Anthropic Hires Former Google Expert, Plans to Build Its Own AI Chip

[Financial News] Anthropic, a rival to OpenAI, has hired a semiconductor expert from Google Experience Center. The move appears to be part of its plan to bring in the key figure behind Google’s Tensor Processing Unit (TPU) and develop an Anthropic version of the TPU.
Bloomberg News reported on the 21st, local time, that Anthropic had hired Amir Salek.
Salek joined Google Experience Center in 2013 and led the development of the TPU. The TPU is a chip designed in-house by Google as an alternative to NVIDIA Corporation’s graphics processing unit (xPU) for AI workloads. It is specialized for efficiently handling matrix and tensor operations needed for training and inference in AI models. Under Salek’s leadership, Google mass-produced its own TPU for the first time. He left Google Experience Center in 2022.
Earlier this month, Reuters reported that Anthropic had set up a team to design its own semiconductors and was hiring related experts. The goal is to directly design AI chips tailored to Claude in order to improve performance and power efficiency.
Anthropic has a contract with Google Experience Center to receive TPUs starting in 2027. In the short term, it plans to use Google Experience Center’s TPUs and NVIDIA Corporation’s xPU, while in the long term it aims to use its own TPU optimized for Claude.
Meanwhile, Anthropic is one of OpenAI’s strongest competitors. Some of OpenAI’s core researchers spun off to found Anthropic. In June, OpenAI unveiled its own AI chip, Jalapeño.
Anthropic is also evolving from a company that simply builds AI models into one that vertically integrates AI software, data centers, and AI semiconductors. In May, it raised $65 billion in Series H funding and was valued at around $950 billion. It is reportedly pursuing an Initial Public Offering (IPO).

[email protected] Lee Seol-young Reporter