Cerebras launches new server chip and system designed to speed AI chatbots

San Francisco, August 18 - Cerebras Systems unveiled on Tuesday a cutting-edge server hardware system equipped with its oversized chips, claiming these innovations will expedite AI chatbot query processing. The company, which competes with Nvidia in the AI hardware and chip sector, specializes in AI inference, the computational task that generates responses in chatbots like Anthropic's Claude.

The novel system, known as CS-4, is a server rack powered by three of Cerebras' bespoke chips, which the company asserts results in enhanced performance. The hardware is built upon Cerebras' Nexus server architecture, featuring modular components that accommodate the chips. Cerebras' chips outperform competitors due to their substantial size, which helps avoid data transfer delays and energy waste between chips.

The new machine is slated for release in the third quarter, with the chips manufactured using TSMC's 5-nanometer process. Cerebras also aimed to simplify setup by reducing components by 50%, according to Chief Technology Officer Sean Lie, which would expedite data center construction. The server rack designed by Cerebras houses the WSE-3 Turbo chip and new networking components that Lie claims will expedite data movement between chips.

The company plans to release another generation of the chip and server in 2027, with CEO Andrew Feldman anticipating 600 megawatts' worth of computing power by the end of 2027. Feldman emphasized Cerebras' engineering focus on accelerating data processing capabilities, stating they aim to be "four times as fast between now and the end of the year, and 20 times more throughput" by the end of 2027. In a recent report, Cerebras disclosed an adjusted loss of $6.9 million on sales of $180.1 million.