Capability, control and growth achieving sovereignty through open-source ai
Explore this report
Executive Summary
Artificial intelligence is central to the UK’s ambitions for economic growth, higher productivity and better outcomes for its citizens. The challenge is no longer whether the UK adopts AI, but whether it does so in a way that builds capability, maintains control, and delivers sustained growth. That, ultimately, is a question of sovereignty.
Open-source AI (OSAI) is central to this shift. Unlike closed models, it gives organisations the best visibility and control over how AI works, how data is used, and how systems are tailored over time — allowing skills, capability and decision-making to be built onshore. A strong OSAI ecosystem enables the UK to better turn its research strengths, talent base and startup ecosystem into genuine AI sovereignty.
The UK Government recognises this and put the question of sovereignty front and centre in its AI Opportunities Plan. This is not about ‘owning’ AI systems, but about ensuring the UK has the practical ability to use AI deeply and safely — an area where OSAI has unique advantages.
Yet despite these advantages, OSAI adoption in the UK remains limited. Other countries — including France and India — have made OSAI a core part of their national strategies. The UK has not and risks falling behind its peers without providing the right incentives and support for OSAI.
Against this backdrop, Meta commissioned Public First to examine how OSAI is being used in the UK economy, what is holding it back, and what action the Government could take to unlock its potential.
Our research has found that if we can achieve faster and wider adoption of OSAI the UK has the potential to deliver £40 billion in additional economic growth over the next decade to 2035. This benefit will be particularly felt by small businesses and business sectors critical to the Government’s industrial strategy, such as manufacturing and scientific research.
However, this boost to growth will not happen on its own, it needs to be seized. Currently, firms struggle to find industry-specific OSAI tools; over three-quarters of businesses say they would adopt industry-specific OSAI tools if they were available, but uptake is constrained by gaps in supply, and a failure of demand to be translated into adoption. Regulatory barriers around data access and use, make this harder still. Targeted, policy action to address these barriers could accelerate adoption, unlock productivity gains, and help ensure that the UK captures the economic benefits of AI alongside enhancing sovereignty. This report sets out a practical agenda to do that.
The potential additional economic growth that can be driven by open-source AI adoption over the next decade.
Over three quarters of businesses say they would adopt an industry specific open-source AI tool, if available.
A third of businesses cite business data security and privacy concerns as a reason not to make greater use of AI.
The amount of economic growth that could be delivered in the near term through an open-source AI adoption fund.
Summary of recommendations:
action
Costs versus benefit
Outcome
Key Government partners
Create a British Open-Source AI Investment Credit to increase the number of companies creating OSAI in the UK
Cost £69m per year; expected ROI of over 20:1 through higher private R&D and positive spillover effects into the wider innovation ecosystem
More UK-based development and release of open-source AI tools; creating stronger domestic capability boosting the innovation ecosystem by encouraging more open sourcing of AI technology
HM Treasury and DSIT
Build UK sovereign AI training datasets
Low cost by using existing investment allocated to the National Data Library; potential £1.5bn uplift by 2035 across priority sectors driven by greater open-source AI capabilities
Easier fine-tuning and deployment of open-source AI; reduced duplication and stronger sector-specific tools
DSIT / Cabinet Office
Clarify how data protection rules apply to open-source AI (including training and fine-tuning)
Low direct cost through regulatory change; a 10% boost in confidence to use data for open-source AI and training could unlock £250m per year by 2035
Provides greater confidence to developers and businesses to train and fine tune open-source AI tools on their own data as well as other data available to them
DSIT / ICO
Create an Open-Source AI Sovereignty Fund to drive adoption and uptake of OSAI among SMEs
£100m investment; potential £1.1bn+ economic uplift from faster AI adoption
More SMEs in growth sectors trial, deploy and scale open-source AI; resulting in a faster diffusion of productivity gains
DSIT / DBT
Preferential access to compute for UK open-source AI developers
Low direct public cost, utilising existing AI infrastructure such as AIRR; potential £50m+ spillover value via data centre demand and faster innovation
Reduced development bottlenecks for SMEs and developers; faster iteration and scaling of open-source AI tools
DSIT / UKRI
Create a National Open-Source AI Institute
Moderate ongoing funding, similar costs to the National Quantum Technologies Programme (NQTP); estimated benefit of up to £50m economic impact from added research capacity, with likely wider benefits
Sustained open-source AI research, skills development, and then translation into deployable tools through new research / spinouts
DSIT / UKRI
Embed open-source AI in Business Growth Service guidance
Low cost; but high coverage through existing business support website and services
Clear, practical advice to grow the understanding of OSAI, tackling myths about increased risk or liability from using OSAI. Provide firms with guidance to find, choose and implement OSAI products with confidence
DBT / DSIT
Action
Create a British Open-Source AI Investment Credit to increase the number of companies creating OSAI in the UK
Costs versus benefit
Cost £69m per year; expected ROI of over 20:1 through higher private R&D and positive spillover effects into the wider innovation ecosystem
Outcome
More UK-based development and release of open-source AI tools; creating stronger domestic capability boosting the innovation ecosystem by encouraging more open sourcing of AI technology
Key Government partners
HM Treasury and DSIT
Action
Build UK sovereign AI training datasets.
Costs versus benefit
Low cost by using existing investment allocated to the National Data Library; potential £1.5bn uplift by 2035 across priority sectors driven by greater open-source AI capabilities
Outcome
Easier fine-tuning and deployment of open-source AI; reduced duplication and stronger sector-specific tools
Key Government partners
DSIT / Cabinet Office
Action
Clarify how data protection rules apply to open-source AI (including training and fine-tuning)
Costs versus benefit
Low direct cost through regulatory change; a 10% boost in confidence to use data for open-source AI and training could unlock £250m per year by 2035
Outcome
Provides greater confidence to developers and businesses to train and fine tune open-source AI tools on their own data as well as other data available to them
Key Government partners
DSIT / ICO
Action
Create an Open-Source AI Sovereignty Fund to drive adoption and uptake of OSAI among SMEs
Costs versus benefit
£100m investment; potential £1.1bn+ economic uplift from faster AI adoption
Outcome
More SMEs in growth sectors trial, deploy and scale open-source AI; resulting in a faster diffusion of productivity gains
Key Government partners
DSIT / DBT
Action
Preferential access to compute for UK open-source AI developers
Costs versus benefit
Low direct public cost, utilising existing AI infrastructure such as AIRR; potential £50m+ spillover value via data centre demand and faster innovation
Outcome
Reduced development bottlenecks for SMEs and developers; faster iteration and scaling of open-source AI tools
Key Government partners
DSIT / UKRI
Action
Create a National Open-Source AI Institute
Costs versus benefit
Moderate ongoing funding, similar costs to the National Quantum Technologies Programme (NQTP); estimated benefit of up to £50m economic impact from added research capacity, with likely wider benefits
Outcome
Sustained open-source AI research, skills development, and then translation into deployable tools through new research / spinouts
Key Government partners
DSIT / UKRI
Action
Embed open-source AI in Business Growth Service guidance
Costs versus benefit
Low cost; but high coverage through existing business support website and services
Outcome
Clear, practical advice to grow the understanding of OSAI, tackling myths about increased risk or liability from using OSAI. Provide firms with guidance to find, choose and implement OSAI products with confidence
Key Government partners
DBT / DSIT
Foreword
The United Kingdom has long stood at the forefront of technological innovation — from Turing’s foundational work in computing to its modern, thriving AI ecosystem.
