Table of Contents
Summary
The explicit goal of those driving AI progress is to build highly autonomous systems that outperform humans at most economically valuable work. Hundreds of billions of dollars are being marshalled towards this goal today.
Venture Capital is doubling down: OpenAI has just received a $150 billion valuation, while Anthropic is targeting $40 billion.
Governments are mobilising: The U.S. CHIPS Act allocates $280 billion to boost AI and semiconductor research and the EU's European Chips Act is investing €43 billion.
Infrastructure megaprojects are underway: Microsoft is planning a $100 billion investment into “Stargate”– a massive supercomputing cluster. BlackRock has announced a fund of $100 billion to build infrastructure to support the expansion of AI.
Tech giants are making massive energy plays: Microsoft is reopening America's 837-megawatt Three Mile Island nuclear energy plant to power its AI ambitions. OpenAI wants the U.S. government to help streamline the construction of data centers with power requirements of up to five gigawatts each.
A geopolitical race is emerging to ensure a capability at the frontier of this technology and to ensure access to its benefits. States with the capacity to build the most advanced general-purpose AI systems will obtain massive economic, geopolitical, and military advantages. Such advantages will be key to securing an international liberal democratic order.
Other countries are playing to win. Key Republican figures seem to be supporting a "Manhattan Project" approach to AI development. In her DNC speech, Kamala Harris stated: “I will make sure that we lead the world into the future [...] on artificial intelligence. That America [...] wins the competition for the 21st century”. Since 2017, China has made AI central to its national strategy and is racing to be the “primary AI innovation center” by 2050, driven by vast state-sponsored funding. For example, Chinese Government Guidance funds have reportedly raised $1.86 trillion for investments in AI.
One very close comparator to the UK is France, where cumulative AI government spending since 2018 now totals €7.2 billion according to Tortoise analysis – 60% more than the UK. France’s biggest supercomputer is also three times more powerful than the UK’s equivalent. This has led to France jumping to fifth place in the Global Artificial Intelligence Index, up from 13th last year. France is now only just behind the UK in fourth place, and may soon overtake us. The French National AI Commission recently called for doubling down and spending €27bn over the next 5 years.
The UK risks being left behind.
This massive investment by other countries and companies is not in pursuit of some abstract goal. It is to build systems that can fully automate human labour. There's significant uncertainty around how this technology will develop, but we should take this aim extremely seriously. What would the world look like if large foreign companies replaced the value of our cognitive labour – if we head towards a world with intelligence too cheap to meter?
Previous governments have promised economic transformation but failed to read changes in the global economy, seeing the UK's growth fall relative to our peers who better took advantage of new technologies. In the 20th century, we failed to sufficiently embrace technologies that could automate physical labour, ultimately contributing to the decline of British manufacturing. We should not risk repeating that mistake: if we fail to prepare for the automation of cognitive labour, we risk a white-collar deindustrialisation across the knowledge-intensive services that our economy relies on.
Geoffrey Hinton and Demis Hassabis just won the Nobel Prizes for physics and chemistry. They represent the best of British AI talent. Yet both failed to receive support in the UK, and had to go abroad or seek acquisition by foreign companies in order to achieve their ambitions.
The Prime Minister has said “we can make London the AI capital of the world – drawing on “our talent, expertise and hunger for success [...] the world’s best graduates, respect for the rule of law, and [...] political stability”. But this ambition is not yet reflected in government strategy. If growth is our primary mission for the next decade, proactively confronting AI’s impact on the economy and capitalising on its opportunities must be a top priority. A focus on the sectors we have already lost will prevent us from attending to those we need to win tomorrow, and a retreat from AI infrastructure investments at the first hint of fiscal challenge would be repeating the mistakes of the past.
Seriously preparing for the future will require a level of mobilisation and organisation of resources not seen since the post-war Labour government. This mobilisation is not only critical to achieving economic growth (which underpins all of the government’s ambitions) but also essential for assuring national security in the midst of a geopolitical competition for these capabilities.
The UK must learn from the actions of other nations and capitalise on its strengths to make AI central to its industrial strategy. In doing so, we should have two explicit goals:
The UK should catalyse the creation of at least one AI-first national champion within this decade. These are UK founded and headquartered firms that focus on applying AI to other domains and/or exist within the AI supply chain with a market valuation above £100 billion.
The UK must strengthen its position in the AI supply chain, including in talent, data, and computing hardware. This will strengthen national resilience, drive greater value added, and provide the UK with greater influence in the global economic landscape.
To achieve these goals, the government needs to focus on four key levers: 1) infrastructure, 2) government funding, 3) incentives for adoption and competition, and 4) talent. This piece proposes particularly high leverage policy ideas in each of these levers. However, this document should not be seen as an exhaustive proposal for an AI industrial strategy.
