Cerebras Claims AI Inference 30x Faster Than GPUs

Cerebras Systems CEO Andrew Feldman says the company has built the fastest AI inference in the industry, outpacing GPUs by 10 to 30 times. The company's wafer-scale chip architecture keeps model weights in on-chip SRAM, avoiding memory bottlenecks that slow conventional GPU systems. As agentic AI workflows drive demand for faster, more iterative reasoning, Cerebras counts OpenAI, Cognition AI, Block, and GlaxoSmithKline among its customers. The company plans to scale manufacturing capacity by eight to ten times this year to support global data center expansion.