NVIDIA’s recent work with the UK-LLM project is revolutionizing our perception of AI and languages. By teaching state-of-the-art models to understand English and Welsh. Researchers aren’t just building smarter systems—they’re giving an endangered language a better chance to thrive. Welsh has been dying for over a century but with the assistance of powerful tools like NVIDIA’s Nemotron framework. The UK’s Isambard-AI supercomputer, it may have just found a fresh digital lease on life. For the 850,000 Welsh speakers today, this isn’t just technology—it’s a step toward cultural preservation.
Supercomputers Meet Endangered Languages
UK-LLM, which launched in 2023 as BritLLM, quickly became the quintessential case study in AI’s impact on society. With help from University College London, and fed by NVIDIA’s Nemotron models. It customizes massive data pipelines to process Welsh text. It’s trained on the Isambard-AI supercomputer in Bristol, which was launched in early 2025 as a part of the UK’s AIRR initiative. Each part of this infrastructure is designed to handle enormous workloads, using thousands of GPUs at once for tasks that no single machine could attempt.
Technology of this stripe typically supports research in climate modelling or pharmaceutical simulations, but here it’s being commandeered to protect cultural legacy. It was once a living language for close to 50% of Wales, but that plummeted during the 20th century. Now AI is putting digital gossip Welsh-style at scale. This isn’t just about building translation tools—it’s about creating a foundation where future generations can learn, write, and even think with confidence in a language that once risked falling silent.
A Global Blueprint for Language Survival
But as much as Welsh is where it’s at, the project slots into a much broader battle against cultural dissolution. A 2025 study projects less than 5% of the world’s 7,000 languages might survive in usable digital form if present patterns continue. Such a loss would be cataclysmic, not only for language but for stories, for culture, for individual ways of seeing the world. English, Mandarin, and Spanish dominate digital spaces, leaving minority languages struggling for visibility.
Imbuing Welsh directly into cutting edge AI models, though, UK-LLM researchers demonstrate retention is achievable when technology receives a cultural imprimatur. They’re also looking soon to take their method to Irish, Scottish Gaelic and Cornish as well. And if it does, the framework may hold a blueprint for other countries confronted with disappearing tongues. The UK’s decision to couple this with AIRR, a resource aiming to reach mind-bending levels of 420 AI exaFLOP by 2030, opens the door to scaling these ideas globally.
The subtext is powerful–AI doesn’t have to pander exclusively to the screeching minorties. It can magnify small neglected ones. When combined with public supercomputing resources, there’s no technical barrier left. Just the decision to prioritize endangered languages before they disappear entirely.
A Fresh Direction for AI and Culture
UK-LLM project is a watershed. But rather than entrenching linguistic injustice, this project uses bleeding edge computing and open source infrastructure to even the odds. AI is often criticized for reinforcing global hierarchies. But this time it’s being applied in a way that keeps communities alive.
The importance goes beyond Wales. I love this project because it proves that although we live in an age of linguistic imperialism. The tiny ones can still wriggle out a cyberspace dwelling if countries invest the proper amount of resources. Nemotron’s ability to switch between English and Welsh reasoning proves that machines can adapt, and communities can thrive alongside them.
Welsh speakers now have more than cultural cachet–they’ve got bleeding-edge support. And if the U.K.’s push catches on, it might begin a trend — one where AI is humanity’s unexpected ally in saving thousands of languages for a digital.
About The Author
