AIs near-omnipresence and sustained impact across a broad range of previously disconnected economic sectors has already established AI as the foremost transformative technology of recent times. It comes as no surprise, then, that national policy and investment-drives are beginning to reflect the cultural shift towards autonomous workflows.
The AI Opportunities Action Plan of January 2025 represented an initial ‘call-to-arms’ for the British scientific community, petitioning for an intensification of AI research efforts in light of rapid advancements across the US and China. This frontier thinking translated demonstrably into the subsequently released Modern Industrial Strategy (MIS, Jun 2025), with Industry 4.0 and the Digital Revolution a consistent throughline for national growth planning. Though open-ended, these initiatives outline a renewed focus on strategic economic resilience, in direct response to the growing challenges of contemporary geopolitics and the creeping impacts of climate change. To this end, the Advanced Manufacturing, Defence, and Digital & Technologies sectors stand to share a combined £6bn of dedicated investment through the next decade.
It is the release of the AI for Science Strategy, however, that specifically establishes a framework and vision for AI-assisted scientific discovery in the UK. Published in November 2025, the strategy focuses on three key mission pillars - Compute, Data, and People & Culture - to deliver on digitalisation policy, underscoring the ambition of the government to revitalise the UK’s standing as a world-leader in technological innovation. Of the £2bn foundational investment earmarked for AI initiatives, up to £137m has been made available for targeted funding across five key ‘priority areas’, including Fusion Energy, Quantum Technologies, Engineering Biology, Medical Research and - pertinently for our community - Materials Science.
These sectors, wide-ranging as they are, have been identified as the likeliest to benefit from ‘AI-driven progress’ in the short- to medium-term. From a materials perspective, this is perhaps unsurprising: high-profile generative-AI models from the private sector (namely Microsoft’s MatterGen, and Google DeepMind’s GNoME) are already impacting the national materials discovery landscape considerably. Notably, too, the Royce Institute’s National Materials Innovation Strategy (NMIS, Jan 2025) - which underpins the £50m National Materials Innovation Programme - has specifically highlighted the role that AI will play in establishing a Materials 4.0 ecosystem. Over the next decade, it is indeed likely that AI and associated technologies will change the face of the materials sector substantially.
Compute is, of course, an intuitive barrier to the expansion of AI research initiatives. Held against the US and China - with a combined 60-70% Global Capacity Share (GCS) of GPU - the UK’s computational infrastructure is comparatively diminutive. With the national GCS sitting in the region of 3-5%, any attempt to match the raw output of international data centres would make little economic or political sense. Prudently, the government appears to have acknowledged this: it cannot win an AI arms-race built on access to resource alone.
Instead, planning appears to have settled on a narrower, more focused approach. Much like national steel policy - which pivoted from large-scale manufacture to specialised, high-performance steel production in the wake of increased international competition - new initiatives will focus on computational efficiency above scale. In short, the strategy commits to the production of a federated computational network (‘Sovereign Compute’) and the formation of ‘AI Growth Zones’ (AIGZs) to provide a rapid-access platform for future science.
The Dawn (Cambridge, 2024) and Isambard AI (Bristol, 2025) supercomputers - together forming the basis of the AI Research Resource (AIRR) - have seen £300m of start-up investment to date, and the strategy commits to a further £1bn investment in their expansion. In time, access to the AIRR network will provide anywhere from 10,000 to 1.4m GPU hours to public research projects, dependent upon scale and specification. In parallel, the first Growth Zone has been christened at the UKAEA’s Culham HQ, designed to strip back data centre planning, provide priority access to the energy grid, and encourage investment in construction. Reading between the lines, the government sees here a circular relationship between compute, growth, and economic resilience: AI built on several decades’-worth of nuclear data brings newfound innovation to materials and energy research, leading to powerful new developments in energy production; in time, these innovations sustain the energy demands of data centres, providing access to greater AI compute.
Compute without Data, however, is redundant, and increasing the useful output of AI research is the implicit goal of the policy drive. To this end, the strategy also seeks to frame data - its generation, collection, and storage - as a significant strategic asset, proposing a unified cultural shift in the handling of scientific datasets. In much the same fashion as prior compute considerations, commitment is made to the development of large-scale data infrastructure, with upscaling projects at existing repositories (such as Harwell’s Diamond Light Source) and fabricating new storage in Bristol and Edinburgh in proximity to sovereign compute.
