Cerebras unveiled the CS-4 server system to accelerate AI chatbots
Cerebras Systems has unveiled its new CS-4 server system with large chips, which the company says is designed to speed up the processing of requests to AI-powered chatbots. Channel NewsAsia reported this, citing Reuters.
Cerebras develops hardware and chips for AI and competes with Nvidia in the inference segment—the computational process of generating chatbot responses. Reuters cites Anthropic’s Claude as an example of such a service.
The CS-4 is a server rack equipped with three large Cerebras chips. The system is based on the Nexus server architecture, which features interchangeable modules for housing the chips. The company stated that the large size of its chips helps avoid energy consumption and slowdowns associated with data transfer between individual chips.
For more breaking news, follow the UA.News Telegram channel.
The server rack is equipped with a WSE-3 Turbo chip and new networking components. According to Cerebras CTO Sean Lee, these components are designed to accelerate data exchange between chips. The chips are manufactured using TSMC’s 5-nanometer process, and the system will be available in the third quarter.
Lee said the new system was designed to simplify installation: it has 50% fewer components. He added that this could speed up the construction of data centers. Cerebras also plans to release the next generation of chips and servers in 2027.
Cerebras CEO Andrew Feldman stated that the company expects to deliver a total of 600 megawatts of computing power by the end of 2027. Last week, Cerebras reported an adjusted loss of $6.9 million on revenue of $180.1 million.
