has been reported by the technology media The Information and TechCrunch based on statements from sources related to the company, who have detailed that Anthropic has been in contact with Samsung to explore a possible collaboration with the aim of develop its first chip in 2nm manufacturing process and utilize its advanced packaging facilities.

However, the technology company has not specified the purpose that will be given to its personalized AI chip, since it is still You have to decide if you will use it for training or inferenceas well as specifying how it will be included in the server or what power it will have.

An Anthropic spokesperson has confirmed that the company’s current relationships with the three companies mentioned will continue to be fundamental to its computing strategy, vital to securing, managing and optimizing the processing power necessary for the development, training and execution of its AI models.

Beyond this supply relationship, both Samsung, SK Hynix and Micron participated in the financing round in May Anthropic H Series, which has given the AI ​​lab a closer relationship with the chipmaker.

Samsung currently occupies a unique position in the semiconductor supply chainnot only for being a key partner of Nvidia in the supply of high-bandwidth memory (HBM), but for its versatility to be a contract chip manufacturer by competing directly with TSMC, the dominant manufacturer worldwide, as reported by TechTimes.

These conversations follow OpenAI’s announcement of ‘Jalapeño’ a little over a week ago, a custom inference processor that it co-developed with Broadcom, and which is part of the strategy of AI laboratories to reduce their dependence on Nvidia (company that dominates approximately 74 percent of the AI ​​chip market).