The Challenge & Opportunity
AI Will Reshape our Economic and Geopolitical Landscape
Those developing AI have a clear goal: create autonomous systems that outperform humans in economically valuable tasks. Hundreds of billions of dollars are being directed toward this effort. Microsoft capital spending has soared to $19 billion in 2024. This surge is driving a massive expansion of data infrastructure, resulting in a sharp rise in energy consumption. In Ireland, for instance, data centres now account for 21% of total electricity use. AI companies are directly investing in energy infrastructure to power their models' immense computational demands (Figures 1 & 2).
Figures 1 & 2: US Tech Companies are Spending Billions to Develop Energy Sources for Data Centres
Progress in AI across domains continues to be rapid and unprecedented. At the start of the previous Parliament, the most advanced AI could barely string together coherent sentences. Today, widely accessible AI can write poetry, do maths, build websites, and narrate environments for the blind. AI systems already very nearly match or exceed human performance in tasks like reading comprehension and competition-level mathematics (Figure 3). If the progress of the last five years continues at a constant pace, then we should prepare for the possibility of "highly autonomous systems that outperform humans at most economically valuable work" to emerge under this Labour Government.
Figure 3: AI Systems are Already Outperforming Human Capabilities on a Number of Tasks
Source: Our World in Data
AI will be the defining technology of our time. It will be critical to achieving economic growth, which underpins the government’s ambitions of national renewal. AI will also be essential for national security in a time of heightened geopolitical competition. Rapid progress in its development could create a more uncertain, potentially volatile world: AI could become a major source of conflict and inequality. Failing to build capacity in the UK presents a serious security risk.
AI will transform economic activity and drive innovation in a host of other technologies. Since AI is a general-purpose technology, it will have an impact throughout the economy, enabling the emergence of new firms at the forefront of adoption and causing the closure of incumbents that don’t implement AI. This dramatic shift presents both risks and opportunities.
Our economy is service-oriented (services contribute 82% of our GDP and 55% of our exports) and therefore particularly vulnerable to increased competition if others adopt AI and adapt faster than we do. Europe looks to China as the great threat to its traditional industrial base, but, for Britain, it is the US tech companies that will destroy today’s jobs. The IMF estimates that around 60% of jobs in advanced economies are exposed to AI – and the true number will likely be even higher as the technology develops and models improve in their capabilities. We are already seeing the effects of AI on jobs. For example, following ChatGPT’s release, copywriter wages fell by 9%, and filmmaker Tyler Perry halted an $800mn investment in new film studios because of the emergence of Sora. Klarna, a European fintech company, has frozen hiring, saying its new chatbot replaces the work of 700 full-time staff.
Despite its late start, the UK is well positioned to capitalise some of the benefits of this technological revolution – if we act quickly. We should place focused bets on areas we are well placed to compete in like inference and edge compute, AI-first R&D-intensive companies, or the AI assurance market. Gaining prime position in such niches could allow us to capture massive wealth, raising living standards for working people. Critically, these tools could be used to deliver on Labour’s missions and transform public service delivery in sectors like health and defence, helping to unlock billions in value for the exchequer. Many of these goods and services could bolster national resilience — for example, tools for cybersecurity and defence, extreme weather prediction, and self-driving labs for drug discovery. These tools can help states weather a crisis, or increase their international leverage. But it will take concerted government action to make sure we capture this value.
The UK is Already Falling Behind
The UK's leadership position in frontier AI capabilities is fragile. Our status is entirely due to one American-owned company, Google DeepMind, without which we have no frontier AI capability. Without DeepMind, the UK's share of citations among top 100 recent AI papers drops from 7.84% to 1.86%. Leadership in development is also disconnected from deployment and adoption – DeepMind deployed its Gemini model in the UK months after in the US.
The UK is not well placed to develop advanced AI. The cost to be competitive in the development of AI is increasingly beyond the means of UK companies or the state. The amortised cost of training GPT-4 is estimated to have been around $100m, and capital investments are rapidly growing – OpenAI and Microsoft are planning to build a $100b supercomputer in 2028. Meanwhile, the R&D budget of Nvidia alone was $8.7b in 2024, and the US government's CHIPS Act provided $39b in subsidies for semiconductor manufacturing and crowded in $327b of private investment. Training compute costs are doubling every nine months for the largest AI models.
