EU Expands Support for AI Startups, Offering Supercomputers for Model Training
-
The European Union plans to support local AI startups by providing supercomputing power for model training. The initiative was announced last September and implemented last month.
According to the latest EU news, France's Mistral AI has participated in the early pilot phase. However, early experiences indicate that the program needs to provide specialized support for AI startups to train them on how to fully utilize the EU's high-performance computing capabilities.
An EU official stated at today's press conference: "One issue we've observed is that we not only need to provide access but also facilities, particularly in terms of the skills, knowledge, and experience available at hosting centers, to help AI startups develop training algorithms using the best architectures and computing power." They added that plans are underway to establish "centers of excellence" to support the development of specialized AI algorithms capable of running on the EU's supercomputers.
Image source note: Image generated by AI, image licensed by Midjourney
Compared to using supercomputers as training resources, AI startups are more accustomed to using dedicated computing hardware provided by US hyperscale vendors. Therefore, according to EU officials, plans to use high-performance computing resources for AI training will include support components. "We are developing facilities for our SMEs to help them understand how to best utilize supercomputers, how to access supercomputers, and how to parallelize their algorithms to develop their own models," said a Commission official. "By 2024, we expect more such projects."
They added: "AI is now considered a strategic priority for the EU. At a time when AI has become a strategic priority, we provide innovation capabilities, or we hope to offer our SMEs and startups broad space for innovation, so they can fully utilize our machines and the public infrastructure we are creating to compete internationally in developing safe, trustworthy, and ethical AI algorithms."
Another EU official confirmed the upcoming "AI Support Center" and stated that it would provide a "special track" for SMEs and startups to help them make the most of the EU's supercomputing resources. "What we need to recognize is that the AI community has not been using supercomputers over the past decade," they noted. "They are familiar with GPUs, but they still need some additional skills and assistance to interact with supercomputers and fully realize their potential."
Over the past five years, the European Union has significantly increased its investments in the field of supercomputing, developing the hardware into a cluster of eight machines distributed across Europe. Plans are underway to interconnect these via gigabit networks, creating a unified supercomputing resource accessible via the cloud to serve users throughout Europe.
Applications being developed on the EU's high-performance computing hardware include a project to simulate Earth's ecosystem for better modeling of climate change and weather systems, named "Destination Earth," as well as a project to design digital twins of the human body. These initiatives aim to advance medical science by supporting drug development and even enabling personalized medicine. Following the EU President's announcement of a plan to provide computational access for AI model training, leveraging supercomputing resources to boost AI startups has become a recent strategic priority.
Last month, the EU also announced a competition called the "Large-Scale AI Grand Challenge," which aims to select up to four experienced European AI startups and provide them with a total of four million hours of supercomputing resources to support foundational model development. A prize fund of one million euros is expected to be awarded to the winners, who will release their developed models under open-source licenses or publish their research findings.
The EU already has programs that provide industry users with access to core computing resources through project applications. However, the EU is increasing its focus on commercial AI, offering specialized programs and resources to transform its growing supercomputing network into a strategic force driving "Made in Europe" general AI. Therefore, it's no coincidence that French AI startup Mistral AI has benefited from the EU's supercomputing access program. (Although, a tech company that just secured €385 million in Series A funding, including from US investors like Andreessen Horowitz, General Catalyst, and Salesforce, being at the front of the queue for EU computing resources might raise some eyebrows. But this is yet another sign of the high-level strategic bet on "large-scale AI.")
The EU's "AI Supercomputing" initiative is still in its early stages, so it's unclear whether positive progress has been made in model training (we contacted Mistral for comment but hadn't received a response by the time of writing). However, the European Commission hopes that by supporting AI startups to leverage its investments in high-performance computing, combined with plans to build supercomputing hardware increasingly targeting AI model training, this will provide a competitive advantage for the relatively disadvantaged local AI ecosystem.
"Since we don't have large-scale suppliers like the U.S. to train these foundational models, we're utilizing our supercomputers and will continue developing generations of supercomputers increasingly suited for AI," said a committee official. "We're not just working with what we have now—by 2024, our goal is to progress in this direction, enabling our SMEs to better leverage supercomputers for developing these foundational models."
It remains to be seen whether this plan aligns with or contradicts the ambitions of certain EU member states to nurture national AI champions. During recent challenging negotiations on EU AI regulations, France led efforts to push for regulatory exemptions for foundational models, a move criticized by SMEs. However, Mistral's early involvement in the EU's supercomputing access initiative may indicate a convergence of thinking.