The UK government has set out an ambitious vision in the AI Opportunities Action Plan, placing sovereignty, onshore capability and resilience at the heart of its national AI strategy. The AI minister has set out a vision to make Britain the home of global open-source AI talent.
At Meta, we share this conviction in the potential for AI to unlock innovation and power growth. We have invested significantly in AI research and development — from open-source models like Llama that have been downloaded hundreds of millions of times to collaborative initiatives that aim to broaden participation in AI innovation globally.
Our experience shows that open-source approaches enable wider participation in AI development. By making core components accessible and adaptable, they allow organisations to experiment, customise tools to their needs and build capability in-house. That process embeds tools and skills more deeply within the domestic economy — strengthening resilience and long-term sovereignty.
Around the world, governments, academic communities, and businesses are embracing open-source AI as a vehicle for inclusion and economic opportunity. Initiatives in Europe, India and beyond are harnessing the ingenuity of open ecosystems to build tailored, locally relevant solutions that drive productivity and help small and medium-sized enterprises compete more effectively in global markets. These efforts demonstrate that when nations lower barriers to experimentation and collaboration, they unlock not just innovation but resilience — ensuring their economies have the capabilities to be adaptable in the face of rapid technological change.
Against that backdrop, we commissioned this report by Public First to better understand how the UK can seize the opportunity that open-source AI presents.
The findings are clear. Open-source AI creates new opportunities for smaller businesses and high-growth sectors in Britain to innovate, compete and scale. By lowering barriers to entry and enabling tailored deployment, it helps British firms to embed AI more deeply into their operations — building confidence in data use and accelerating the next phase of adoption across the economy.
Importantly, this report does not simply diagnose barriers; it offers practical, costed policy recommendations designed to maximise the benefits that open-source AI can bring to UK businesses as a catalyst for growth and prosperity. We hope this will be an important contribution to the UK’s success in the AI era.
Chris Yiu,
Director of Public Policy Northern Europe, Meta
About this research
Meta commissioned Public First to explore the current landscape of open-source AI adoption in the UK, and the benefits increased adoption could bring to the economy. Our independent findings emphasise the opportunity open-source AI presents, and detail proposals to ensure the UK makes the most of this opportunity.
Research Methods
We used a mixture of methods to conduct our research:
- Public First ran an online poll of 541 senior decision-makers in UK businesses between 27th November and 8th December 2025. Participants qualified if they were at director level or above and had decision-making responsibility for at least one key business area. Quotas and weighting were used to ensure a broadly representative balance of business size and sector based on the proportion of employed UK adults working for each business type.
- Public First spoke with several businesses who have adopted or developed AI tools.
- Public First then created new quantitative models exploring the economic impact of open-source AI and demonstrating the impact of the various policy proposals we have made. A full methodology for these is available at the end of this report.
About Public First
Public First is an independent consultancy that works to help companies and organisations develop new policy proposals, better understand public opinion, and model their economic and social impact. Public First is a member of the Market Research Society and the British Polling Council and abides by their rules and guidelines.
the case for open-source ai
The UK has a strong track record of turning open technologies into economic growth, from publicly funded internet protocols to early investment in digital infrastructure that helped British businesses thrive. Open-source AI presents a similar moment of opportunity.
what is open-source ai?
Open-source AI refers to AI systems developed and shared in a way that allows others to access, use, modify and build on them, rather than being locked into a platform. There are various definitions of open-source AI, but in practice, it means key elements such as model weights, code and documentation are made available under open licences.
This openness gives organisations far greater control and transparency than closed-source alternatives. Instead of relying on externally managed tools with limited visibility over how models work or how data is handled, organisations can deploy open-source AI within their own environments, tailor it to their needs, and integrate it into core workflows. This helps build capability onshore by building up the skills required to fine-tune and apply an AI model, retaining control over data, and creating a greater range of AI suppliers to businesses.
As a result, open-source AI is essential to the UK’s ambitions on sovereignty, resilience and long-term economic growth. Where concerns over data security, vendor lock-in and one-size-fits-all tools are limiting deeper AI adoption, open-source approaches directly address these barriers by enabling more tailored, trusted and controlled deployment across business, public services and research institutions.
Despite our strengths, the UK is falling behind on Open-Source AI
The UK enters the current phase of global AI development from a position of real strength, with a leading research base, deep technical talent, and one of Europe’s most vibrant AI ecosystems spanning science, industry and services. It has also played a prominent global role in AI governance, hosting the world’s first AI Safety Summit and establishing institutions to support responsible development.
These advantages have translated into strong private investment, with the UK now home to Europe’s largest AI startup ecosystem and regularly ranked third globally for AI investment, behind only the United States and China.
Private sector AI investment (2013-2024)
However, despite these strengths, the UK is falling behind on open-source AI, particularly when compared with countries such as France, India and the USA, which are backing open-source approaches through clearer strategy, dedicated funding and national infrastructure.
UK
Dedicated public funding for open-source AI
No dedicated funding stream. Support sits within wider AI and R&D programmes, with no ring-fenced investment for open-source models or tools.
National open-source AI champion
No government-backed open-source AI champion or anchor institution; activity remains largely market-led and fragmented.
Sovereign compute and infrastructure for open-source
Limited access to sovereign compute, with no explicit prioritisation for open-source model training or fine-tuning.
Open-source-first national AI strategy
Open-source AI is referenced but not a strategic priority; focus remains on adoption and safety.
Outcomes
Strong research base but lower contributor growth with few large-scale open-source champions and slower translation into commercial scale.
France
Dedicated public funding for open-source AI
Targeted public funding for open-source AI, including state backing for Mistral AI to crowd in private investment.
National open-source AI champion
Clear national champion in Mistral AI, supported as a strategic asset for domestic and European capability.
Sovereign compute and infrastructure for open-source
Partial sovereign compute provision, aligned with national and EU initiatives supporting open-source developers.
Open-source-first national AI strategy
Open-source AI embedded in national AI and industrial strategy, linked to sovereignty and competitiveness.
Outcomes
Rapid rise of Mistral to multi-billion valuation; strong venture capital inflows; Paris emerging as a leading European hub for AI research labs and startups.
India
Dedicated public funding for open-source AI
Large, centrally coordinated funding treating open-source AI as national digital infrastructure.
