Denmark has officially joined the AI supercomputing race with Gefion, its first national AI supercomputer. King Frederik X and Nvidia’s CEO, Jensen Huang, were on stage in Copenhagen to power up Gefion, a project that combines Nvidia’s DGX SuperPOD architecture with a setup of 1,528 H100 GPUs. Built through a collaboration between Nvidia and the Danish Centre for AI Innovation (DCAI), Gefion will support research across various fields including healthcare, environmental science, and quantum computing. Its hardware design allows for rapid processing of complex AI models, aimed at propelling Denmark’s scientific and industrial research.
Nvidia’s DGX Cloud and AI Infrastructure
Before Denmark’s launch of Gefion, Nvidia had expanded access to high-performance AI through its DGX Cloud platform, a cloud-hosted supercomputing service that became widely available in July 2023. The DGX Cloud provides a monthly subscription model, allowing businesses to rent powerful AI resources without needing their own supercomputer setup. The cloud platform initially launched on Oracle’s cloud infrastructure and is set to expand to Microsoft’s Azure, followed by Google Cloud, reflecting Nvidia’s approach to making AI more accessible for industries lacking dedicated infrastructure.
Each DGX Cloud instance comprises 80-gigabyte Tensor Core GPUs linked through a high-performance networking framework, creating a configuration that can handle immense data processing. Nvidia’s cloud infrastructure supports training complex AI models used in various sectors, from healthcare and finance to logistics. Gefion’s in-country infrastructure complements DGX Cloud by enabling Danish organizations to maintain data sovereignty while accessing a similar scale of computational resources domestically.
Partnerships Powering Denmark’s AI Future
Gefion is operated by the Danish Centre for AI Innovation, a collaboration established through substantial investments from Denmark’s Novo Nordisk Foundation and the Export and Investment Fund of Denmark (EIFO). With Novo Nordisk’s contribution of DKK 600 million ($87 million) and EIFO’s DKK 100 million ($14.5 million), the Danish government now holds a 15% stake in DCAI. DCAI’s leadership, headed by quantum computing expert Nadia Carlsten, sees Gefion as a model of national AI autonomy. By creating a sovereign AI infrastructure, DCAI aims to support research and development tailored to Danish interests, allowing the country to operate independently from third-party tech providers for critical data applications.
Denmark’s investment strategy reflects an aim to become self-sufficient in AI research and infrastructure. Carlsten highlighted that Gefion’s domestic location allows Denmark’s researchers to retain control over sensitive data, including research in health and climate. Gefion’s resources will serve a range of organizations across Denmark, opening up AI applications for research, industry, and education sectors alike.
Key Pilot Projects in Science and Industry
Gefion’s initial rollout already involves notable organizations such as the Danish Meteorological Institute (DMI) and the University of Copenhagen. DMI plans to use Gefion’s processing power to make weather forecasting faster and more energy-efficient, while university researchers aim to simulate quantum circuits that bring Denmark closer to achieving “quantum supremacy.” This term denotes a quantum computer’s ability to perform calculations beyond what’s possible with conventional computers, which Gefion’s computing power will help approach.
Collaborative research projects, such as a new genomic foundation model led by Danish universities and Novo Nordisk, are also underway. This project seeks to improve the ability to identify disease mutations and develop vaccines by training large language models (LLMs) that analyze genomic data. Gefion’s resources will enable this genomic model to run at scale, providing deeper insights into genetic research.
AI Development Beyond Denmark: A Look at Global Supercomputers
Gefion’s activation comes amid a global expansion of supercomputing power, with recent entries like Elon Musk’s xAI Colossus supercomputer and Cerebras and G42’s Condor Galaxy network. Colossus, located in Tennessee, includes 100,000 Nvidia H100 GPUs and is designed to train Musk’s large language model, Grok-3, with plans to expand its capacity further by 2024. Cerebras, meanwhile, has partnered with UAE-based G42 to launch Condor Galaxy, a network that will ultimately consist of nine interconnected AI supercomputers. The first, Condor Galaxy 1, boasts 4 exaFLOPs of computing power and supports AI-driven applications in healthcare, climate science, and language processing.
Gefion also aligns with international initiatives such as IBM’s integration of quantum computing with the Fugaku supercomputer in Japan, which aims to push the limits of hybrid quantum-classical computing. These developments underline an ongoing shift in the tech landscape, as countries and corporations build powerful AI infrastructures to support high-stakes research and industrial applications.
Expanding AI Reach in Danish Research and Startups
Alongside major research bodies, Danish startups are gearing up to use Gefion’s computational power to develop AI-driven solutions. Local firm Go Autonomous, for instance, is focused on models that can interpret mixed data inputs, while Teton, another startup, is building an AI tool for healthcare applications. With Gefion, both companies can train models on a larger scale, expanding the potential of Danish startups in AI-driven industries.
Gefion’s domestic availability grants Denmark greater control over AI development, offering a national-level AI infrastructure that aligns with Danish regulatory standards. As Gefion becomes operational, it will likely influence the direction of AI research and development not only in Denmark but across Europe. Through Nvidia’s collaboration and DCAI’s oversight, Gefion provides a unique framework for balancing local needs with the advantages of global-scale AI infrastructure.
Last Updated on November 7, 2024 2:20 pm CET