The European Union has announced winners of the “AI Big Grand Challenge,” which launched earlier this 12 months with the goal of accelerating the pace of homegrown innovation by creators of large AI models.
The 4 startups will share a €1 million money prize and, perhaps more importantly, 8 million GPU hours to train their models on several of the bloc’s high-performance computing (HPC) supercomputers over the next 12 months. The commission believes this can enable them to reduce model training times “from years to weeks,” as its PR put it.
The winning 4 startups are in alphabetical order: French fintech The language of care, which processes financial documents using natural language processing (NLP); Belgian start-up Text reinforcement, which also uses NLP for text processing, but focuses on analyzing unstructured data, akin to monitoring social media conversations for hate speech; Latvian startup Tilde, one other language specialist specializing in Balto-Slavic languages, offering machine translations and AI-based chatbots in goal languages; and Portugal Unbabelwhich has historically combined machine translation with the knowledge of native speakers and leveraged artificial intelligence for customer support and productivity applications for enterprise clients.
The Commission reported that a total of 94 proposals were received under the AI Challenge.
Unbabel probably has the highest status of the 4 winners. The Y Combinator-backed translation company has been in business for over a decade almost $100 million was raised throughout the period, according to Crunchbase.
It’s debatable whether Unbabel needs an extra quarter of a million euros, or maybe even 2 million free GPU training hours, but even seasoned AI startups may find that any help counts, given the rapid growth of generative AI over the last 1 .5 years.
At the end of the training period, the EU expects that each one winners will open-source their models for non-commercial use or publish the results of their research.
EU supercomputers to support AI startups
The EU unveiled a plan to expand startup access to the bloc’s supercomputing equipment in President Ursula von der Leyen’s State of the Union address last fall, saying at the time that it wanted “ethical and responsible AI startups” to be the first to queue to use the support computing power.
The Joint Undertaking on European High Performance Computing (also generally known as EuroHPC JU) – to give the bloc’s supercomputing initiative its full name – now has eight working (nine purchased) supercomputersand two of them will allocate 8 million GPU hours to the 4 winners: Finland’s Lumi and Italy’s Leonardo (each pre-exascale HPC supercomputers).
The fifth startup based in Spain Multiverse computingwhich focuses on trying to improve the energy efficiency and speed of large language models using “quantum-inspired tensor networks,” narrowly missed out on the prize money. But there is some consolation: it should be allocated 800,000 computing hours on one other supercomputer, the Spanish (pre-exascale) MareNostrum 5.
These handful of European startups building large-scale AI models won’t be the first to explore the capabilities of HPC hardware. French general-purpose artificial intelligence modeler Mistral was a participant in an early pilot phase of supercomputing services last summer, using Leonardo to “run some small experiments,” co-founder and CEO Arthur Mensch told TechCrunch in December — though he said it wasn’t getting used for model training.
In the past, the EuroHPC Joint Undertaking also provided some capability to business entities. However, demand for supercomputers typically far outstrips supply, so AI startups are essentially pushed to the front of the queue.
EU policymakers have also recognized the need to reconfigure and re-equip HPC infrastructure for the era of generative AI. That’s why in January the Commission announced a package of measures on “artificial intelligence innovation”, which included proposals to modernize supercomputers and build a support layer to improve accessibility so that AI start-ups can more easily use the infrastructure.