National open-source AI champion
No single champion, but a deliberate ecosystem-led approach supporting startups, academia and developers at scale.
Sovereign compute and infrastructure for open-source
Sovereign compute is a core pillar of India’s approach, with dedicated infrastructure for open-source training and deployment.
Open-source-first national AI strategy
Open-source AI treated as foundational to national AI development, skills and digital sovereignty.
Outcomes
Open-source contributor base growing 32% year-on-year; second-largest GitHub community globally; $3.4bn raised by AI startups in 2024.
USA
Dedicated public funding for open-source AI
Mixed federal funding through DARPA, NSF and DoE programmes; significant defence and research investment supporting open-source model development and infrastructure.
National open-source AI champion
Multiple globally dominant open-source AI firms (e.g. Hugging Face) with strong venture backing and close links to leading universities and federal research labs.
Sovereign compute and infrastructure for open-source
World-leading hyperscale compute capacity and federally funded supercomputing centres supporting model training, research and open-source collaboration.
Open-source-first national AI strategy
No formal “open-source-first” mandate, but open models central to commercial ecosystem and national AI competitiveness strategy.
Outcomes
Global leader in open-source AI development; highest contributor base; concentration of frontier labs, venture capital and ecosystem spillovers driving commercial scale.
UK
France
India
USA
Dedicated public funding for open-source AI
No dedicated funding stream. Support sits within wider AI and R&D programmes, with no ring-fenced investment for open-source models or tools.
Targeted public funding for open-source AI, including state backing for Mistral AI to crowd in private investment.
Large, centrally coordinated funding treating open-source AI as national digital infrastructure.
Mixed federal funding through DARPA, NSF and DoE programmes; significant defence and research investment supporting open-source model development and infrastructure.
National open-source AI champion
No government-backed open-source AI champion or anchor institution; activity remains largely market-led and fragmented.
Clear national champion in Mistral AI, supported as a strategic asset for domestic and European capability.
No single champion, but a deliberate ecosystem-led approach supporting startups, academia and developers at scale.
Multiple globally dominant open-source AI firms (e.g. Hugging Face) with strong venture backing and close links to leading universities and federal research labs.
Sovereign compute and infrastructure for open-source
Limited access to sovereign compute, with no explicit prioritisation for open-source model training or fine-tuning.
Partial sovereign compute provision, aligned with national and EU initiatives supporting open-source developers.
Sovereign compute is a core pillar of India’s approach, with dedicated infrastructure for open-source training and deployment.
World-leading hyperscale compute capacity and federally funded supercomputing centres supporting model training, research and open-source collaboration.
Open-source-first national AI strategy
Open-source AI is referenced but not a strategic priority; focus remains on adoption and safety.
Open-source AI embedded in national AI and industrial strategy, linked to sovereignty and competitiveness.
Open-source AI treated as foundational to national AI development, skills and digital sovereignty.
No formal “open-source-first” mandate, but open models central to commercial ecosystem and national AI competitiveness strategy.
Outcomes
Strong research base but lower contributor growth with few large-scale open-source champions and slower translation into commercial scale.
Rapid rise of Mistral to multi-billion valuation; strong venture capital inflows; Paris emerging as a leading European hub for AI research labs and startups.
Open-source contributor base growing 32% year-on-year; second-largest GitHub community globally; $3.4bn raised by AI startups in 2024
Global leader in open-source AI development; highest contributor base; concentration of frontier labs, venture capital and ecosystem spillovers driving commercial scale.
What is the Current State of AI Adoption Among British Businesses?
The majority of UK businesses are now using AI in some form, indicating that the technology has moved beyond early experimentation and into mainstream awareness. Across the economy, firms are engaging with AI tools to support a range of activities, from basic automation to more advanced analytical tasks.
However, the depth of adoption varies significantly by company size. Large businesses are far more likely to have embedded AI directly into core functions or integrated it into internal systems that support company-wide automation. By contrast, smaller firms are more likely to use AI occasionally or for simple, standalone tasks rather than as part of core workflows
Using the scale below, please tell us which comes closest to how your business uses generative AI
Businesses already see clear and tangible benefits from using generative AI tools. When asked what the benefits of using generative AI were, the most common answers among UK businesses were:
said increased efficiency and productivity.
said creating content (e.g. text, images, or reports) more quickly.
said speeding up product or service development.
Looking ahead, businesses are broadly positive about the role AI will play in their organisations and expect their use of AI to deepen over the next two to three years. But this shift is far from guaranteed. Only 27% of UK businesses expect AI to be embedded in the core functions of their organisation over that period, while one in five (22%) do not expect to use AI at all.
These numbers are even more concerning when we look at SMEs, where just 18% expect AI to be embedded as a core function, and 29% do not expect to use it at all. This creates a risk that adoption could remain shallow or uneven across the economy.
A larger, healthier OSAI ecosystem could help prevent this by providing a community of businesses and developers that are constantly experimenting and providing a wider range of customisable tools that they can tailor to their needs.
As businesses consider embedding AI more deeply into their operations, concerns around data privacy become a more binding constraint.
When AI is used for limited or low-risk tasks, questions about data handling can often be more easily managed. But as systems begin to interact with sensitive commercial or customer data, firms become more cautious about where that data goes, how it is used, and whether they retain effective control over it. In many cases, reliance on closed or externally managed AI systems means giving up visibility over data flows and decision-making. This helps explain why 30% of businesses cite concerns over data security or privacy as a reason they are not making greater use of generative AI.
Open-source approaches can mitigate these concerns by allowing organisations to deploy and govern AI systems within their own environments, retaining greater control over data use while still accessing advanced capabilities.
Without greater clarity and confidence around how data can be used safely and compliantly, many firms risk delaying or limiting the next phase of AI adoption.
what are the main things that put your business off from making greater use of generative AI, or investing in its use in future?
“The benefit of open-source AI is that you don’t need to build something from scratch, you already have a baseline and then you’re building on top of that, which means you can focus on what really matters rather than rebuilding the basics. For us, having the ability to use an open source AI and then change it the way we want works very well from an innovation point of view.”
Representative of a large telecoms company
Furthermore, many businesses are relying on generic AI tools rather than solutions tailored to their specific industry or operational needs. This pattern is particularly pronounced among smaller firms, which are significantly more likely than large businesses to use off-the-shelf tools rather than industry-specific applications. While generic tools can deliver quick and accessible benefits, they are less well suited to supporting complex workflows.
As a result, smaller businesses risk capturing fewer of the long-term productivity gains that more tailored AI solutions can deliver, reinforcing existing gaps in capability and performance between firms of different sizes.
Does your business use any generative AI tools that are specifically designed for your type of business?
