SEOUL, July 03 (AJP) - Artificial intelligence startup Anthropic is considering Samsung Electronics' 2-nanometer manufacturing process for its first custom AI chip, potentially handing the South Korean company another marquee customer in its bid to challenge Taiwan Semiconductor Manufacturing Co.'s dominance of advanced chip production.
U.S. technology publication The Information, citing people familiar with the matter, reported that Anthropic is evaluating Samsung Foundry's 2nm manufacturing process and advanced packaging capabilities for a proprietary AI processor. The project remains at an early stage, with the chip's architecture, performance specifications and server deployment strategy yet to be determined, while detailed design work has not begun.
The development comes as leading artificial intelligence companies increasingly design proprietary silicon to reduce dependence on Nvidia's graphics processing units, which continue to dominate AI computing infrastructure.
Anthropic has recently strengthened its semiconductor expertise by hiring Clive Chan, an early member of OpenAI's custom chip team. OpenAI last week unveiled its first in-house inference processor, Jalapeño, developed with Broadcom, while Alphabet, Amazon, Meta and Microsoft are all expanding development of their own AI chips as they seek greater control over computing costs and supply chains.
For Samsung, securing Anthropic as a foundry customer would represent another potential boost for its efforts to revive its contract chip business, which has trailed Taiwan Semiconductor Manufacturing Co. in the race for leading-edge process technology.
The discussions also come as Samsung steps up its push to attract AI chip developers to its 2nm manufacturing platform. At its SAFE Forum held this week in Seoul, the company highlighted its design technology co-optimization strategy and advanced packaging capabilities, underscoring efforts to improve performance, power efficiency and manufacturing yields for next-generation AI processors.
Separately, Google has also been reported to be considering Samsung's 2nm process for part of a future tensor processing unit, adding to signs that major AI companies are looking beyond TSMC as demand for advanced semiconductor manufacturing continues to accelerate.
Anthropic did not confirm the reported discussions. In a statement provided to The Information, the company said Amazon Web Services' Trainium chips, Google's tensor processing units and Nvidia's GPUs would "remain central" to its compute strategy, indicating that any future proprietary processor would complement, rather than replace, its existing hardware partnerships.
Any eventual agreement would strengthen Samsung's efforts to expand its portfolio of leading-edge foundry customers and narrow the competitive gap with TSMC. The discussions, however, remain exploratory, and there is no assurance they will lead to a manufacturing contract.
Copyright ⓒ Aju Press All rights reserved.