Furthermore, electricity in the UK is more expensive than in the US, East Asia, and much of Europe, with prices rising 210% over the last two decades. The high electricity cost discourages companies from running frontier AI training here, forcing firms like Google DeepMind to use US-based data centres. The power required to train frontier AI models is doubling annually. By 2030, training runs will require energy equivalent to a city of 1 million people. In preparation, US tech giants are securing their own energy sources—for example, Microsoft is reopening the Three Mile Island nuclear plant and Meta is building geothermal energy to power their models. If the UK does not lower energy costs by ensuring stable, reliable power from firm power sources like nuclear, it will continue to be an unattractive place to build data centers and train AI models.
The broader academic community has failed to keep up with or update their worldviews based on recent advances such as LLMs, especially in the UK. This has stymied government support for the technology as the UK directs the majority of state AI investment via UKRI, an organisation largely oriented towards the support of academic research rather than broader strategic aims. The British AI community has been constrained by years of neglect for infrastructure investment in areas such as compute, as well as by a brain drain to industry and to other countries.
Universities have failed to prepare the next generation to lead and respond to transformations in AI. Every year Oxford University trains 3x more English Literature undergraduates than computer scientists, contrasting with the US’ recent spike in computer science majors. Meanwhile, Cambridge only offers 30 places per year in its class about natural language processing – the key discipline behind advances like ChatGPT. The comparable class at Stanford has an enrollment of over 500 students.
The UK lacks an economic strategy that can support industrial success in AI. The UK does not have significant control of any of the critical components in the AI supply chain, which reduces our geopolitical leverage and removes us from some of the greatest economic payoffs. UK businesses face immense energy costs and limited lab space in key innovation clusters, which places our firms at a significant disadvantage to international peers. Additionally, we lack the domestic capital for late-stage investment, causing many UK firms to list abroad. Government looked on passively while our key tech firms like ARM and DeepMind were forced into foreign takeovers to secure growth capital. Foreign capital will continue to be important for the UK tech sector, but our ambivalence about corporate control should not continue.
AI Should be Central to the UK’s National Strategy
The UK needs to prepare for the near-term availability of "intelligence too cheap to meter" that will underpin productivity and growth across the economy. Those nations best positioned to facilitate rapid, equitable adoption of these capabilities are most likely to gain a competitive edge. Those that best rapidly achieve adoption and diffusion will gain the benefits of productivity growth and national power. Current government priorities do not reflect this need for urgency. We are at the cusp of a technological revolution: it should be our top priority to prepare the country for comprehensive AI adoption.
We must seek strategic advantage through controlling critical components of a coordinated multilateral supply chain, across the "AI triad" of compute, data, and talent (Figure 4). The UK should position itself to lead these multilateral industrial efforts. Leadership in key Western alliances (European, G7) would give us a significant stake in the future of AI, just as Airbus and CERN help small (but highly advanced) European states engage in the largest industrial projects and push the frontier of fundamental science.
Figure 4: The AI Triad: The Three Technical Inputs to AI – Data, Algorithms, and Compute
Source: Computing Power and the Governance of Artificial Intelligence
For too long companies have ceased to be British-owned, in part, because they receive little advantage from it. The government must demonstrate that this is no longer the case. The state should actively incubate our strategically significant firms by prioritising them when taking investment or procurement decisions. The forthcoming Industrial Strategy should provide the framework to support this.
We will also need to take a more forward approach to exporting our technology. Other countries have had industrial policies that have been aggressive in subsidising and exporting strategic products such as steel, solar panels, telecommunications infrastructure, and electric vehicles. They have also proactively shaped international standards around them. We should be aiming to have the rest of the world be consumers of British AI models or products, because we build them better and can export them easily.
We must catalyze the formation of AI-first national champions that drive commercialization and diffusion across our economy. Given our size, we cannot lead in every domain or expect to have multiple $1tn companies like the US. Instead, we should focus on housing a series of firms that can dominate a few key technology niches. This will boost our national productivity, and increase our strategic autonomy and resilience.
We should make focused bets in AI sectors in which the UK could have comparative advantage. We cannot compete in every sector in the AI supply chain – especially given its concentration and high barriers to entry. Nor do we have the massive internal markets and fiscal resources of the USA, EU or China. We must instead maximise our competitive advantage by making focused bets which leverage our existing strengths, such as our skilled labour and knowledge-intensive industries. These focused bets should prioritise:
Box 1: AI sectors in which the UK could have comparative advantage
Our services economy is particularly exposed to AI, but we can transform this risk into an opportunity by being pioneers in AI adoption. Our legal, financial, and consulting sectors are well-positioned to lead adoption by leveraging their skilled talent and proprietary datasets. To achieve this aim, we need a government willing to procure AI-first services, regulators agile enough to adapt to rapidly changing industries, and legislative reforms that support AI adoption (e.g. by enabling open access to legal data).
