NVIDIA GB300 NVL72 Power frontier AI workloads at rack scale with NVIDIA's most advanced Blackwell Ultra platform. Beyond.pl AI Panel Send an inquiry Immense power in a rack-scale system The NVIDIA GB300 NVL72 integrates 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs into a single, fully liquid-cooled rack, connected by a 72-GPU unified NVIDIA NVLink domain delivering 130 TB/s of low-latency GPU communication. This architecture eliminates the inter-node communication overhead typical of multi-server clusters, enabling the entire rack to operate as a single high-bandwidth compute platform. NVIDIA GB300 can be scaled up to tens of thousands of superchips, creating a massive shared memory space that accelerates the performance of the world’s largest AI models. Beyond.pl makes this level of computing performance available within an enterprise-ready environment designed for demanding AI workloads. Whether supporting a single large-scale deployment or a long-term AI transformation program, our AI Factory combines enterprise-grade infrastructure with the expertise needed to develop, deploy and scale AI projects. Contact us to explore its capabilities and discuss how we can support your AI project. Graphics: GB300 motherboard Source: NVIDIA The benefits of scaling AI infrastructure with Beyond.pl Faster access to AI capacity Launch AI projects without long infrastructure lead times. With 2 MW of GPU capacity available in Q1 2027, an additional 12 MW of high-density, liquid-cooled GPU capacity scheduled for H2 2027, and another 24 MW available in 2028, Beyond.pl provides a clear capacity roadmap for organizations planning to scale their AI infrastructure over the long term. Designed for large-scale Built for dedicated AI environments - not one-size-fits-all deployments. Beyond.pl supports large-scale AI projects with access to NVIDIA B300, GB300, GB200 or Vera Rubin platforms, enabling organizations to deploy the architecture that best matches their performance, operational, and growth requirements. AI infrastructure on your terms Your AI infrastructure should adapt to your business - not the other way around. We work closely with your team to deliver the right platform, capacity, and deployment model for your workloads, ensuring your AI environment evolves alongside your long-term strategy. Use cases for NVIDIA GB300 NVL72 Train larger AI models Accelerate the development of foundation models and large-scale AI architectures with infrastructure designed to handle the highest memory, compute, and networking demands throughout the training lifecycle. Scale reasoning and agentic AI Support AI systems that analyse, plan, and evaluate multiple outcomes before responding. GB300 NVL72 provides the performance required for advanced reasoning, agentic workflows, and compute-intensive inference. Deliver enterprise AI at scale Deploy AI applications that serve thousands of concurrent requests while maintaining high throughput and predictable latency. With large GPU memory capacity and rack-scale architecture, GB300 NVL72 enables efficient production inference for modern AI services. NVIDIA GB300 NVL72 specification Download datasheet Form factor Fully liquid-cooled, rack-scale system GPU 72 Blackwell Ultra GPUs 36 NVIDIA Grace CPUs GPU Memory | Bandwidth 20 TB HBM3e | 576 TB/s NVLink Bandwidth 130 TB/s Fast Memory 37 TB CPU Cores 2,592 Arm Neoverse V2 cores FP4 Tensor Core Performance 1,440 PFLOPS | 1,080 PFLOPS FP8/FP6 Tensor Core Performance 720 PFLOPS Individual Blackwell Ultra GPU Specification FP4 Tensor Core per GPU 20 PFLOPS | 15 PFLOPS FP8/FP6 Tensor Core per GPU 10 PFLOPS INT8 Tensor Core per GPU 330 TOPS FP16/BF16 Tensor Core per GPU 5 PFLOPS TF32 Tensor Core per GPU 2.5 PFLOPS FP32 per GPU 80 TFLOPS FP64/FP64 Tensor Core per GPU 1.3 TFLOPS GPU Memory per GPU 279 GB HBM3e GPU Memory Bandwidth per GPU 8 TB/s Interconnect Fifth-generation NVIDIA NVLink™: 1.8 TB/s Source: NVIDIA datasheet The launch of the first commercial AI Factory in CEE In the second half of 2025, Beyond.pl launched the first commercial AI Factory in the CEE region in collaboration with NVIDIA. The platform was built on NVIDIA DGX SuperPOD reference architecture powered by NVIDIA B200 GPUs, providing enterprise-grade infrastructure for AI training and inference workloads Read the press release Explore more NVIDIA AI infrastructure for your next deployment NVIDIA B300 Next-generation accelerated compute for advanced AI applications. NVIDIA B300 supports demanding AI workloads across training, fine-tuning and inference. It is designed for organizations that need scalable performance, enterprise-grade reliability and high-performance NVIDIA accelerated computing for large-scale AI deployments. Explore NVIDIA GB200 Rack-scale AI infrastructure for large, distributed workloads. NVIDIA GB200 is built for organizations developing and operating complex AI environments at scale. It supports large model training, advanced inference and compute-intensive workloads that require tightly integrated GPU, networking and software resources. Explore NVIDIA B200 High-performance compute for training, fine-tuning and inference of advanced AI models. NVIDIA B200 is designed for organizations running demanding generative AI, large language model and high-performance computing workloads. It provides access to NVIDIA Blackwell architecture in an integrated enterprise AI system. Explore FAQ What is NVIDIA GB300 NVL72? The NVIDIA GB300 NVL72 is a fully liquid-cooled, rack-scale system built on NVIDIA Blackwell Ultra architecture. It integrates 72 Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs into a single platform, connected by a 72 NVLink technology, purpose-built for AI reasoning, frontier model training, and large-scale inference. What is the key difference between NVIDIA GB300 NVL72 and NVIDIA B300? GB300 NVL72 is a rack-scale system where 72 GPUs are unified in a single NVLink domain, delivering up to 130 TB/s of intra-rack bandwidth and enabling the entire rack to operate as a single high-bandwidth compute platform. In contrast, NVIDIA B300 is an 8-GPU AI server that scales out by connecting multiple nodes through high-performance cluster networking, such as NVIDIA InfiniBand or Spectrum-X Ethernet. GB300 NVL72 is ideal for workloads that benefit from the tightest possible GPU integration and maximum performance, while B300 is better suited for flexible, modular, horizontally scalable deployments. What is the key difference between NVIDIA GB300 NVL72 and NVIDIA GB200 NVL72? GB300 NVL72 is built on Blackwell Ultra, the next generation after Blackwell. Each Blackwell Ultra GPU offers 1.5× more HBM3e memory, 1.5× more dense FP4 compute, and 2× higher attention-layer performance compared to the Blackwell GPU in GB200 NVL72. GB300 NVL72 is purpose-built for AI reasoning workloads and delivers higher throughput per rack for the similar power scope. Is NVIDIA GB300 NVL72 available as a self-service offering? No. NVIDIA GB300 NVL72 is deployed as a tailored infrastructure solution. Because each rack-scale deployment is designed around specific workload and performance requirements, our team will help you determine the right configuration and deployment timeline. How do I get access to NVIDIA GB300 NVL72? Contact the Beyond.pl technical team to explore the best deployment strategy for your AI initiatives. We’ll help you assess workload demands and build a solution tailored to your requirements. Ready to run AI at scale? Deploy demanding AI workloads on NVIDIA GB300 NVL72 infrastructure at Beyond.pl AI Factory. Check available capacity and discuss dedicated compute designed around your AI workloads and long-term requirements. Contact us
