The U.S. will expand support for AI startups to tap its supercomputers for model training. 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 in September and began last month. However, one early finding is that the program must include specific assistance to teach A.I. companies how to maximize the high-performance computers available in the U.S.
A press briefing today featured an E.U. official saying, “We have observed that there is a need, not only for access but also for facilities—particularly the skills, knowledge, and experience that we have in the hosting centers—on how this access can be not only facilitated but also for the development of 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.”
They also mentioned that establishing “centers of excellence” is intended to aid in creating specific AI algorithms compatible with the European Union’s supercomputers.
Instead of employing supercomputers’ processing capacity as a training resource, A.I. companies are more likely to be used to using specialized computing gear supplied by American hyperscalers. According to E.U. officials speaking in the 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 the A.I. training program is therefore being enhanced with a support wrapper.
“We are creating resources to help our small and medium-sized enterprises (SMEs) learn how to utilize supercomputers efficiently, gain access to them, and parallelize their algorithms for A.I. applications so they can create models,” a Commission representative stated. “We anticipate using a lot more of these kinds of approaches than we do now by 2024.”
“The Union now views A.I. as a strategic priority,” they continued. “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,” the statement reads. “With AI becoming a strategic priority, next to the AI Act,”
Another E.U. official said an “A.I. support center” is coming. It would include “a special track” for SMEs and startups to receive assistance in making the most of the E.U.’s supercomputing power. They stated, “We need to acknowledge that the A.I. community has not used supercomputers for the last ten years.” They require our assistance even though they are not new to using GPUs—they are inexperienced with interacting with supercomputers.
“The A.I… community frequently stems from a wealth of knowledge regarding the number of GPUs that may fit in a single box. They’ve been excellent at it as well. However, the supercomputers are just large boxes with GPUs, and to scale up and make the most of the supercomputer, other skill sets and assistance are required.
Over the past five years, the bloc has significantly increased its investment in supercomputers, building the hardware for a cluster of eight machines spread throughout the region. It also intends to connect these machines via terabit networks, forming a federated supercomputing resource available in the cloud for users throughout Europe.
The first exascale supercomputers in the E.U. are also scheduled to be online the following year; the first will likely be located in Germany (perhaps next year) and the second in France (anticipated in 2025). The Commission plans to purchase several quanta co-located with supercomputers the Commission plans to purchase several quanta co-located with the suit. This will allow the quantum computers to serve as, in other words, “accelerators” for the classical supercomputers.
A project called Destination Earth aims to simulate Earth’s ecosystems to better model climate change and weather systems. Another application being developed on top of the E.U.’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 U.S. president announced this fall’s compute access for the A.I. model training initiative, its supercomputing capabilities to jump-start A.I. firms have been a recent strategic focus.
The bloc announced a competition aimed at European A.I. startups “with experience in large-scale A.I. models” last month. Its name, “Large AI Grand Challenge,” is intended to select up to four promising homegrown startups that will receive 4 million hours of supercomputing access to support the development of foundational models. A €1 million prize fund is also set up for the winners, who, according to the Commission, must publish their research findings or share their created models under an open-source license for noncommercial usage.
Through a request for projects procedure, the E.U. already had a program to give industry users access to core hours of supercomputing resources. However, the E.U. is increasing funding and focusing on commercial A.I. through specialized initiatives and projects. It is seeing an opportunity to position its expanding supercomputing network as a strategic power source for scaling general-purpose A.I. “Made in Europe.” Therefore, it appears that Mistral, an A.I. firm based in France that hopes to take on U.S. foundation model giants like OpenAI and that offers “open assets” (if not entirely open source itself), was one of the first recipients of the Commission’s supercomputer access program. However, the fact that a tech company that just raised €385 million in Series A funding—including from American investors like Andreessen Horowitz, General Catalyst, and Salesforce—is in the lead for an E.U. computing giveaway may cause some people to wonder. However, it indicates the strategic wagers placed on “big A.”
The E.U.’s “supercomputer for A.I.” program is still in its infancy, so it’s uncertain how much model training can benefit from dedicated access. (When writing, Mistral has not responded to our request for comment.) However, the Commission hopes that by providing support to A.I. 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 A.I. model training in mind, this will give the local A.I. ecosystem that is just getting started a competitive edge over hyperscaler-proximate U.S. A.I. giants.
According to a Commission official, “We use our supercomputers and we will develop a new generation of supercomputers that will be more and more A.I. compliant since we do not have the large hyperscalers that Americans have in the case of training these kinds of foundational models.” “The goal is to move in this direction and have even more of our SMEs use the supercomputers for developing these foundational models by 2024, not just the ones that we currently have.”
According to them, part of the strategy would be to buy “more specialized A.I. supercomputing machines that will be more based on accelerators than the standard CPUs.”
It remains to be seen whether the E.U.’s. A.I. support strategy aligns with or deviates from the goal of some member states to develop national A.I. champions. During the recent contentious negotiations to establish the bloc’s A.I. 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 E.U.’s supercomputing access initiative could indicate a convergence of ideas.
