A couple of months ago, UK’s latest AI supercomputer, the Isambard-AI has went live and started to provide all its computing power towards projects in need. Today, NVIDIA will be helping the system expand further, while announcing something new along the way.
The new construction is said to be the next-generation AI factories in partnership with CoreWeave, Microsoft, and U.K.-based Nscale. By 2026, these facilities will integrate 120,000 NVIDIA Blackwell Ultra GPUs and up to £11 billion in local data center investments, marking the largest rollout in the nation’s history.
OpenAI is expected to serve its most advanced models, including GPT-5, on this infrastructure, while Nscale expands globally with 300,000 NVIDIA Grace Blackwell GPUs. Microsoft and Nscale are also set to build the U.K.’s most powerful supercomputer in Loughton with 24,000 NVIDIA Grace Blackwell Ultra GPUs, bringing advanced Azure AI services to the country.
Meanwhile, CoreWeave will further strengthen this ecosystem with a renewable-powered data center in Scotland, while BlackRock has pledged up to £500 million to modernize data centers with NVIDIA-ready hardware. Beyond infrastructure, NVIDIA is also driving collaborations in quantum computing with institutions such as Oxford, Edinburgh, and Imperial College London, as well as advancing robotics and workforce training through techUK, QA, and Quanser.
Alongside these industrial initiatives, the sovereign AI effort is also focused on linguistic inclusivity. The UK-LLM project, originally launched as BritLLM and led by University College London, has released a new AI model based on NVIDIA Nemotron that can reason in both English and Welsh. Developed with Bangor University and Nscale, this model is designed to support public services from healthcare to education through the Welsh language, spoken by about 850,000 people today.
If you’re wondering if UK-LLM is trained with Isambard-AI, then yes, it actually is. Specifically, the Welsh model is backed by £225 million in government investment and powered by hundreds of NVIDIA GH200 Grace Hopper Superchips. Using Nemotron’s open datasets and NVIDIA NIM microservices, the team translated more than 30 million English entries into Welsh, creating one of the most comprehensive AI training datasets for the language to date.
By uniting sovereign AI infrastructure with linguistic inclusivity, the U.K. is not only laying the groundwork for advanced industrial applications and quantum breakthroughs but also ensuring that cultural identity remains part of its digital future.