Despite the UK’s strong position in AI, evidence from global open-source activity suggests it is not keeping pace with leading competitors. On GitHub, the US remains the dominant hub for open-source generative AI development, with around 80,000 contributors, while India has rapidly scaled to around 30,000 contributors and is growing faster than any other major developer ecosystem. The UK has around 13,000 contributors to open-source generative AI projects. This is a strong ecosystem, but places the UK behind not only the US and China, but also India, Hong Kong, Germany and Japan. Users are also growing at a much slower rate in the UK (6%) than in other countries like India (32%).
This gap matters because participation in open-source ecosystems is a key channel through which skills are developed, tools are shared, and standards are shaped. Without stronger engagement, the UK risks becoming more dependent on AI capabilities developed elsewhere, rather than helping to set the direction of the technologies its businesses increasingly rely on.
The OSAI opportunity for the UK’s Industrial Strategy sectors
Open-source AI offers a practical response to many of the barriers currently limiting deeper and more widespread adoption of AI across the UK economy. By changing how tools are developed, shared, and adapted, open-source approaches can help businesses move beyond reliance on generic solutions and towards more tailored, integrated use.
In particular, open-source AI can help because it:
Offers greater control and transparency over data use, giving organisations more confidence about how models are trained, fine-tuned, and deployed, and making it easier to align AI systems with data protection and governance requirements.
Encourages faster diffusion of innovation, allowing improvements to spread across firms, sectors, and regions more quickly than proprietary approaches alone.
Enables customisation without starting from scratch, allowing businesses to adapt existing tools to their sector or workflow rather than relying on one-size-fits-all products.
Lowers barriers to adoption for smaller firms, reducing the need for large upfront investment or highly specialised in-house teams.
Supports collaboration and reuse, helping businesses learn from and build on tools developed by others rather than duplicating effort.
“When we are working with sensitive data, we cannot allow it to leave organisational boundaries. In those cases, we prefer locally run models and agents. Everything runs on our own servers, in our own lab environment. None of that touches the cloud. If the data stays inside, we maintain control. That’s the benefit of open-source AI”
Representative of a major car manufacturer
In our survey, businesses recognised these benefits:
Thinking about such an AI tool you have full control over, what do you see as the main potential benefits for your business, compared with using only off-the-shelf commercial tools?
Case study: How PwC is using Open-Source AI to Transform Document Processing
Generative AI is increasingly being adopted in professional services to improve efficiency and accuracy. PwC has integrated open-source AI into its document workflows through Digital Workmate, an automation solution built on Llama. The system extracts and validates data from invoices, purchase orders and contracts, reducing manual review. By fine-tuning an open-source model to recognise industry-specific terminology and formats, PwC has achieved over 90% accuracy in field extraction and enabled more than 70% of documents to be processed straight through without manual intervention.
Open-source AI has also provided greater flexibility in implementation. Because the model can be deployed on-premises or in private cloud environments, PwC is able to meet strict data security and regulatory requirements. The ability to customise the model without rebuilding it from scratch has reduced processing times by up to 70%, cut processing costs by 60–70%, and saved approximately 800 hours per month. This demonstrates how open-source AI can deliver measurable productivity gains while maintaining control over data and deployment.
While these benefits are most obvious at the level of individual businesses, their importance extends far beyond firm-level efficiency. When more companies are able to adopt and tailor AI tools to their needs, the cumulative effect is wider productivity growth, stronger competitiveness, and faster diffusion of innovation across the economy.
Open-source approaches help spread the benefits of AI beyond a small group of large or highly digital firms, supporting broader adoption across sectors and regions. In doing so, they offer a practical way to translate business-level improvements into economy-wide growth and resilience.
In total, we estimate open-source AI could create £40 billion for the UK economy by 2035.
This impact would be felt across Britain’s key industries:
The Impact of open-source ai by 2035 on the following sectors:
The economic impact of open-source AI is concentrated in sectors that sit at the core of the Government’s industrial strategy. Manufacturing, professional and business services and the financial sector account for a large share of the estimated uplift, reflecting both their scale and their exposure to data-intensive processes. In these sectors, open-source AI has the potential to support more efficient production, better asset utilisation, and faster innovation by enabling firms to develop and deploy tools that are tailored to complex operational environments rather than relying on generic, off-the-shelf systems.
At the same time, the importance of open-source AI is not limited to sectors with the largest absolute gains. In parts of the economy where productivity growth has historically been slow, even relatively modest uplifts can be significant.
In water supply, sewerage and waste, for example, the estimated £400 million impact would amount to almost 1.2% of the sector’s current value. This illustrates how open-source approaches can deliver meaningful improvements across a wide range of industries.
The relative impact of open-source AI by 2035, as a % of total GVA on the following sectors:
“The UK operates in an AI landscape largely shaped by competition between the United States and China. They are unlikely to build frontier models at scale, but that does not exclude them from influence. The opportunity sits downstream. Much of AI’s long-term economic and strategic value will be captured through adaptation, fine-tuning and sector-specific deployment. Open-source gives middle powers a way to shape that layer. By investing in developer communities, last-mile tooling and domain expertise, they can build comparative advantage in areas where they already have strengths and retain agency in how AI diffuses through their economies.
Sovereignty in this context is about ecosystem capability. It rests on the ability to adapt and govern models, curate strategic data, and integrate AI into national priorities without overdependence on external platforms. Open-source supports that aim by widening access to the tools and infrastructure that sit across the AI stack. The strategic question is not who owns the largest model. It is who controls the capacity to apply, improve and embed AI where it matters.”
Keegan McBride
Director – Science and Technology Policy, Tony Blair Institute for Global Change
Case study: How TLC LIVE is using Open-Source AI to Expand Access to Tutoring
Generative AI is increasingly being adopted in education to widen access to personalised learning and reduce cost barriers. TLC LIVE has developed an AI tutor, ‘Manda’, built on an 8B parameter Llama 3 model that has been fine-tuned using 550,000 minutes of transcribed explanations from over 300 fully qualified UK teachers. The model has been trained to support Key Stage 3 and 4 maths and English using national curriculum-approved methods. By drawing on structured teaching explanations rather than generic internet data, the system is designed to reflect established classroom practice while remaining accessible to students globally.
The open-source foundation model enables TLC LIVE to customise and continuously improve the system through its internal academic team. ‘Manda’ is priced at £10 per month per student, significantly lowering the cost of structured tutoring, and builds on TLC LIVE’s experience delivering almost 900,000 sessions to more than 51,000 students. All personal data was removed from training transcripts, and student interactions are stored in encrypted databases with safeguards in place. This demonstrates how open-source AI can be adapted to local curriculum standards, expand access to high-quality support and complement, rather than replace, qualified teachers.
The Barriers to OSAI adoption
and why intervention is needed
Open-source AI presents a huge opportunity, both for individual businesses and for the wider UK economy. Yet there is a clear gap between this potential and what is happening in practice. While many businesses express strong interest in adopting open-source AI tools, actual take-up remains limited, particularly among smaller firms.
of UK businesses would be more likely to adopt industry specific open-source AI tools if there were more products readily available, and clearer guidance on how to adopt and run them.
