Cerebras Challenges NVIDIA with New AI System That's Multiple Times Faster, Memory-Free Design a Key Highlight

Cerebras Systems, Inc. Class A (CBRS-US) held a launch event in San Francisco on Tuesday (the 18th), unveiling a new computing system powered by its proprietary chip, claiming it offers significantly faster processing speeds. The company asserts this advancement will give it a competitive edge against NVIDIA (NVDA-US) hardware.

The new system, named CS-4, is currently being tested by a limited number of customers and is expected to expand availability later in the third quarter. CS-4 promises processing speeds multiple times faster than its predecessor, and Cerebras previously claimed its prior-generation product outperformed systems built with NVIDIA's mainstream processors.

This product upgrade marks a crucial step for Cerebras in challenging NVIDIA's dominant position in the AI data center hardware market. Although Cerebras completed its initial public offering (IPO) earlier this year and remains much smaller in scale than NVIDIA, the company aims to triple its revenue by next year.

Since its IPO in May, Cerebras' stock price has risen 19%.

Cerebras currently operates two primary businesses: selling computing systems equipped with its own chips, and providing data center computing services powered by its proprietary technology. Both businesses target the rapidly growing market for AI infrastructure spending.

At the event, CEO Andrew Feldman and other executives demonstrated the new hardware in action, stating it produces AI-generated results significantly faster—multiple times faster—than alternative solutions based on NVIDIA's architecture.

Cerebras is heavily promoting 'fast result generation' as its core differentiator. The company explains that because it can store all parameter values required to run AI models directly on its massive chip, there's no need to access memory chips via an interface, dramatically accelerating computation.

The new-generation system also features improved thermal design, with future compatibility for an upgraded chip expected next year. Enhanced cooling allows a single unit to house three of Cerebras' current processors, enabling them to operate at higher speeds.

Feldman stated that lessons learned from past generations enabled a modular design, making upgrades easier and allowing configuration adjustments to support various external systems.

He noted this innovative computer design will help Cerebras meet surging demand. First, it doesn't require memory chips, which are currently in short supply. Second, the manufacturing process technology it uses has more abundant supply, enabling the Sunnyvale, California-based company to deploy systems more quickly.