A European Union plan to support domestic AI startups by giving them access to processing power to train models on the bloc’s supercomputers, announced in September and launched last monthsaw France’s Mistral AI take part in an early pilot phase, according to a briefing from the EU. However, an early lesson is that the program should offer dedicated support to AI startups to train them how to get the most out of them EU high-performance computers.
“One of the things we’ve seen is the need, not just to provide access, but to provide ease — especially skills, knowledge and experience we have in the hosting centers — on how this access can not only be facilitated but also develop training algorithms that use the best architecture and computing power currently available in each supercomputing center and on our machines,” said an EU official speaking during a press conference today.
The plan is to create “centres of excellence” to support the development of proprietary AI algorithms that can run on the EU’s supercomputers, they added.
AI startups are more likely to get used to using dedicated computing hardware provided by US superscalers to train their models rather than leveraging the processing power that supercomputers offer as a training resource. So, high-performance computing access for the AI training program is being boosted with a wrap of support, according to EU officials who spoke on the sidelines before the official ribbon-cutting for Mare Nostrum 5a pre-hexascale supercomputer to be inaugurated on Thursday at the Barcelona Supercomputing Center in Spain.
“We are developing facilities for our SMEs so that they can understand how to better use supercomputers and how to access supercomputers and how to parallelize their algorithms in the case of artificial intelligence so that they can develop their models,” an official said. Commission. . “From 2024, we expect many more of these approaches than we have now.”
“Artificial intelligence is now considered a strategic priority for the Union,” they added. “With AI becoming a strategic priority, alongside the AI Act, we provide the innovation capability — or we want to provide a large innovation window for our SMEs and startups to better use our machines and this public infrastructure we have”. have created so that they can compete internationally in the development of safe, reliable and ethical artificial intelligence algorithms.”
An “AI support center” is on the way, another EU official confirmed — saying it will have “a dedicated section” for SMEs and start-ups to get help making the most of the EU’s supercomputing resource. “What we have to recognize is that the AI community has not been using supercomputers for the last decade,” they noted. “They’re not new to GPUs, but they’re new to how to interact with a supercomputer, and so we need to help them.
“In many cases the AI community comes from a huge knowledge of how many GPUs can you put on a box? And they were very good at it. But what we have in supercomputers are multiple boxes of GPUs, and there are some additional skill sets and some additional help that are needed to scale and use the supercomputer to its full potential.”
The bloc has significantly stepped up its investment in supercomputing over the past five+ years — ramping up hardware to a cluster of eight machines located in the region, which it also plans to interconnect, via terabit networks, to create a federated supercomputing resource that will is accessible in the cloud so that it is available for users across Europe.
The EU’s first exascale supercomputers are also due to go live in the next two years — one in Germany (likely next year) and a second machine in France (expected in 2025). The Commission also has investments in quantum computing, with plans to acquire a clutch of quantum simulators to be co-located with supercomputers to provide a hybrid resource combining both types of hardware so that quantum computers can act as ‘accelerators’ for classic supercomputers, as the Commission says.
Applications being developed on top of the EU’s high-performance computing hardware include a project to simulate Earth’s ecosystems to better model climate change and weather systems, called Destination Earthand another to invent a digital twin of the human body — which will hopefully further aid medical science by supporting drug development and even enabling personalized medicine. Using its supercomputing resources to spark AI start-ups has emerged as the latest strategic priority following the EU president’s announcement of computing access for the AI model training program this autumn.
Last month, the bloc also announced what it called the “AI Grand Challenge”: A competition aimed at European AI startups “with experience in large-scale AI models” that aims to select up to four promising domestic startups to receive a total of 4 million hours of supercomputer access to support fundamental model development. A €1 million prize is also earmarked for distribution to the winners — who are expected to release their developed models under an open-source license for non-commercial use or through the publication of their research findings, according to the Commission.
The EU already had a program to provide industry users with access to core hours of supercomputing resources through a call for projects process. However, the bloc is focusing on commercial AI with dedicated programs and resources – spying an opportunity to turn its growing network of supercomputers into a strategic powerhouse for scaling up general-purpose AI ‘Made in Europe’. Therefore, it seems no coincidence that France’s Mistral – an artificial intelligence startup that aims to compete with American foundation model giants such as OpenAI, and which claims to offer “open assets” (if not itself fully open source ) – is an early beneficiary of the Commission’s supercomputer access programme. (Although it may be troubling that a tech company that just raised €385m in Series A funding, including US investors such as Andreessen Horowitz, General Catalyst and Salesforce, is at the front of the queue for a free EU account. But, well, it’s another sign of high-level strategic bets being made on “big AI”.)
It’s still early days for the EU’s “supercomputer for artificial intelligence” program, so it’s unclear if there’s much model training yet for the exclusive access report. (We reached out to Mistral for comment, but at the time of writing it had not responded.) But the Commission’s hope, at least, is that by providing support to AI startups so they can leverage its investment in high-end computing performance, combined with the creation of supercomputing hardware that he says will increasingly be procured and configured with AI model training in mind, this will translate into a competitive advantage for a local AI ecosystem starting from behind compared to the AI giants of the US approaching the hyperscalers.
“Since we don’t have the large superscalers that the Americans have in the case of training these kinds of basic models, we are using our supercomputers and will develop a new generation of supercomputers that will be increasingly compatible with artificial intelligence,” a Commission official noted. “Not just the ones we have now, but, from 2024, the aim will be to go in that direction — and have even more of our SMEs using supercomputers to develop these fundamental models.”
The game plan will include acquiring “more dedicated supercomputing AI machines, relying more on accelerators than standard processors,” they added.
Whether the EU’s AI support strategy aligns with or diverges from the ambition of some member states to foster national AI champions — something we heard a lot about during the recent tough talks to define the bloc’s AI handbook, in which the France led to a push regulatory engraving for fundamental models that drew criticism from the media – remains to be seen. But Mistral’s early presence in the EU’s supercomputer access program may indicate an alignment in thinking.