Only a third of UK businesses have adopted an open-source AI tool.
of UK businesses have taken an open-source tool, developed by another organisation, and adopted it in-house.
of UK businesses have taken an open-source tool, developed by another organisation, and built upon it in-house.
This participation gap is most pronounced among small and medium-sized enterprises. Just 15% of SMEs have taken an open-source tool, developed by another organisation, and built upon it in-house compared to 28% of large businesses. This reflects the practical realities smaller businesses face; limited in-house technical capacity and tighter resource constraints.
Businesses that could benefit from open-source AI are not adopting it at scale, and firms that are capable of developing useful tools are not releasing or sustaining them in ways that support wider uptake. The result is an under-provision of open-source AI tools and a persistent adoption gap, despite strong underlying interest.
This matters for the UK economy because it slows the diffusion of productivity-enhancing technologies and concentrates the benefits of AI among a small number of firms, limiting the choice and spillovers across sectors and regions. Without wider use and adaptation, investment in AI risks delivering fragmented gains rather than the broad-based productivity growth the UK needs.
“Not all adoption creates value equally. There is a clear difference between widespread use of proprietary chatbots to streamline routine tasks and the deeper integration of open-source AI into core systems. When researchers fine-tune open models on their own datasets, or when firms build specialised tools around open infrastructure, they create a durable advantage. Open-source matters here because it allows adaptation. It gives organisations visibility into how systems work, flexibility to modify them, and the freedom to integrate them into existing workflows without relying entirely on a closed vendor stack. That form of adoption builds capability over time rather than simple dependency.
Sector context shapes the outcome. A consumer using a chatbot through a search interface engages with AI in a limited way. A manufacturer embedding open models into industrial robotics operates at a different level entirely, with greater scope for customisation and control. Policymakers should therefore be precise about their aims. The objective is not adoption for its own sake, but adoption that strengthens domestic capability, supports sector-specific innovation and captures more of the value generated by AI.”
Guy Ward-Jackson
Senior Policy Advisor, Tony Blair Institute for Global Change
Taken together, this pattern points to a clear market failure with supply, demand and information challenges – a classic problem that governments can address with targeted interventions.
Supply-side barriers
A significant proportion of businesses report that they are unable to find AI tools tailored to their industry or use case. This is not because firms are uninterested in more specialised solutions, but because relevant products are either not available, not visible, or not supported in a way that makes adoption viable. Others have not searched at all, suggesting that the difficulty in finding appropriate tools, uncertainty about quality, and limited confidence in what is available are suppressing engagement. A majority of UK businesses either can’t find, or haven’t looked for industry-specific AI tools.
of UK businesses say they have looked for a tailored or industry-specific AI tool but have been unable to find a suitable option.
of UK businesses have not looked for a tailored or industry-specific AI tool at all.
Where the market does not provide suitable tools, in-house development might be expected to fill the gap. Yet here too, supply is constrained:
Only 16% of businesses have developed their own in-house industry-specific AI tool
This highlights a fundamental bottleneck: most firms lack the specialist skills, time, and resources required to design, build, and maintain bespoke AI systems. As a result, businesses are often left with tools that are either too generic to deliver real value, or entirely absent.
of businesses who say they are unlikely to develop an AI tool in-house cite a lack of in-house knowledge or skills as a key barrier.
of businesses who say they are unlikely to develop an AI tool in-house cite a lack of other resources as a key barrier.
Open-source AI has the potential to break this deadlock — but only if businesses can access both ready-to-use tools and the support needed to adopt them. Most organisations do not build or implement new technologies on their own; they rely on specialist suppliers, integrators and service providers to help select, adapt and run them.
Today, that support layer is largely missing in the open-source AI ecosystem. If it were strengthened — through a larger pool of reusable tools and a stronger market of UK-based specialists able to deploy them — more businesses could adopt open-source AI without starting from scratch. This would lower the skills barrier, support SMEs by giving them a greater range of choices, and help create a new generation of companies focused on implementing, supporting and scaling open-source AI across the economy.
Among businesses who would consider open-sourcing tools they’ve built:
would do so to improve their reputation or stand out as innovative.
to attract potential collaborators for the future.
to influence industry standards or best practice.
“We are building on top of open-source stacks, frameworks and libraries that have been created by many people before us. We stand on the shoulders of giants. If we benefit from that ecosystem and do not contribute back by open-sourcing our own tools, the innovation momentum does not build. Philosophically, contributing back is part of the process.”
Representative of a major car manufacturer
However, too many businesses are reticent to open-source such tools in the first place.
Only 1 in 5 UK businesses would definitely consider open-sourcing an AI tool they developed in house. 28% probably / definitely would not consider doing so.
These businesses often cite a mix of perceived risks, uncertainty, and limited capacity. Together, this contributes to a self-reinforcing gap in supply, where useful tools are scarce, development is fragmented, and innovation struggles to scale across the economy.
Demand-Side Barriers
Demand-side barriers are not driven by a lack of interest in open-source AI. Most businesses say they are in theory open to adopting open-source tools and many expect to do so in the near future. This indicates that appetite exists across the economy, but that practical obstacles continue to prevent firms from translating interest into action.
of businesses say they are likely to consider taking an open-source tool, developed by another organisation, and adopt it in-house in the next 2-3 years.
of businesses say they are likely to consider taking an open-source tool, developed by another organisation, and build upon it in-house in the next 2-3 years.
Yet we know that only 1 in 3 of businesses have adopted open-source AI, meaning there are clear barriers preventing businesses adopting these tools.
61% of businesses that have adopted open-source tools in-house say they required some initial adjustments. These adjustments often involve fine-tuning models, integrating tools with existing systems, and adapting them to specific business processes, all of which demand technical expertise.
This requirement for additional fine-tuning creates a practical demand-side barrier. Firms need access to skilled, qualified staff who can configure and maintain open-source tools, or the resources to bring in external support. Larger businesses are more likely to have this capability in-house, while small and medium-sized enterprises are more exposed to the cost, time, and risk associated with implementation. If UK businesses lack the capability or confidence to deploy AI systems themselves, they are more likely to default to a small number of externally developed, proprietary technologies to support core activities. Over time, this can reduce choice, raise costs, and limit the ability of firms to adapt systems to their own needs. As AI becomes more central to how businesses operate, ensuring that UK firms have viable alternatives is essential for the Government’s drive for AI sovereignty.
“You can always hire developers or contractors to help with integration, but the real challenge begins when the system has to operate reliably at scale. A proof of concept may work perfectly in a lab environment with a small number of users. But scaling to millions of users is a different challenge. Some issues only appear at scale. When you rely on an external API provider, they carry much of that scaling and infrastructure burden. With open-source AI, that burden sits with you.”
