Cerebras has deployed trillion-parameter frontier AI models by mapping them across its WSE-3T wafers. This approach enables highly scalable inference, achieving impressive generation speeds of 1,000 to over 2,000 tokens per second.