Beyond private datasets, the UK also holds valuable, large-scale public datasets that give us an edge in ‘narrow model’ AI development. Data from sources like the Biobank and NHS health records can kick-start a sector focused on AI applications in pharmaceutical development, medical research, and healthcare delivery—areas of enormous potential value. This development not only contributes directly to the economy but also supports the government’s missions to build an NHS for the future and improve public service delivery.
The UK cannot match the scale of US compute giants or compete in chip manufacturing with East Asia, but we have a rich history in cutting-edge chip design. The Cambridge cluster, or Silicon Fen, emerged from the success of semiconductor firms like ACORN, Cambridge Silicon Radio, and the global tech leader ARM. These historical competencies, coupled with continued deep talent pools, position the UK to lean into R&D for AI accelerators, particularly in inference. The focus should be on emerging technology stacks where incumbents, like NVIDIA, have yet to build significant IP and infrastructure moats.
AI will also spawn entirely new industries, some of which the UK can lead. We already have a head start in AI Assurance Technologies (AIAT), which includes model evaluations, pre-training risk assessments, and post-deployment monitoring—fields open for the development of new companies. This would build on the talent cluster being built up around the world-leading AI Safety Institute and existing UK strengths in professional services. Deloitte’s 2024 acquisition of Gryphon Scientific (the group behind AI biosecurity evaluations for OpenAI and Anthropic) is an early indicator of the growing value of this market. The UK Government previously estimated the AIAT market could grow to £4bn a year in the UK, and investors are estimating a $276 billion AIAT market by 2030. Startups like Apollo Research and those being incubated in Entrepreneur First’s AI assurance cohorts are promising signs that the UK could become a key player in this sector.
There are also opportunities in ‘AI for science’ – the application of AI to power fundamental research and accelerate discovery. We have already seen promising examples, such as AlphaFold, which uses AI to perform predictions of protein structure and played a crucial role during the COVID-19 pandemic by speeding up our understanding of the SARS-CoV-2 virus. There are exciting opportunities for AI to be applied in drug discovery, genomics and DNA sequencing, climate science, quantum chemistry, and materials discovery. These advances could unlock new treatments for diseases, offer new strategies to combat climate change, and revolutionise industries, supporting UK strengths in sectors like life sciences and aerospace. The resulting IP from this research is also likely to confer huge economic benefits. The UK should be pioneering these AI-first R&D labs to lead the next phase of scientific discovery.
The UK should also actively court US tech companies to establish their second offices here, rather than losing them to competitors like Ireland. By capitalising on our world-class professional services in accountancy, law, finance, cybersecurity, and emerging fields like AI evaluations and assurance, we can offer critical support to AI and tech firms seeking a European base. This model cannot be the totality of an AI industrial strategy, but it should nevertheless be part of the portfolio, as it would allow us to capture some of the benefits that will inevitably flow to US tech giants who are currently building frontier AI models.
Scaling up our ambitions will require positioning the UK as an attractive destination for AI investment. Following the US, we should engage with international funding sources, including Gulf States. We should aim to strengthen our partnerships with our allies and limit China’s sphere of influence on AI, by facilitating investment access to UK AI startups and sharing the benefits of the technology.
Plan of Action
1) Infrastructure
Boost national AI capabilities by increasing investment into public-sector supercomputing capabilities, specifically the AI Research Resource and the Exascale programme. There are significant positive externalities to public supercomputing resources, which will enable the UK government to accelerate AI development and adoption, strengthen academic AI research to stem a brain-drain of talent abroad and to industry, while supporting early-stage entrepreneurship. Compute allocation could also be used to drive research priorities in line with the Government’s Missions.
We should aim to spend £1.5bn over 2 years, then repeat this process annually over the next five years, spending up to £10bn. We suggest this include:
One publicly funded compute provider building on AIRR and exascale using an "AI-first" access model, with (1) a comprehensive software and skills package to ensure proper migration for user community, (2) some well-defined research challenges that will accelerate utilisation, (3) private sector involvement.
Several machines for AI-enabled research, close to the academic community, with access awarded competitively.
The allocation of compute to research projects and the private sector should be performed by expert programme directors with the autonomy to appraise high-potential projects. The programme should not be devolved to the research community or funded through traditional UKRI allocation mechanisms.