Representative of a major car manufacturer
“Because we operate national infrastructure, we need very strong guardrails. Whether a model is open or closed, we have to be confident it will not impact the resilience, availability, or reliability of the network, and that places limits on how quickly and widely we can deploy these systems.”
Representative of a large telecoms company
Case study: How the Llama Impact Grants are Supporting Open-Source AI Innovation
Open-source AI is increasingly being used to drive economic and social impact across sectors and regions. The second round of Llama Impact Grants recognised 10 international recipients and awarded over $1.5 million to support companies, startups and universities building on Llama. The programme is designed to accelerate the development of open-source AI solutions that address real-world challenges, ranging from agricultural insights and rural connectivity to fraud detection and healthcare innovation. Because Llama is openly available, recipients are able to build and scale tools without the licensing costs or access barriers associated with proprietary systems.
The funded projects demonstrate the breadth of open-source AI applications. In the United States, Solo Tech is using Llama to provide offline, multilingual AI support to underserved rural communities, with plans to equip 50 rural centres with AI tools. In the UK, Doses AI is developing an autonomous pharmacy system that automates prescription processing and stock management while maintaining pharmacist oversight. Since opening applications, the initiative has also supported the wider ecosystem through nearly 60 global events, including accelerator programmes and hackathons. Together, these efforts illustrate how open-source AI can lower barriers to innovation, enable locally tailored solutions and support economic opportunity at scale.
Regulatory Environment and Risk
Data privacy concerns play a central role in how businesses approach open-source AI. For firms developing their own tools, decisions about whether to release them openly are shaped less by technical performance and more by uncertainty around data use, compliance, and risk. As AI becomes more embedded in sensitive datasets and core business processes, confidence in how data is handled becomes increasingly important.
While many recognise the benefits of open-sourcing their tools to their external reputation, concerns about data protection, misuse, and unintended exposure often outweigh these incentives. Uncertainty about how existing rules apply in practice leads some firms to take a cautious approach, helping to explain why supply remains constrained despite clear interest. In our survey, concerns about security and misuse were the number one barrier to firms open-sourcing tools they’ve built.
You said that you would be unlikely to consider open-sourcing any tool you develop in-house. Is this for any of the following reasons?
Taken together, these concerns have wider consequences for the open-source AI ecosystem. When businesses hesitate to release tools because of data privacy uncertainty, the supply of reusable, industry-relevant open-source AI remains limited, even where strong technical capability exists. This in turn slows adoption by other firms and reinforces existing gaps between those with the resources to build and manage AI internally and those without.
Addressing data privacy confidence is a necessary condition for scaling open-source AI and ensuring its benefits can diffuse more widely across the UK economy.
policy solutions
The economic potential of open-source AI will not be realised by default. While interest among businesses is growing, persistent market failures around supply of products, translating demand into adoption, and regulatory confidence continue to limit progress. The policy challenge is to create the conditions in which open-source AI can develop and scale where it offers clear advantages.
The measures below focus on lowering barriers to development and adoption, improving access to skills, data, and compute, and providing greater clarity on data use. They are designed to accelerate diffusion and ensure the benefits of open-source AI are spread more widely across the UK economy.
Taken together, they can help the UK seize the £40bn of additional growth available from OSAI over the next decade.
Supply-Side Barriers
The UK is not producing enough open-source AI tools that businesses can easily adopt and build upon. High development costs, limited access to key inputs, and weak incentives to release and maintain tools openly continue to constrain supply, particularly among SMEs and early-stage firms.
Supply-side interventions should therefore focus on lowering the cost and risk of developing open-source AI in the UK, and strengthening the conditions for sustained domestic innovation. The measures below are designed to increase the availability of high-quality tools and anchor more development activity within the UK.
Establish a British open-source AI Investment Credit
The Problem
Developing open-source AI involves significant upfront costs in skilled staff, compute, data, and infrastructure, while the economic benefits are often shared widely rather than captured by the firm that invests. This weakens incentives for businesses—particularly SMEs and start-ups—to develop and release open-source AI tools in the UK, leading to systematic underinvestment despite clear spillover benefits.
The Solution
A targeted investment credit would help correct this market failure by lowering the effective cost and risk of open-source AI development. By improving the risk-reward profile for firms that build and release tools openly, open-source AI activity can be anchored within the UK, which would help achieve sustained investment.
Under this proposal, businesses undertaking qualifying open-source AI R&D in the UK would be eligible for a 10% corporation tax deduction, or up to a 10% cash credit if not paying corporation tax. Eligible expenditure would include staff, compute, data, and equipment directly related to AI research, training, and product development. To qualify, firms would commit to releasing defined elements of their AI product under recognised open-source licences. This would include a UK nexus requirement to ensure benefits are captured in the UK.
The government should amend the existing R&D tax credit and add this 10% incentive as an additional benefit (on top of existing R&D tax credit deductions) making it easy to apply for and linked to a scheme many businesses already use.
Our modelling shows this kind of targeted incentive is effective because lowering the upfront cost of open-source AI development leads to a disproportionate increase in investment, generating spillover benefits as reusable tools are adopted and adapted across many firms rather than captured by a single provider.
Public First estimates that this investment credit would increase UK open-source AI R&D by 25.5%, raising total annual investment from £694 million to £871 million.
A 10% tax credit directed at open-source AI R&D projects would directly increase GVA by £71m per year. Our modelling shows the indirect benefits are likely to be an order of magnitude higher than this. We estimate this measure would cost £69 million a year, with an expected ROI of at least 20:1.
Provide Preferential Compute Access for UK open-source AI
The Problem
Access to compute is an increasingly binding constraint on AI development, particularly for businesses working on open-source tools where costs are harder to recoup. Training, fine-tuning, and testing models require significant computational resources, and limited or expensive access can slow development, discourage experimentation, and prevent promising tools from scaling, especially among SMEs and application developers.
The Solution
Providing preferential access to compute would directly address this constraint by lowering the cost and friction associated with high-performance computing. This would enable more frequent experimentation, faster iteration, and more ambitious open-source AI development by UK-based firms, while helping anchor activity domestically rather than overseas.
UK businesses developing open-source AI models or applications should be offered preferential access to public compute resources, such as reserved capacity within national infrastructure like AIRR or priority access to publicly supported compute environments. Eligibility should be limited to firms carrying out training or fine-tuning activity in the UK, and extend to both model developers and companies building open-source applications that require periodic access to significant compute.
Our modelling shows this has a strong economic case because compute is a binding constraint on open-source AI development, and easing it unlocks activity that would not otherwise happen, generating spillover benefits as models and tools are reused across firms rather than delivering returns to a single provider. Unlike general compute subsidies, preferential access for open-source AI targets a clear market failure by supporting activities with high spillovers that are currently being underprovided.
The spillover value from meeting the additional data centre demand created by open-source AI development could generate around £50 million in additional economic activity, reflecting increased investment in UK-based data centre capacity, construction and operations, as well as wider knock-on effects through supply chains.
UK Sovereign AI Training Sets
The Problem
High-quality, sector-specific data is essential for training and fine-tuning AI models, but it is currently fragmented, costly and hard to access — especially for smaller firms.
Our modelling shows that this is a particular barrier for open-source AI, where value comes from adapting models to UK-specific workflows, regulations and operating conditions rather than relying on generic, centrally trained systems. As a result, many potentially high-value open-source applications never get built.
The Solution
The Government should invest in UK sovereign AI training datasets aligned to priority industrial strategy sectors, using existing and emerging infrastructure such as the National Data Library to deliver them.
Making high-quality, governed datasets available for open-source development would lower development costs, reduce duplication, and allow many firms to build and fine-tune AI tools on shared foundations. Because open-source tools can be reused and improved across the economy, this approach generates much larger spillovers than firm-specific data access.
Our analysis shows that removing this data constraint has an outsized impact on adoption and productivity, because access to shared, UK-relevant datasets significantly reduces development costs and allows open-source tools to be reused and improved across many firms rather than built repeatedly in isolation.
Enabling open-source AI to meet just 10% of data-intensive AI needs in three key sectors, manufacturing, finance and professional services, could unlock up to £1.5bn in additional economic value by 2035.
National open-source AI Institute
The Problem
The UK has strong early-stage AI research, but much of this value is fragmented, absorbed into proprietary systems, or fails to translate into reusable tools that businesses can adopt. Our modelling shows that without a dedicated mechanism to sustain open-source research and retain talent, knowledge spillovers are weaker and productivity gains take longer to materialise. This limits the long-term impact of the UK’s existing research strength.
The Solution
A National Open-Source AI Institute would provide a focal point for long-term investment in shared research capacity, with a clear mandate to develop and maintain open-source AI tools that can be reused across the economy. By attracting and retaining top AI researchers and anchoring their work in open, reusable outputs, the Institute would accelerate the accumulation of domestic AI capability and increase spillovers into businesses and public services.
Using established evidence on the link between research capacity and total factor productivity, we estimate an institute with 100 OSAI researchers could add around £50 million in economic value, with wider benefits as open-source tools are adopted, adapted and built upon across multiple sectors.
Demand-Side Barriers
Even where open-source AI tools are available, many businesses struggle to move from interest to adoption. Gaps in skills, information, and implementation support, alongside concerns about risk and disruption, continue to slow uptake, particularly among smaller firms.
Demand-side interventions should focus on reducing uncertainty, building capability, and supporting businesses to trial and deploy open-source AI in real-world settings. The measures below aim to help firms turn potential into practice and support wider diffusion across the economy.
Create an open-source AI Sovereignty Fund
The Problem
Many businesses are interested in adopting open-source AI but struggle to move from awareness to implementation. Skills gaps, uncertainty around integration, and the perceived risk of disruption often stall adoption at an early stage, particularly for SMEs with limited in-house capability.
The Solution
A dedicated Open-Source AI Sovereignty Fund would help address these barriers by supporting firms to trial, deploy, and scale open-source AI tools in real operational settings. By reducing the cost and risk of early adoption, the fund would help businesses turn existing open-source capabilities into tangible productivity gains.
Our modelling draws on evaluation evidence from the Made Smarter pilot and official R&D multipliers, adjusted for the lower costs and higher spillovers of open-source AI, showing that hands-on adoption support combined with targeted challenge funding delivers early productivity gains that compound over time, generating returns of more than £11 for every £1 invested.
Investing £100 million in a Made Smarter style government programme to catalyse open-source AI adoption could support economic growth in the UK of over £1.1 billion.
Updates to Business Growth Service guidance to include OSAI
The Problem
Many businesses are interested in adopting AI but lack clear, practical information on when and how open-source approaches are appropriate. This is particularly true for SMEs, which often struggle to navigate a complex and fast-moving ecosystem and may default to familiar proprietary tools or delay adoption altogether if they cannot get the right information or guidance.
The Solution
Updating Business Growth Service guidance to include open-source AI would address this information gap at low cost. Clear, non-technical guidance would help businesses understand the benefits and trade-offs of open-source tools, identify suitable options, and make informed decisions about adoption, governance, and implementation.
This guidance should also seek to tackle common myths about OSAI, for example that it is riskier, or that its use entails greater liability over outputs and the use of OSAI products when compared to closed-source models.
The Government should expand existing Business Growth Service guidance to cover open-source AI as part of its wider advice on digital adoption, productivity, and growth. This should focus on practical considerations rather than technical detail, and be included within mainstream business guidance rather than treated as a specialist add-on.
Improving access to clear, practical guidance reduces adoption risk and accelerates uptake. Our survey shows that 76% of businesses would be more likely to adopt open-source AI if suitable tools and guidance were available, meaning better information directly supports faster diffusion, higher productivity, and stronger economic returns.
Regulatory Environment
Open-Source AI Data Access and Use
The Problem
The UK’s data protection regime is creating unnecessary friction for OSAI development and adoption. Under the current guidance, where a firm downloads weights and fine-tunes or adapts a model on its own infrastructure, it is more likely to be treated as a distinct controller for that processing, with corresponding obligations to document compliance, manage transparency and enable rights. This is very onerous to evidence and is particularly challenging for smaller firms.
Meanwhile, the EU is moving to clarify responsibilities and controls across the AI supply chain, while the UK missed an opportunity in the Data (Use and Access) Act 2025 to set out clearer, more practical rules.
The Solution
The UK should make data protection rules and regulatory guidance clearer, more proportionate, and easier to apply in practice for OSAI — especially for SMEs. The focus should be on outcomes and effective safeguards, not process-heavy regulatory compliance. This should include clearer specification of where model outputs may present personal data risks, and guidance around practical steps to mitigate those risks, providing users of OSAI with an assurance that they have taken the necessary steps to comply with data protection rules.
As part of a review of the Data (Use and Access) Act 2025, the UK Government should amend and clarify the law on lawful data use for training and fine-tuning, including proportionate flexibilities for low-risk testing and experimentation by SMEs as well as exemptions for using data to address bias and discrimination.
In addition, the UK legislation around text and data mining (TDM) creates significant uncertainties for firms who want to use publicly available data or data they hold with unclear copyright protections for fine-tuning. The EU and other jurisdictions like Japan and Singapore have already resolved this issue by creating exemptions for text and data mining for commercial purposes.
Our modelling isolates open-source AI use cases where UK-specific data and workflows are essential (using our task-based analysis) and shows that clearer, simpler rules on lawful data use would remove a binding barrier to training and fine-tuning. Even a 10% increase in business confidence to use data translates into a material uplift in adoption and productivity, worth around £250m a year by 2035.
methodology
Below we set out the analytical approach underpinning our assessment of the economic opportunity from open-source AI and the impact of the policy interventions proposed in this report. It combines original business polling with established economic modelling to quantify how deeper, more controlled and more widespread AI adoption — enabled by open-source approaches — could translate into productivity gains, onshore capability and long-term growth for the UK economy.
