Overview
- The company unveiled the CS-4 on Tuesday while reporting second-quarter results that missed top-line expectations and showed an adjusted loss, triggering a roughly 12–13% intraday drop in its stock.
- Cerebras says the CS-4 is a rack-scale Nexus system built around three WSE-3 Turbo wafer-scale processors, built on TSMC 5nm, available in Q3 and designed with 50% fewer components to speed data-center deployment.
- The firm claims the CS-4 can deliver up to 30x faster inference and greatly higher tokens-per-second and throughput per watt versus production GPU systems, citing wafer-scale SRAM, massive on‑wafer I/O and 125 petaflops per wafer figures.
- Financials showed $180.1 million in Q2 revenue, a core revenue figure of $209.9 million with core cloud and services up 287% to $127.7 million, and an adjusted Q2 loss of about $6.9 million, yet management raised full-year core revenue guidance.
- Cerebras projects 600 megawatts of capacity by end-2027 and a multi-fold speed and throughput roadmap, a plan that tests whether its capital‑intensive, wafer‑scale approach can convert partnerships into steady cloud revenue and displace GPU incumbents on latency‑sensitive inference workloads.