Unlocking the power of NVIDIA GB200 NVL72: How Crusoe makes next-gen AI a reality

The AI revolution is powered by a new class of supercomputers, and NVIDIA GB200 NVL72 represents a monumental leap in performance. This is the hardware that will train the reasoning-based models of tomorrow and solve problems that were previously out of reach.

But with great power comes new engineering challenges. For builders, questions around new architectures, power density, and scale are valid and important.

At Crusoe, we see these not as hurdles, but as the exact challenges our vertically integrated AI infrastructure is purpose-built to solve. We’ve designed our entire stack (from our environmentally aligned energy sources to our AI-native cloud platform) to unlock the full potential of NVIDIA GB200 NVL72. This is how Crusoe provides a strategic advantage for builders pushing the boundaries of what's possible.

Seamlessly migrate to the NVIDIA Grace™ CPU

The shift to the Arm-based NVIDIA Grace™ CPU is one of GB200 NVL72’s most powerful advancements, and can present some workload migration challenges. The critical question for engineers: will my code still run, and how much work will it take?

The Crusoe advantage: Our platform is designed to make this transition a planned, low-risk upgrade. The NVIDIA AI platform, including CUDA® and the major frameworks, have native Arm support, ensuring a familiar software environment where the vast majority of applications will run seamlessly.

More importantly, Crusoe de-risks the entire process. You can access GB200 NVL72 on our platform today and test your workloads in a hybrid deployment, running Grace (Arm) right alongside your existing x86 instances. This process allows for a smooth, staged transition, enabling you to validate performance and refactor only where necessary, all without disrupting your current workflows. It’s not a coding challenge; it’s a strategic, controlled evolution.

Improve reliability and efficiency with advanced liquid cooling

For the unprecedented compute density of the GB200 NVL72 requires liquid cooling for peak performance, reliability, and total cost of ownership (TCO) savings.

The Crusoe advantage: We are experts in deploying and managing advanced cooling technologies at scale. Our GB200 NVL72 clusters in Iceland are a testament to this, running today in one of the world's most innovative energy environments.

Powered by 100% renewable geothermal and hydro energy, the synergy between our advanced, closed-loop liquid cooling systems and the naturally cool climate allows us to maximize efficiency. This isn't just theory; it's a live deployment that demonstrates how Crusoe's engineering cuts energy use by up to 25% while enabling the GB200 NVL72 to deliver up to 30x the performance of previous generations for certain workloads. With Crusoe, liquid cooling is a proven, reliable pathway to more sustainable and cost-effective AI.

Future-proof your ambition with a highly scalable architecture

A common question is whether the NVIDIA Blackwell architecture is only for massive, multi-thousand GPU clusters.

The Crusoe advantage: Today's large models are tomorrow's standard. We believe builders shouldn't have to re-architect their infrastructure every time their ambition grows. The GB200 NVL72 is a future-proof investment, and Crusoe's platform makes it accessible and scalable from day one.

Our rack-scale systems are designed to behave as a single, unified super-GPU. This gives you the power of a massive, coherent memory pool and advanced NVIDIA NVLink™ interconnects, even with a smaller initial deployment. As your needs grow, you can seamlessly scale up without hitting the performance bottlenecks of legacy architectures. Early workloads from our customers in finance and generative AI running on our GB200 NVL72 systems in Iceland are already demonstrating the power of this unified memory model for complex, data-intensive tasks.

Harness every performance advantage of the Blackwell architecture

The Blackwell architecture introduces cutting-edge capabilities, GB200 NVL72 delivers 30X faster real-time trillion-parameter LLM inference. 

The Crusoe advantage: Our platform is engineered to maximize every feature of the Blackwell architecture, ensuring you get the best possible performance out of the box.

•     Massive unified memory: GB200 NVL72 is equipped with fifth-generation NVLink and PCIe Gen5 for faster GPU-to-GPU communication. Our platform’s tooling and our team's expertise ensure your workloads are configured to leverage this unified memory pool to its fullest potential, which is critical for training and inferencing the largest models.

•     Next-generation NVIDIA NVLink™: With up to 1,800 GB/s of inter-GPU bandwidth, NVLink is a game-changer. Our purpose-built data centers and network fabric are designed to maximize this massive leap in bandwidth for parallel training and distributed inference at scale.

•     Real-time decompression: The advanced decompression engine accelerates analytics and data pipelines. We ensure our platform's out-of-the-box experience eliminates data pipeline bottlenecks, so you can leverage this feature from day one.

Build the future faster with Crusoe

The NVIDIA GB200 NVL72 represents a strategic, future-proof investment for any organization in the AI race. The perceived complexities of migrating architectures is precisely the challenges our vertically integrated AI infrastructure is designed to eliminate.

At Crusoe, we provide a fully integrated, purpose-built platform that is the ideal environment for GB200 NVL72. You can access GB200 NVL72 on our platform now, with trial runs available to minimize migration risk. Test your workloads and see the performance gains firsthand before making any long-term commitment.

Download our full NVIDIA GB200 NVL72 explainer or contact our experts today to discuss your specific AI workloads and discover how Crusoe, accelerated by NVIDIA Blackwell, can accelerate your path to innovation.