← Capa | Mercado

Crusoe Lands Multi-Year Deal to Train and Serve Perplexity's Frontier Models

2026-11-04 · Fonte:Unite AI

Crusoe Lands Multi-Year Deal to Train and Serve Perplexity's Frontier Models

Perplexity will run its full model lifecycle on Crusoe Cloud, training on NVIDIA GB300 NVL72 clusters and serving via Managed Inference; Crusoe's 1,800 staff adopt Perplexity Enterprise Pro and Max.

AI infrastructure firm Crusoe has signed a multi-year partnership with AI search company Perplexity, announced on September 15, 2026. Under the agreement, Perplexity will consolidate its entire model lifecycle on Crusoe Cloud: frontier-model training will run on dedicated NVIDIA GB300 NVL72 clusters, while production serving will be delivered through Crusoe's Managed Inference service. The structure places both halves of Perplexity's compute operation — research-scale training and live product inference — under a single infrastructure provider.

The deal, datelined San Francisco, spans multiple years, giving Perplexity reserved access to GB300 NVL72 hardware rather than shared or on-demand capacity. That commitment covers training of frontier models — a workload class that typically requires dedicated accelerator capacity — and extends through deployment, since the company's production workloads will be served through the managed platform for the life of the contract. Reserved, dedicated clusters of this type differ from spot or shared arrangements in that a customer's training runs do not compete with other tenants for accelerator time.

The arrangement also flows in the opposite direction. In the same release, Crusoe committed to adopting Perplexity Enterprise Pro and Max across its workforce of 1,800 employees. Per the announcement, the deployment is meant to equip staff with the enterprise subscriptions, effectively turning the GPU supplier into a paying software customer of the AI developer it serves.

For Crusoe, the agreement secures a multi-year tenant for its GPU cloud, with dedicated frontier-training clusters as the anchor of the commercial relationship. For Perplexity, consolidating training and serving with one supplier simplifies operations and shifts production uptime responsibility onto Crusoe. The two-way structure — compute capacity in one direction, enterprise subscriptions in the other — illustrates how AI infrastructure deals increasingly bundle hardware access with reciprocal software adoption between partners.

Tags: Crusoe · Perplexity · NVIDIA GB300 NVL72 · Managed Inference · AI Cloud


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado