tl;dr: Cerebras announced the CS-4, a rack-scale AI system packing three WSE-3 Turbo wafers that claims up to 30x faster inference than GPUs and 10x better throughput per watt than the prior CS-3. Key architectural changes include a modular "backpack" design with power delivery just 0.5mm from the processor, a new programmable I/O subsystem enabling 2-microsecond wafer-to-wafer latency, and support for 1,000+ tokens/sec on 10T+ parameter models. First shipments begin this quarter.