Crusoe expands NVIDIA collaboration across AI infrastructure stack
Crusoe has enhanced its collaboration with NVIDIA to support multiple NVIDIA hardware and software components integral to its Artificial Intelligence (AI) infrastructure. This includes the addition of NVIDIA Vera CPUs, Nemotron 3 Super and VoiceChat models, and integration with NVIDIA's Omniverse DSX Blueprint, along with a contribution to the NVIDIA Dynamo open-source framework.
The collaboration addresses operational demands for AI environments that require simultaneous Central Processing Unit (CPU) and Graphics Processing Unit (GPU) performance alignment, supporting concurrent sandbox environments at scale. Crusoe's use of NVIDIA hardware and design standards aims to facilitate efficiency and energy optimization in AI infrastructure deployment.
Technically, Crusoe plans deployment of the NVIDIA Vera CPU in its cloud services alongside GPUs like NVIDIA Rubin and Vera Rubin NLV72, with the Vera CPU designed to support over 22,500 concurrent environments per rack. The company intends to incorporate the Vera Rubin DSX Reference Design principles for AI factory layouts, power, cooling, and mechanical integration to maintain infrastructure redundancy and adaptability for future chip advances.
In terms of AI models, Crusoe is an early adopter of the NVIDIA Nemotron 3 Super, a hybrid Mixture of Experts (MoE) model capable of processing up to one million token context windows, and the Nemotron 3 VoiceChat model, a speech-to-speech solution supporting simultaneous talking and listening. These models are accessible through Crusoe's Managed Inference service, which integrates MemoryAlloy technology to improve inference speed and throughput. Additionally, Crusoe has integrated its tokenizer into the NVIDIA Dynamo library, offering performance improvements in token processing and inference start times.
Erwan Menard, Senior Vice President of Product for Crusoe Cloud, said, “As an early adopter of NVIDIA Nemotron 3 Super and NVIDIA Nemotron 3 VoiceChat, and with the Crusoe Managed Inference tokenizer integration available in NVIDIA Dynamo and open to the community, we're giving AI-native developers everything they need to move from model to production without compromise.” He added, “Open models, open infrastructure, and performance that scales — that's the architecture of immediacy in practice.”
Chase Lochmiller, Co-Founder and CEO at Crusoe, said, “We’re building every layer of Crusoe's stack in lock step with NVIDIA's roadmap in order to manufacture intelligence with unprecedented speed and scale.” He also stated, “We’re aligning at every level – from the physical design of our AI factories to how we support their latest models and hardware. This tight integration is how we continue to deliver the exceptional infrastructure solutions our customers require. Crusoe is proud to work with NVIDIA to power the future of agentic AI.”