Business survey
Public First conducted an online survey of 541 senior decision-makers in UK businesses, covering a range of company sizes and sectors. Fieldwork was carried out between 27th November and 8th December 2025, with respondents drawn from established business research panels.
Participants were screened to ensure they held senior roles with responsibility for technology adoption, operations, or strategic decision-making within their organisation. The sample included both SMEs and large businesses, enabling analysis by firm size where relevant. Quotas were applied by sector and company size to ensure broad coverage of the UK business population.
The survey explored businesses’ current use of AI, including generative and open-source tools, depth of adoption, perceived benefits, barriers to wider and deeper use, and views on data use, skills, and regulatory risk. Results reported are weighted by business size, sector, and region using Iterative Proportional Fitting, or ‘Raking’.
Economic Modelling
Overall economic benefits of open-source AI
- Our main model exploring the economic impact of AI is a task-based model, following the precedent of Eloundou et al (2023), Microsoft / Public First (2024), and Felten, Raj and Seamans (2021).
- We draw on the US O*NET database, which sets out the task composition of different types of occupation. For each combination of task and occupation, we use an LLM to classify:
- How likely it is that task can be augmented by today’s AI and machine learning technology
- How sensitive a task is likely to be for legal, cultural or ethical reasons, limiting the potential for automatability
- To what extent fine-tuning or customization through open-source AI would improve performance on the task, based on
- Domain-specific knowledge or terminology
- Organization-specific workflows or processes
- Specialized output formats or styles are required
- How important UK specific workflows or data sources are for training on the task
- We then aggregate up to the occupation level, looking at the proportion of tasks within each occupation that are potentially augmentable in each scenario.
- We convert from US to UK occupation data, using a crosswalk at the 4-digit level, and then aggregate up from there to 2-digit occupation and individual sectors, based upon lower level occupations’ share of total wagebill from ONS data. We use an S-curve diffusion model to project baseline future adoption, which then is adapted depending on the relevant scenario.
- Using our sector model, we look at the impact new government-supported datasets could create to support 10% of use cases in three key sectors: manufacturing, finance and professional services.
- Our focus in this modelling is looking at use cases where open-source AI is likely to be more technically suited to a workflow. We do not take account of other potential gains that could come from open-source AI being cheaper to run than closed-source AI, accelerating adoption and uptake, and reducing inference costs. That suggests that our estimates should be regarded as a lower estimate of the potential benefits from open-source AI.
Impact of attracting additional AI researchers through a new National Open-Source AI Institute
We estimate the impact of attracting additional AI researchers to the UK by:
- Converting into effective research FTEs, and assigning a plausible sector split between business, higher education and government, benchmarked against the UK’s existing R&D workforce.
- Using a perpetual inventory approach to estimate the accumulation of knowledge, and then drawing on the literature on the elasticity between overall knowledge stock and TFP level.
Spillover value from additional data centre investment
In order to calculate this, we:
- Use our AI and labour force models to convert our estimates of the overall economic impact of open-source AI into an equivalent number of additional inference queries by 2035. We then use conservative estimates of average token-per-query and inference efficiency to convert this into an overall estimate of required data centre capacity.
- We draw on industry datasets to estimate average construction and operating costs by MW of nameplate capacity, and then apply indirect and induced multipliers calculated from the ONS’ input-output tables.
Economic benefits from a Made Smarter style scheme
In this claim, we look to estimate the benefits of a £100 million UK government programme modelled on Made Smarter, focused on supporting open-source AI adoption across economic sectors. The programme includes:
- Adoption support: Business advice, digital mentoring, leadership training, and matched funding grants for SMEs to integrate open-source AI
- Challenge funding: Competitive grants for developing new open-source AI products and applications in industrial strategy sectors
In order to calculate this, we:
- Use evaluation evidence from the Made Smarter North West pilot to estimate GVA returns per pound of government investment in technology adoption programmes. For the challenge funding component, we apply DSIT’s official £8 per £1 multiplier for public R&D investment.
- Adjust both multipliers for open-source AI characteristics—lower licensing costs and stronger spillover effects from freely reusable tools—applying a net 25% uplift to adoption support and a 50% uplift to challenge funding. On top of this, we make allowances for additionality (reflecting that some firms would adopt AI without support) and displacement (some productivity gains come at competitors’ expense), respectively.
- Benefits are modelled over a 10-year realisation period, with adoption gains front-loaded (years 1-5) and innovation benefits back-loaded (years 4-10), then discounted to present value at the Green Book rate of 3.5%.
Economic benefits from improving access to data for open-source AI workflows
In order to isolate this benefit we:
- Draw on our categorisation run of how important UK specific data is, to isolate likely open-source AI workflows where international data is unlikely to be sufficient.
- Apply an adjustment to our associated estimate of potential economic impact, drawing on other Public First modelling on the importance of data availability to AI capability.
Direct and indirect benefits from creating a tax credit for open-source AI
In order to calculate the direct benefits of a tax credit for open-source AI, we:
- Use HMRC evaluation evidence to estimate the elasticity of R&D expenditure to R&D tax credits. HMRC’s 2020 evaluation of the R&D Expenditure Credit found elasticities of 2.4-2.7, meaning each £1 of tax relief generates £2.40-£2.70 of additional private R&D spending.
- Estimate the share of UK business R&D attributable to open-source AI by combining ONS data on R&D expenditure by SIC division with DSIT’s AI Sector Study. We identify SIC divisions containing dedicated AI firms, calculate the AI sector’s share of GVA within those divisions, and apply this proportion to divisional R&D expenditure.
- Model the impact of a 10 percentage point enhancement to R&D tax relief for qualifying open-source AI investments, applying the HMRC elasticity to estimate the resulting uplift in private R&D expenditure.
- Convert additional R&D spending into GVA impact using DSIT’s published estimates of returns to public R&D, which suggest approximately 20% of R&D investment translates directly to productivity gains.
In order to estimate the indirect ROI from an open-source tax credit we:
- Calculate the cost of developing UK-specific fine-tuned open-source models, drawing on industry benchmarks for data preparation, compute, and specialist expertise.
- Estimate firm-level adoption costs using business population data and industry benchmarks for AI integration. Because foundation models already exist, per-firm costs are primarily integration and workflow adaptation—significantly lower than if each firm fine-tuned independently.
- Calculate annual productivity benefits by applying our adoption curve to the estimated productivity value at full adoption from our main model, netting off data centre operating costs. Benefits are discounted to present value.
While there are larger uncertainties to this estimate, we find that even under our most conservative assumptions the overall return is greater than 20:1. The largest driver of uncertainties in our modelling is how wide a set of use cases each open-source model is likely to serve – with less conservative assumptions here, our overall estimate of ROI could easily be many times higher.