Away from hardware, the strategy also reiterates the need for improved data stewardship. FAIR principles (Findability, Accessibility, Interoperability, and Reusability) are to ideally be applied to all scientific data by 2030, though such a sea-change in the monitoring of data collection and processing inside four years is perhaps ambitious. Further emphasis is also placed on the retention of ‘dark data’ - unpublished or inaccessible data produced during the research process. Recent estimates suggest that over 50% of all scientific data held by UK-based organisations falls under this banner, with negative experimental results or even standard experimental procedure deemed ‘dark’ by virtue of their inaccessibility. Though this has long been considered problematic in the scientific community - with the journal-centric model for scientific reporting often incentivising the over-promotion of ‘successful’ science - dark data has long-term ramifications for the AI community. Not only is access to larger datasets preferable by default, but the inclusion of negative results during training is necessary to prevent bias in model outputs.
It is here that the Henry Royce Institute is capable of playing a substantial role in the future of digital materials research. The strength of Royce is in its position as a central ‘hub’ for materials science, leading an interdisciplinary landscape in both academia and industry. By extension, Royce can further benefit the community by acting as a unifying leader for data generation and best practices. The institute is specifically highlighted in the AI for Science Strategy for its Data Curation programme, with the team successfully delivering workshops on Data in the Materials Community in 2025, and a number of data-centric events already planned into the 2026 pipeline. Equally, reference is made to the provision of AI-ready data provided by Royce-led initiatives such as the Digital Materials Foundry, which hopes to provide a centralised resource for materials-based Large Language Models (LLMs), Machine Learning models, and materials data repositories. Further value could be added through the hosting of community-driven AI benchmarking - mentioned by the strategy as integral to engagement and innovation - with locally hosted data leveraged for this purpose.
It requires little imagination to foresee further in-house practical contributions to these repositories: time allocated for data collection on state-of-the-art Royce laboratory equipment could substantially bolster the AI-ready data landscape, and provide significant value to the wider materials community. In this sense, it is not difficult to see the transformational impact that Royce could have on future AI-enabled materials science, and we would actively encourage comments or suggestions from the community as to how we might best facilitate frontier research going forward.
The final pillar for consideration is the oft-overlooked requirement for an adequate research Culture, supported by a properly trained ecosystem of professionals. The document outlines a sweeping (at times perhaps vague) set of action points for the rapid provision of technical AI experts, with the intention of producing a work-force capable of delivering on desired innovation. In some ways, these actions are aligned with existing Royce initiatives: the desire to provide upwards of 1000 AI doctoral training and fellowship positions reflects the success of the Royce Materials 4.0 Centre for Doctoral Training, and the push for interdisciplinarity is baked into the structure of Royce. There is little time, however, dedicated to explaining the mechanisms by which these groups will become equipped with the requisite diversity in their backgrounds or expertise. Missing, too, is evidence for the consideration of cybersecurity and the interface of these procedures with the trained workforce. Though this is no doubt at the top of the agenda for future white paper releases, the success of culture-building is likely to hinge on the risk to reward ratio for national security.
Taken collectively, how does this translate to newly coordinated, AI-driven Science? This is, perhaps, unanswerable in the immediate term. A follow-up to the AI Opportunities action plan was published only last week, purporting delivery on 38 out of 50 original aims. Practical delivery for the strategy began with the formation of the Sovereign AI Unit (SAU), a task-force subsidiary of the Department for Science (DSIT) designed to provide £500m to the UK’s AI landscape. The SAU has already delivered open calls for AIRR access and funding for autonomous labs, inspired in part by the £400m sales growth successes of Royce-affiliated autonomous Materials Innovation Factory. This funding ecosystem is also supported by other, more targeted DSIT initiatives, such as the Advanced Research + Invention Agency (ARIA)’s recent AI Scientist call, which raised the Materials Science sector as an exemplar. Further still, Imperial College London - a significant Royce partner in the Modelling & AI Research Area - last month announced the opening of a Centre for AI-Driven Innovation in partnership with the World Economic Forum (WEF), while the National Physical Laboratory hailed a new Centre for AI Measurement. Both schemes are designed to leverage institutional expertise and accelerate AI adoption across the country, though naturally their value will be judged by their long-term value-add to the scientific economy.
‘Methodological’ considerations (i.e. the alignment of scientific decision making with national policy) are purportedly to be handled concurrently by the UK Metascience Unit (UKRI & DSIT), though this initiative is too fresh for appraisal. This, too, without mention of the ethical or social ramifications of widespread AI adoption, particularly across an increasingly opaque private sector. It does, however, represent a strong initial statement from DSIT on its commitments to expand the UK’s offering in the AI sphere, and to capitalise upon potential scientific investment opportunities in the near future.
To conclude, the AI for Science Strategy has provided a strong foundation for the deployment of an AI-ready national research landscape, and the Materials Science community stands to gain considerably from its outcomes and future plans. The Henry Royce Institute is ready to support these initiatives from the ground up, and continues to be an integral part of national scientific digitalisation policy.
For further discussion on the Royce approach to the Materials 4.0 Challenge, or the Modelling & AI Research Area, please get in touch via royce@sheffield.ac.uk, or via our LinkedIn page to continue the conversation.