In the competition for AI computing infrastructure, major model giants are extending their research efforts to the hardware level. Recently, it was reported that the well-known AI company Anthropic has officially started early preparations for its own AI chip development and has had in-depth discussions with Samsung Electronics regarding potential manufacturing collaborations.

According to information, this project is still in its initial stage. The core team at Anthropic is systematically planning the functional positioning, computing specifications, and architectural deployment of the self-developed chip in future server clusters. Although the company has contacted several chip design companies, the project has not yet entered the detailed design phase. According to informed sources, Anthropic's preliminary technical roadmap is relatively clear, and the company plans to use Samsung's advanced 2nm process technology combined with high-integration advanced packaging technology to achieve industry-leading energy efficiency and performance metrics.

To strengthen its hardware R&D capabilities, Anthropic has also been actively building its talent pool. At the beginning of this month, the company successfully recruited Clive Chen, a key member of OpenAI's original self-developed chip team, which is seen as an important signal of its effort to strengthen its hardware "foundation."

At the same time, the competitive landscape in the large model hardware field has become clear. Anthropic's move comes at a time when the joint development of the "Mexican Pepper" chip by OpenAI and Broadcom has made progress. It is reported that the engineering sample of this chip has been running complex machine learning tasks stably under production-standard power consumption and frequency in a laboratory environment, and has successfully run OpenAI's code model GPT-5.3-Codex-Spark launched in February of this year.

From model development to hardware customization, AI giants are showing a significant increase in their "control" over underlying computing infrastructure. As the barriers to chip design and manufacturing technologies continue to be broken down, future competition in the large model market will no longer be limited to algorithm iterations but will deeply extend to infrastructure competition. Anthropic's entry into this field undoubtedly adds a new variable to this computing power race.