Cerebras reveals next-gen wafer-scale AI roadmap at Hot Chips

Cerebras detailed its upcoming Nexus rack design, which triples rack-scale performance compared to current systems. The company also outlined the CS-6 wafer, which will integrate stacked DRAM for improved memory bandwidth. These announcements mark the next two generations of its wafer-scale accelerator roadmap.

Cerebras unveiled its hardware roadmap at the Hot Chips conference, detailing two upcoming generations of wafer-scale AI systems. The Nexus rack architecture is engineered to deliver three times the rack-scale performance of current deployments, representing a substantial density improvement for AI compute clusters. The company also previewed the CS-6 wafer, which will integrate stacked DRAM directly into the wafer design to boost memory bandwidth, a critical constraint for large language model training workloads. These announcements signal Cerebras's continued push into the specialized AI accelerator market, positioning its wafer-scale approach as an alternative to conventional GPU-based systems from established vendors.