Perplexity Signs Multi-Year AI Infrastructure Deal With Crusoe, Uniting Training and Inference on NVIDIA GB300

AI infrastructure company Crusoe and AI search service Perplexity have entered a multi-year partnership. The agreement covers the entire AI service lifecycle, from model training to production inference, running on Crusoe's infrastructure, with each company contributing its own core strength to the collaboration.

Training and Inference on a Single Track

At the center of the deal are dedicated clusters powered by NVIDIA GB300 NVL72 systems that Crusoe operates. Using these clusters, which are connected via high-speed NVIDIA InfiniBand networking, Perplexity will train its frontier models. Once training is complete, the same models move directly into Crusoe's Managed Inference service for production use. The arrangement removes the need to source and manage separate infrastructure for training and inference from different vendors, which is one of the deal's defining features.

When Speed Becomes the Product

For Perplexity, inference speed and reliability are paramount. The company's AI search service reportedly handles more than 1.5 billion questions every month, and any delay in responses directly degrades the user experience. Unlike conventional search engines, Perplexity's answer generation involves heavy real-time computation, so latency and throughput are not just technical metrics — they define the quality of the product itself.

Crusoe Positions Itself as a Full-Stack AI Infrastructure Company

Crusoe describes itself as a vertically integrated AI infrastructure provider, handling everything from hardware procurement and data center operations to inference services. Unlike traditional cloud providers, it builds out power sourcing, data centers, and GPU clusters all in-house. As part of the partnership, Crusoe will also roll out Perplexity's enterprise plans, Enterprise Pro and Max, to its roughly 1,800 employees, deepening the relationship on both the infrastructure and product sides. Neither company has disclosed the value of the contract or the scale of computing resources involved.

AI-Focused Infrastructure Providers Gain Ground

As demand for generative AI grows, competition for GPU cluster capacity has intensified. Alongside major cloud providers, dedicated infrastructure companies like Crusoe — which combine power, facilities, and GPU operations into a single offering — are becoming a compelling option for AI companies. Training and production workloads demand different qualities from infrastructure, and it is not easy to excel at both at once. A partnership that spans training through inference on a single platform, as this one does, offers one path for AI service companies to reduce the burden of sourcing and managing infrastructure, freeing up resources to focus on improving response speed and reliability.

Summary

The Crusoe-Perplexity partnership is an attempt to connect two workloads with very different characteristics — AI model training and inference — on a single infrastructure platform. Perplexity stands to improve the responsiveness of a search service handling roughly 1.5 billion queries a month, while Crusoe gains a proof point for its infrastructure capabilities. As generative AI adoption continues to expand, this kind of deepening interdependence between infrastructure specialists and AI service companies is likely to become more common.