Public-sector supercomputing is important for national scientific and strategic infrastructure and is distinct from typical private sector data centres. Private sector data centres are profit-driven, optimised for efficiency and scalability, and engineered to support a variety of commercial operations—like cloud computing, big data analytics, and online services. Public supercomputing facilities can be designed to support open-ended scientific research and national interests. Some are built to tackle the most complex, high-performance computing challenges that drive breakthroughs in critical areas such as climate modelling, genomics, and AI. Some handle sensitive data and conduct simulations that are vital to national security, requiring stringent security measures and government oversight. Public-sector supercomputing facilities are shared resources, fostering collaboration across research institutions and government agencies. Policies often encourage open access (where appropriate) to accelerate scientific progress and innovation.
As laid out in the Independent Review of the Future of Compute, public sector supercomputing is a normal service provided in many other countries like the United States (the Department of Energy’s supercomputers such as Frontier and Aurora), Japan (Fugaku), Finland (LUMI) and Italy (Leonardo). Just one of the USA’s supercomputer Summit is 7.5 times the size of the UK’s largest. The UK is an outlier as it has missed several ‘investment cycles’ and thus lags behind other states.
We should urgently pursue multilateral compute partnerships with allies to ensure access to resources beyond the capabilities of the UK. This would allow us to pool resources, provide shared access to existing supercomputing resources, and jointly build new supercomputers. Our membership of EuroHPC is a strong step, but this can be deepened through joint investments and co-location. There is also significant potential in pursuing partnerships with other allies such as Canada (which has a strong AI talent base) or Australia.
Incentivise data centre build outs in the UK by removing energy barriers to ensure the UK capitalises on the economic and strategic benefits of having data centres onshore. Market-driven data centre build outs could continue to be positive for the UK. However, we currently have a massive competitive disadvantage in building data centres in the UK because of the high cost of energy.
We should enable AI companies to create their own private energy sources, as is happening in the US. Connecting a new project to the grid in the UK can take 10-20 years. This is simply too long. We need to allow power plants to be built that are not connected to the grid and that only provide electricity to data centres. In doing so, we can tap unused resources: some renewable sources – such as wind turbines off the West Scottish coast – may not be economical if the electricity must be transmitted to cities but would be economical to power a data centre.
Most importantly, however, will be a ‘dash for nuclear’. AI companies need the base-load energy of nuclear energy for training and inference. Investments in energy solutions like nuclear generate high-quality local jobs and build national expertise in infrastructure construction. This expertise could lead to downstream cost savings and boost the efficiency of future civil build-outs.
Given the increased demands of AI infrastructure, the government should update its projections for required electricity generation capacity. The government should use Great British Energy and the National Energy System Operator as an opportunity to increase energy supply, which would improve grid resilience and stability, as well as lowering costs for British businesses and consumers.
The government should remove planning barriers to speed up construction. Data centres should be considered a Critical National Priority, as has been done for some low carbon infrastructure. It should also easily accelerate delivery of data centers by creating a permitted development right for data centres that meet certain criteria. Government should also include data centres in its plans to create ‘planning passports’, passing new legislation to create a specific kind of ‘planning passport’ for data centres.
While we should remove barriers that prevent private-sector construction of data centres, the state should not seek to subsidise private-sector buildout.
2) Government Funding
Leverage the government's purchasing power to boost the UK supply chain of new technologies. This should be done by launching an Advanced Procurement Agency (APA) to drive AI adoption across the private and public sector, which would serve as a nimble organisation to develop, procure, and trial innovative early-stage AI products in public services, acting as a "buyer of first resort" for proof-of-concept work.
This can build on the ‘digital centre’ at DSIT formed by bringing together the Government Digital Service (GDS), the Central Digital and Data Office (CDDO) and the Incubator for AI (i.AI).
The APA should be run with empowered externally-hired ARIA-style programme managers who are "deployed" into the field and have the autonomy to innovate in specific areas of public services, such as trialling AI assistants in a single classroom or school. This approach would create crucial market opportunities for early-stage startups while de-risking, and paving the way for, broader adoption across public services.
Lessons from these programmes would also inform regulatory efforts in different domains of AI applications, making sure regulations are fit for purpose and efficient.
The APA could also run innovative large-scale efforts such as using advanced market commitments to drive innovation, setting targets of products in government-priority technologies, and pre-committing to purchase these if they are developed by UK firms.
In time, the role and powers of the APA should be enshrined in legislation to prevent the APA falling victim to short-term political expediency. However, to allow swift action in advance of legislation, the Government should first establish a minimally viable APA on the basis of existing legal powers. This initial iteration of APA could be targeted at the government’s mission areas.
Evolve and modernise our R&D institutional architecture to ensure it can capitalise on and respond quickly to the growing opportunity around AI research. This should be done by creating a programme to fund new AI-first "disruptive innovation labs" across all areas of the economy. We need to capitalise on rapid acceleration of R&D progress as frontier AI capabilities are applied to science and technology.
