EU will increase its assistance to AI startups so they can use its supercomputers to train models

EU will increase its assistance to AI startups so they can use its supercomputers to train models

According to an update from the EU, France’s Mistral AI is taking part in an early pilot phase of an EU plan to support domestic AI startups by giving them access to processing power for model training on the bloc’s supercomputers. The plan was announced back in September and began last month. However, one early finding is that in order to teach AI startups how to maximize the high performance computing available in the EU, the program needs to include dedicated support for them.

“One of the things that we have seen is the need, not only to provide access but, to provide facility — especially skills, knowledge and experience that we have in the hosting centres — on how this access can be not only facilitated but to develop training algorithms that are using the best of the architecture and the computing power that is available right now in each supercomputing center and in our machines,” said an EU official speaking during a press briefing today.

They also mentioned that the establishment of “centers of excellence” is intended to aid in the creation of specific AI algorithms that are compatible with the supercomputers of the European Union.

Instead of using supercomputers’ processing power as a training resource, AI startups are more likely to be used to using specialized compute hardware supplied by American hyperscalers. According to EU officials speaking on background ahead of the official ribbon-cutting for MareNostrum 5, a pre-exascale supercomputer that will be inaugurated on Thursday at the Barcelona Supercomputing Center in Spain, the high-performance computing access for AI training program is therefore being enhanced with a support wrapper.

“We’re developing facilities for our SMEs to be able to understand how to best use the supercomputers and how to access the supercomputers and how to parallelize their algorithms in the case of AI to be able to develop their models,” said a Commission official. “As of 2024, we expect much more of these kinds of approaches than we have right now.”

“AI is considered now a strategic priority for the Union,” they added. “With AI becoming a strategic priority, next to the AI Act, we’re providing the innovation capacity — or we want to provide a large innovation window for our SMEs and startups to be able to best use our machines and this public infrastructure we’ve been creating so that they can compete internationally in developing safe, trustworthy and ethical AI algorithms.”

An “AI support center” is on the way, confirmed another EU official — saying this will have “a special track” for SMEs and startups to get help to get the most out of the EU’s supercomputing resource. “What we need to recognise is that the AI community have not been using supercomputers for the last decade,” they noted. “They’re not new users to GPUs but they’re new users to how to engage with a supercomputer and therefore we need to help them.

“In many cases the AI community comes out of a huge knowledge about how many GPUs can you get into one box? And they’ve been very good at that. But what we have on the supercomputers is a lot of boxes with GPUs and there’s some extra skill-sets and some extra help that’s needed in order to scale out and use the supercomputer to its full potential.”

Over the past five or so years, the bloc has significantly increased its investment in supercomputers, building the hardware to a cluster of eight machines spread throughout the region. It also intends to connect these machines via terabit networks, forming a federated supercomputing resource that will be available in the cloud for users throughout Europe.

The first exascale supercomputers in the EU are also scheduled to go online in the coming years; the first will likely be located in Germany (probably next year) and the second in France (expected in 2025). In order to provide a hybrid resource that combines both types of hardware, the Commission plans to purchase a number of quantum simulators that will be co-located with supercomputers. This will allow the quantum computers to serve as, in its words, “accelerators” for the classical supercomputers.

A project called Destination Earth aims to simulate Earth’s ecosystems in order to better model climate change and weather systems. Another application being developed on top of the EU’s high-performance computing hardware is to create a digital twin of the human body, which is intended to advance medical science by facilitating drug development and possibly even enabling personalized medicine. Since the EU president announced this fall’s compute access for AI model training program, using its supercomputing resources to specifically jump-start AI startups has become a more recent strategic priority.

A competition aimed at European AI startups “with experience in large-scale AI models” was also announced by the bloc last month. Its name, the “Large AI grand challenge,” is intended to select up to four promising homegrown startups that will receive a total of 4 million hours of supercomputing access to support the development of foundational models. A €1 million prize fund is also set aside for the winners, who, according to the Commission, must publish their research findings or release their developed models under an open source license for noncommercial use.

The EU’s “supercompute for AI” program is still in its infancy, so it’s unclear how much model training can benefit from dedicated access just yet. (At the time of writing, Mistral had not responded to our request for comment.) However, the Commission hopes that by providing support to AI startups so they can benefit from its high-performance computing investment and by constructing supercomputer hardware that it claims will increasingly be acquired and configured with AI model training in mind, this will give a local AI ecosystem that is just getting started a competitive edge over hyperscaler-proximate U.S. AI giants.

“Since we do not have the large hyperscalers that Americans have in the case of training these kinds of foundational models we use our supercomputers and we will develop a new generation of supercomputers that will be more and more AI compliant,” noted a Commission official. “Not only the ones that we have right now but, as of 2024, the purpose would be that we will go in this direction — and have even more of our SMEs use the supercomputers for developing these foundational models.”

According to them, part of the strategy will be to buy “more specialized AI supercomputing machines, that will be more based on accelerators rather than the standard CPUs.”

It remains to be seen whether the EU’s AI support strategy aligns with or deviates from the goal of some member states to develop national AI champions. During the recent contentious negotiations to establish the bloc’s AI rulebook, France spearheaded a push for a regulatory carve-out for foundational models, which drew criticism from SMEs. However, Mistral’s early inclusion in the EU’s supercomputing access program might indicate a convergence of ideas.