Crusoe Signs Multi-Year Deal with Perplexity to Power Frontier AI Models on NVIDIA GB300 NVL72 Clusters

Crusoe and Perplexity announced a major multi-year infrastructure partnership on September 15, 2026. Under the agreement, Perplexity will run its entire model lifecycle on Crusoe Cloud, deploying dedicated NVIDIA GB300 NVL72 liquid-cooled supercomputing clusters to train next-generation frontier search models and serving real-time production traffic via Crusoe Managed Inference.

Full Lifecycle AI Deployment on Next-Gen NVIDIA Hardware

The collaboration unites Perplexity conversational search engine with Crusoe climate-aligned AI compute infrastructure. Rather than splitting development across disparate clouds, Perplexity is standardizing on Crusoe Cloud to handle both massive pre-training workloads and production-scale query answering. The training tier utilizes dedicated clusters of NVIDIA GB300 NVL72 systems, interconnecting cutting-edge Blackwell Ultra GPUs over high-bandwidth NVLink backplanes to accelerate training runs for complex multi-step reasoning models.

Infrastructure Highlights

• Dedicated NVIDIA GB300 NVL72: Scalable compute clusters providing high-density FP4 and FP8 precision tensor throughput for frontier model training.
• End-to-End Managed Inference: Custom kernel optimizations and dynamic batching built specifically to cut end-to-end token latency for live search queries.
• Clean Energy Compute: Data centers powered by stranded, renewable, and clean energy sources to minimize carbon footprint at hyperscale.
• Cross-Company Adoption: Crusoe will deploy Perplexity Enterprise Pro and Max across its 1,800-employee workforce for research and internal knowledge discovery.

Meeting Production Latency and Throughput Demands

For real-time search engines serving millions of active users daily, inference latency directly impacts user engagement and answer quality. Crusoe Managed Inference provides hardware-level optimizations tuned specifically for the low-latency requirements of agentic reasoning, web synthesis, and real-time citations. By co-locating frontier model checkpoints directly within Crusoe high-throughput data fabric, Perplexity significantly reduces transit overhead between model weight updates and live production endpoints.

Product Profile Sheet