Modern metascientific analysis shows that innovation and diversity in organisational form and function allow for new frontiers of progress to be opened up. Today, we largely see this institutional entrepreneurship and experimentation in AI abroad, through new philanthropically-funded institutions such as Future House (that aims to create an AI scientist, led by an ex-Crick researcher who left London to build in San Francisco), Arc Institute and Kyutai. Funding Research Ventures Catalyst is a good start, but we need to go further.
The state should fund the creation of new AI-first research institutes led by ambitious junior scientists in narrow domains, which could also crowd-in significant philanthropic funding. These new institutions should be tailored to the gaps and opportunities in the UK landscape.
We should also create a resilient ecosystem by actively cultivating and scaling new startups that can commercialise the inventions created by our R&D ecosystem. Google DeepMind is single-handedly responsible for our global lead, but is ultimately American-owned. We need to create conditions for a more resilient and diverse ecosystem. This could be done by working with philanthropists on projects such as supporting the creation of a physical AI community hub in Kings Cross, similar to the €250mn investment by a French billionaire to build the Station F startup campus in Paris.
3) Incentives for Adoption and Competition
Facilitate the creation and diffusion of AI in the UK economy by giving the Regulatory Innovation Office (RIO) the mandate to accelerate adoption, both via accelerated sectoral pathways (e.g. AI in healthcare, one of its initial priority areas) but also by surfacing and resolving sector-agnostic regulatory concerns (e.g. enterprise liability risks).
The RIO should add AI adoption as one of its priority areas. It should provide clear guidance and rapid responses to businesses on regulatory compliance for AI deployment, make liability issues clear to them, and drive the development of new markets for AI risk assessment and insurance capabilities which will support further business adoption.
The Government should also gather feedback on the frontier AI Bill and introduce it in the next King’s Speech. This Bill will monitor for, and prevent, large-scale harm. It can be presented as a package for frontier AI companies alongside other reforms: the UK will balance additional scrutiny against unleashing their potential for growth. Such a tradeoff has historical precedent – in 2015 the UK’s four mobile network operators extended mobile coverage in exchange for reform of the Electronic Communications Code.
Regardless, we must not let regulatory barriers hamstring our ability to innovate or we will be left behind. This capability will be built – whether we like it or not. If the UK fails to be a place friendly to AI, talent will continue its exodus to places like the US.
Incentivise adoption across our current industries, with a particular focus on services. Our heavily service-oriented economy is particularly exposed to AI, but we can transform this risk into an opportunity by being pioneers in AI adoption.
Government should encourage UK firms and AI companies to collaborate, especially in services. Large AI companies have the freedom to choose almost any company on the planet to partner with. We want them to collaborate with UK firms. To achieve this goal, the UK government can play a brokerage role, convening meetings between AI companies and key UK firms, or an enabling role, providing public compute as an incentive for joint ventures. Target UK firms should have large pools of proprietary data and large pools of talented staff. Focuses should include professional services (accountancy, law, finance, insurance, management consultancy), biotech, advanced manufacturing and defence. Examples could include: AstraZeneca & GSK; HSBC, Lloyds, Barclays, Aon; Deloitte, EY, KPMG, and PwC; A&O Shearman, Clifford Chance, Freshfields Bruckhaus Deringer, Linklaters, and Slaughter and May;Rolls-Royce & BAE.
By fostering collaboration between UK services firms and AI companies, government will enable the UK to capture more of the economic value of success.
We should actively court US tech companies to establish their second offices in the UK. We can do this by offering a ‘one-stop shop’ for professional services – everything these companies need in terms of accountancy, law, finance, insurance, management consultancy, cybersecurity and AI assurance. We can also continue to make the UK an attractive place to live and work through supporting our universities and talent base, and building more housing in the tech hubs of London, Cambridge and Oxford. This would allow us to capture some of the benefits that will flow to tech giants who are currently building frontier AI models, and further encourage adoption and talent agglomeration.
Leverage the UK’s valuable datasets to support areas of competitive advantage in AI. This should be done by mandating the National Data Library's (NDL) primary role to be the collection, curation, and processing of high-quality datasets and data collection infrastructure. The costs of key datasets that have catalysed the most significant AI advances of the past decade have been in the range of hundreds of thousands to the low millions of dollars, representing a huge return on investment and crowd-in effect. The NDL should aim to create data we want but don't yet have and curate and process high value existing information sets that are currently inaccessible, non-machine-readable, and highly disparate. To do this, the NDL should:
Provide access to state of the art data collection infrastructure for all of government, UK universities or research institutions, UKRI-funded researchers and UK startups. It should aim to build this infrastructure where possible, via GDS or i.AI, but should otherwise contract with leading data labelling firms for access.
