The Wafer-Scale Engine (WSE) is Cerebras's AI processor, built as one chip the size of a whole silicon wafer instead of being cut into many small dies.
The numbers from Cerebras
Cerebras's chip page gives a die size of 46,225 mm2, 4 trillion transistors and 900,000 AI-optimized cores. The CS-3 page lists 44GB of on-chip SRAM and 21 PB/s of memory bandwidth.
In August 2026 Cerebras announced the CS-4, built on a "WSE-3 Turbo" wafer. Its press release gives 250 PFLOPS per wafer and 43.2 PB/s of memory bandwidth, with the same transistor and core counts. Its headline claim of up to 30 times faster than GPU solutions comes with Cerebras's own caveat that "actual throughput varies by model architecture, context length, precision, and serving configuration."
Reading the claims
These are vendor figures from Cerebras. For the full comparison, see Cerebras vs NVIDIA. To see how you program it, read whether Cerebras supports PyTorch, and for the field, NVIDIA's main competitors in AI.
