The European Union’s initiative to bolster domestic AI startups by granting them access to processing power on the bloc’s supercomputers, introduced in September and initiated last month, has involved the participation of Mistral AI from France in an early pilot phase, as per the latest update from the EU. However, a key takeaway from the early phase is the necessity to incorporate dedicated support for AI startups to train them in optimizing the use of the EU’s high-performance computing.
An EU official, speaking during a press briefing, emphasized the importance of not only providing access but also offering facilities, including skills, knowledge, and experience from hosting centers. The goal is to guide AI startups in developing training algorithms that leverage the architecture and computing power available in each supercomputing center.
The plan involves establishing “centers of excellence” to support the creation of specialized AI algorithms compatible with the EU’s supercomputers. Given that AI startups typically utilize compute hardware from U.S. hyperscalers for model training, the program is complemented with a support framework to assist startups in utilizing the processing power offered by supercomputers effectively.
A Commission official highlighted the development of facilities to help SMEs understand how to optimize supercomputers, access them, and parallelize their algorithms for AI model development. The EU aims to make such approaches more widespread by 2024, recognizing AI as a strategic priority and providing innovation capacity for SMEs and startups to compete globally in developing safe, trustworthy, and ethical AI algorithms.
The EU plans to establish an “AI support center” with a dedicated track for SMEs and startups to maximize the use of supercomputing resources. Acknowledging that the AI community is relatively new to supercomputers, the EU emphasizes the need for additional skills and support to fully utilize the potential of supercomputers.
The EU has significantly increased its investment in supercomputers over the past five years, with plans to interconnect them to create a federated supercomputing resource accessible in the cloud across Europe. The introduction of the EU’s first exascale supercomputers in Germany and France in the coming years and investments in quantum computing further underscore the strategic focus on advanced computing technologies.
The EU’s recent emphasis on supporting AI startups, as demonstrated by the supercomputing access program and the “Large AI grand challenge,” indicates a concerted effort to leverage its growing supercomputing network as a strategic asset for advancing “Made in Europe” general-purpose AI. While Mistral AI’s inclusion in the program may raise eyebrows given its recent substantial funding, it underscores the EU’s commitment to promoting large AI initiatives. The EU envisions the integration of AI into its supercomputing infrastructure, with plans to acquire more dedicated AI supercomputing machines based on accelerators, aligning with the broader goal of fostering a competitive local AI ecosystem.