Expand existing data efforts at the NHS, DfE, and HMRC and build synthetic versions of the most interesting government datasets in a privacy-preserving manner for public release.
Put out a public call on what datasets to create, and run data creation challenges through UKRI to drive AI research and attract talent.
Consider changes to UK institutions that enable valuable new datasets to be leveraged including providing open access to legal data.
Provide access to its datasets through a tiered approach, from free access for UK startups and academics based in UK institutions to significant fees for foreign companies.
Take a strategic bet on hardware design for purposes other than AI training by using public procurement to purchase hardware designs through a new Advanced Procurement Agency and directing R&D funding through the Semiconductor Institute. AI training chips have received by far the most international attention, despite being just one type of relevant hardware. It is infeasible for us to independently compete on chip manufacturing or designing hardware for AI training, given the immense costs required to enter these spaces now (see UK efforts via Graphcore, or the scale of US CHIPS act subsidies). However, we possess significant chip design expertise, primarily centred around ARM and with an additional cluster in Bristol. We should capitalise on that advantage by:
strategically focusing on inference/serving hardware design—a much earlier-stage area with no clear leaders yet. This segment is particularly promising, as demand for inference hardware currently accounts for 60-80% of compute allocation at frontier AI labs.
investing in other areas of chip design such as sensors and embedded processing (e.g. Pragmatic Semiconductor) in preparation for growth in edge AI in peripherals and robotics.
leveraging UK expertise in hardware security by design (e.g. CHERI at the University of Cambridge) to offer secure inference hardware for safety and security critical applications.
Foster national and international collaborations and partnerships to grow a thriving AI Assurance industry in the UK. This sector will provide pre-training risk assessments, model evaluations, system conformity assessments and post-deployment monitoring. The UK Government estimated it could soon grow into a £4bn a year market in the UK, and investors are estimating a $276 billion AI Assurance Technologies (AIAT) market by 2030. The UK has a head start in AI Assurance thanks to our world-leading professional services firms, the talent cluster around the world-leading AI Safety Institute (AISI), and emerging start-ups in the sector.
The Government can support this emerging sector by encouraging three-way collaboration between the AISI, assurance start-ups and professional services firms; supporting the sector with existing start-up grant funding; continuing the successful secondment system with the AISI to ensure we can stay in the lead on AIAT; and ensuring that the Frontier AI Bill supports the AI Assurance sector.
The UK must be set up as an international assurance partner so we can export to bigger markets. For example, this means ensuring that UK ‘conformity assessment bodies’ are authorised to run assurance tests under Article 39 of the EU AI Act. Authorisation would allow us to export AI assurance, products, and services to the EU with minimal frictions. We should also aim for observer status at the EU-US Trade and Technology Council, a forum where our two leading export markets are setting shared standards. We need to be at the table.
4) Talent
Find and support the next generation of outlier global and domestic talent that will lead invention and innovation by creating an Exceptional Talent Office (ETO). ETO should bring in staff with recruitment expertise from leading AI companies or quantitative trading firms, and take a Moneyball-inspired approach to finding this talent. Projects to prioritise immediately include:
Creating a flagship international scholarship programme in STEM (equivalent to the Rhodes, Marshall, or Chevening) to give full scholarships for exceptional STEM talent (for example, International Math and Science Olympiad medallists) to pursue undergraduate study in the UK. We spend £60M/year to support foreign policy objectives via Chevening, and should aim to spend at least £10m/year comparably to attract and integrate the world’s best STEM talent.
Funding and running domestic STEM competitions to discover and support hidden talent of all ages, particularly from under-represented communities in the UK that might otherwise be overlooked.
Driving visa policy reform, particularly through:
Giving ETO a yearly allocation of Global Talent Visas, with some direction on priority areas (based on industrial strategy), such that the ETO, in concert with departmental policy teams and research bodies, can identify and directly offer an immigration pathway to outlier global talent without additional layers of Home Office bureaucracy.
Widening access to the High Potential visa, increasing the list of universities qualifying for the High Potential Visa by using additional metrics like top 200 universities for graduate earnings and the top 30 universities on each continent, and the top 20 research institutions.
Working with universities to drive adoption of industry/startup internships as a core part of an undergraduate education, drawing upon the extremely successful co-operative education model of Canada's University of Waterloo, or France's CIFRE PhD model. This will shortcut the transfer of information about frontier capabilities from industry to academia. The ETO should aim to partner with AISI, Google DeepMind, and other top AI companies for an initial cohort. Programmes should be developed where students do placements in Silicon Valley (or comparable regions for different disciplines).
