Gefion: The First AI Supercomputer

Powering New Era of Innovation.

How Gefion Is Transforming Danish Research, Industry, and the Global AI Landscape

Keywords: Gefion AI supercomputer, Denmark AI, Danish supercomputer, NVIDIA DGX SuperPOD, AI innovation Denmark, renewable energy supercomputer, AI research Denmark, quantum computing Denmark

Denmark has taken a giant leap into the future of artificial intelligence with the launch of Gefion, its most advanced AI supercomputer. Officially ranked 21st on the global TOP500 list, Gefion is not only Denmark’s first AI-dedicated supercomputer but also a catalyst for innovation across research, industry, and society.

Gefion is built on the NVIDIA DGX SuperPOD architecture, powered by 1,528 NVIDIA H100 Tensor Core GPUs and connected via NVIDIA Quantum-2 InfiniBand networking. This setup delivers multi-exaflop performance, enabling researchers to tackle complex AI workloads, train large-scale models, and accelerate breakthroughs in record time.

In line with Denmark’s commitment to green technology, Gefion operates entirely on renewable energy, making it one of the world’s most sustainable supercomputing facilities.

• A National Asset for AI Sovereignty: Gefion is operated by the Danish Centre for AI Innovation (DCAI), a public-private partnership funded by the Novo Nordisk Foundation and the Export and Investment Fund of Denmark. This ensures that Denmark has sovereign control over its AI infrastructure, allowing it to develop solutions tailored to local needs, language, and culture

Gefion’s immense computational resources are already being applied to:

Pushing the boundaries of quantum simulations, bringing Denmark close to “quantum supremacy” and supporting companies like Kvantify in quantum chemistry and drug discovery

• Life Sciences and Drug Discovery: Accelerating the development of new medicines and treatments through AI-powered simulations and data analysis
• Climate Science and Green Transition: Supporting the Danish Meteorological Institute in developing advanced weather and climate models, and aiding the transition to a greener economy with AI-driven solutions
• Providing Danish businesses with the computational muscle to stay competitive globally, develop tailor-made solutions, and drive commercialization of research

Remarkably, Gefion went from project disclosure to operational status in just six months — a testament to Denmark’s strategic focus and collaborative spirit. The supercomputer is now entering a pilot phase, with select customers and researchers already exploring new AI use cases in fields ranging from deep learning to societal challenges

Gefion is more than a technological milestone; it’s a factory of intelligence that positions Denmark at the forefront of the global AI revolution. As AI continues to reshape industries, having a sovereign, sustainable, and world-class supercomputing resource gives Denmark a decisive edge in research, innovation, and economic growth.

Gefion is not just Denmark’s supercomputer — it’s the engine driving the nation’s AI-powered future.

: Gefion AI supercomputer, Denmark AI, Danish supercomputer, NVIDIA DGX SuperPOD, AI innovation Denmark, renewable energy supercomputer, AI research Denmark, quantum computing Denmark

From a technical perspective, Gefion is built from 191 NVIDIA DGX H100 systems, each housing 8 NVIDIA H100 Tensor Core GPUs, for a total of 1,528 H100 GPUs. These GPUs are interconnected using NVIDIA Quantum-2 InfiniBand networking, ensuring ultra-fast data transfer and scalability for demanding AI and HPC workloads

• (based on the Hopper architecture) are among the most advanced AI accelerators available. Each H100 is designed to excel at large-scale AI training, inference, and high-performance computing.
• Up to 51.2 teraflops (FP64 vector) and 102.4 teraflops (FP64 tensor core) per GPU for double-precision tasks.
• For AI workloads at reduced precision (FP16), the aggregate system performance reaches 1.51 exaflops; this can double with FP8 precision, and double again if sparsity is exploited

also includes two Intel “Sapphire Rapids” Xeon SP CPUs and high-speed networking, making each unit a powerful standalone AI workstation or a node in a massive distributed system: The NVIDIA DGX H100 system has a market price of approximately $500,000 to $600,000 USD per unit (2024 pricing; actual prices may vary based on configuration and contracts). Each unit contains 8 H100 GPUs, so the cost per GPU is in the range of $60,000 to $75,000 USD.: For 191 DGX H100 units, the hardware cost alone is estimated at $95 million to $115 million USD. This does not include additional infrastructure, storage, networking, installation, or operational costs, which can be substantial for a supercomputer of this scale.: Each DGX H100 node can independently train large language models, run generative AI workloads, or perform deep learning at a scale suitable for enterprise and research needs.: When networked as in Gefion, these units collaborate to train models with billions of parameters, simulate quantum systems, accelerate drug discovery, and process massive datasets in parallel.

• Hybrid Quantum-Classical Workflows: With support for platforms like NVIDIA CUDA Quantum, each unit can participate in hybrid workflows that combine classical and quantum computing resources, crucial for next-generation research in quantum chemistry and materials science

ComponentQuantityUnit Cost (Estimate)Total Hardware Cost (Estimate)Key CapabilitiesDGX H100 Systems191 units$500k–$600k$95M–$115M8x H100 GPUs, 2x Xeon CPUs, InfiniBand, AI/HPC readyNVIDIA H100 GPUs1,528 units$60k–$75kIncluded aboveUp to 102.4 TFLOPS (FP64 tensor), multi-exaflop AI

Gefion’s architecture and investment put Denmark at the forefront of AI and HPC, offering researchers and industry access to world-class computational resources for breakthroughs in AI, quantum computing, and beyond