Perplexity is moving its entire model workflow to Crusoe Cloud:
• training frontier models on dedicated NVIDIA GB300 NVL72 clusters;
• deploying models to production via Crusoe Managed Inference;
• a single platform spanning from training to inference.
Perplexity processes over 1.5 billion queries per month, and every millisecond of latency is noticeable to the user. Consequently, the inference provider directly determines the quality of the product.
The deal is partially reciprocal: Crusoe is adopting Perplexity Enterprise Pro and Max for its 1,800 employees. The parties have not disclosed the financial terms.
Why is this a significant signal?
In early September, Bloomberg reported on a five-year contract between Crusoe and Jane Street valued at approximately $13 billion. While the companies did not officially comment on the deal, it represents the largest client to date for Crusoe’s cloud business.
Crusoe’s client base is expanding across various segments, and managed inference is emerging as a standalone business line alongside its compute-for-training rental services.