TLDR Tech / AI
The Inference Shift
Cerebras' surging IPO signals a coming split between "answer inference" optimized for token speed and "agentic inference" optimized for memory hierarchy. Cerebras' WSE-3 has 44GB of on-chip SRAM at 21 PB/s, roughly 6,000 times the memory bandwidth of an H100, making it perfect for human-facing low-latency answers including voice and AI wearables but unsuitable when KV caches and model weights exceed on-chip capacity.
Read the full story