The European High-Performance Computing (EuroHPC) Joint Undertaking is a collaborative initiative between the European Union, European countries, and private partners to develop a world-class European supercomputing ecosystem. The amendment to Council Regulation (EU) 2021/1173, existing EU legislation related to the joint procurement of high-performance supercomputers, expands the capabilities of the EuroHPC in response to rapid developments in AI.

The European Union's high-performance computing (HPC) strategy and AI strategy has faced challenges in adapting to the rapid emergence of generative artificial intelligence (AI) technologies. The EU's network of supercomputers has primarily catered to scientific research. However, these systems were not initially optimized for training large-scale generative AI models, revealing a significant gap in the EU's AI strategy that policymakers are now working to address.

The EU's approach to rectifying this situation involves two key elements:

1. Infrastructure optimization: Investments are being directed towards enhancing existing supercomputing capabilities and reconfiguring them to support generative AI model training.
2. Democratized access: The EU is maintaining its policy of providing free access to its supercomputing resources for research purposes and extending this to AI startups and SMEs engaged in model pre-training or training.

The amendment to the Council Regulation is a first step in tackling this situation. The key changes introduced by the amendment include:

1. New definitions: The regulation introduces definitions for "AI-dedicated supercomputer" and "AI Factory," expanding the scope of EuroHPC supercomputers.
2. Additional objective: A new objective is added for EuroHPC JU to develop and operate AI Factories. The factories refer to large-scale, high-performance computing facilities specifically designed and optimized for AI workloads. AI factories can be used for a wide range of AI applications, including natural language processing, computer vision, scientific simulations, and more.
3. New pillar of activity: The amendment establishes an "AI Factory pillar" focused on providing AI-oriented supercomputing services, including the acquisition and operation of AI-dedicated supercomputers, upgrading existing supercomputers with AI capabilities, and providing access to these resources.
4. Selection criteria for hosting entities: New criteria are introduced for selecting hosting entities (organizations or institutions that are selected to house, operate, and manage the supercomputers and related infrastructure) for AI-dedicated supercomputers. Factors such as proximity to data centres and capabilities in supporting the AI ecosystem are herein emphasized.
5. Acquisition and ownership: The Joint Undertaking will acquire and own AI-dedicated supercomputers, with the EU covering up to 50% of acquisition and operating costs.
6. Access and usage: AI-dedicated supercomputers will primarily be used for developing, testing, and validating large-scale AI models and applications. Special access conditions will be defined for AI startups and the research and innovation ecosystem Such access will be facilitated by a yet-to-be-established one-stop shop, which should also help SMEs and scientific users.
7. Emphasis on trustworthy AI: Only proposals for developing trustworthy and ethical AI models aligned with EU values will be eligible for access to these resources.
8. Collaboration and talent development: The initiative encourages interaction between AI Factories, EuroHPC Competence Centres (regional hubs to promote and facilitate the use of high-performance computing across various sectors), and other EU AI initiatives, as well as efforts to attract and train talent in AI and supercomputing.
9. Upgrades to existing supercomputers: The amendment provides more flexibility for upgrading existing EuroHPC supercomputers, including enhancing their AI capabilities.
10. No additional EU budget: The initiative will be implemented through redeployment of existing EuroHPC JU resources, without requiring additional funding from the EU budget.

The success of the supercomputing initiative depends on two critical factors:

1. Rapid infrastructure upgrades: The EU must implement the necessary enhancements to its supercomputing facilities to meet the specific demands of AI model training. Access to accelerators (i.e. Graphic Processing Units (GPUs)) poses an additional challenge here
2. Resource allocation: With an anticipated increase in demand from AI startups seeking to utilize these free computing resources, effective management and allocation of supercomputing capacity will be crucial.

The effectiveness of the EU's supercomputing-for-AI strategy will largely depend on its ability to execute these infrastructure upgrades efficiently and manage the increased demand for computational resources.