Ensure government has the necessary expertise to capitalise on this techno-economic revolution by recruiting experts in AI into government, embedding them at the highest levels of authority, and giving them autonomy to deliver. Government needs to recruit people with technical expertise that understand the gravity of the moment and can help shape policy. This expertise will not only come from the civil service. We should be drawing in a wider set of junior technical talent who understand the frontier of research rather than relying on a small pool of senior external advisors.
The network of Chief Scientific Advisers should expand to include at least one prominent expert in AI, appointed as Government Chief AI Adviser.
The Prime Minister should establish an AI cabinet committee which takes regular stock of the shifting landscape and approves cross-government action.
Private sector companies have starting salaries of $800k+. To compete with these offers and attract top talent, we must provide them with autonomy, a promise to cut red tape, and a competitive salary. The government science profession's pay scales need to be updated to reflect the more competitive job market for AI companies, and an exceptional pay process needs to be put in place to hire top AI talent into key roles where needed.
FAQs
How are other governments investing into AI?
Here is a non-exhaustive list of investments currently being undertaken by other countries. Note that this is a list of publicly announced investments. There are likely a number of investments that, for national security reasons, have been kept confidential.
How does this impact Labour’s 5 missions?
Kickstart Economic Growth: Recent technological developments put Britain’s economic future on the line. More than 60% of jobs in advanced economies are exposed to AI. AI impacts high-skilled jobs and, in particular, service-oriented roles. The UK will not grow if it does not succeed in capturing the opportunities afforded by AI, and may even shrink if it fails to adapt to this technological change.
Make Britain a Clean Energy Superpower: Developments in technologies like AI mean that the UK is likely to need more electricity than current projections suggest. We need to be more ambitious about our clean energy goals to provide compute and build out AI in the UK. The government should focus on decreasing energy costs. Energy costs are more expensive than in the US, East Asia, and much of Europe and have increased by 210% in the last two decades. High costs negatively impact investment and are likely depressing gains in AI in the UK. A global increase in demand for energy (driven by AI and other factors) also presents a significant economic opportunity to export clean energy. Manchester AI Prize finalists Aiolus and Quartz are applying AI solutions to boost wind energy and solar energy production respectively. New AI-first firms could further boost British clean energy production.
Take Back our Streets: AI can provide an opportunity to improve public sector delivery while creating market opportunities for British AI startups. Crime mapping supported by AI can better inform “hot spots policing”, an evidence-based approach to reducing crime. Developed correctly and leveraging AI Assurance Technologies, AI can bolster our criminal justice system.
Break Down Barriers to Opportunity: Government should be actively recruiting and investing in ‘lost einsteins’ in the UK – people (primarily from underrepresented groups) who would have had high-impact inventions had they been exposed to the right opportunities. This talent identification can be done through domestic STEM competitions.
Build an NHS Fit for the Future: AI can provide an opportunity to improve public sector delivery while creating market opportunities for British AI startups. AI solutions in healthcare could reduce the administrative burden on doctors and nurses, giving them more time with patients. Developed correctly, AI can provide capacity for our NHS.
How is public-sector supercomputing distinct from private-sector data centres? Why should government policy be different in relation to each?
Public-sector supercomputing is important for national scientific and strategic infrastructure, fundamentally different from the typical data centres found in the private sector. While private data centres are engineered to support commercial operations—like cloud computing, big data analytics, and online services—public-sector supercomputing is built to tackle the most complex, high-performance computing challenges that drive breakthroughs in critical areas such as climate modelling, genomics, and AI.
Supercomputing facilities support open-ended scientific research and national interests that extend beyond the profit-driven objectives of private-sector data centres. These supercomputers often handle sensitive data and conduct simulations that are vital to national security, requiring stringent security measures and government oversight. Unlike private-sector data centres, which are optimised for efficiency and scalability, public-sector supercomputing facilities are shared resources, fostering collaboration across research institutions and government agencies. Policies often encourage open access (where appropriate) to accelerate scientific progress and innovation.
Acknowledgements
Our thanks to Ben Johnson, Liam Patell, and Andrew Bennett for invaluable comments, editing, fact-checking and research.
Julia Willemyns is Founding Co-Director of UK DayOne, where she focuses on science and tech policy, high-skilled STEM immigration, and energy infrastructure. She previously worked at Schmidt Futures on science and tech policy and, before that, was an entrepreneur. Julia holds degrees from the University of Oxford and the London School of Economics.
Haydn Belfield is co-Chair of the Global Politics of AI project and Research Fellow at the University of Cambridge’s Leverhulme Centre for the Future of Intelligence. He has also for the past seven years been a Research Associate and Academic Project Manager at the University of Cambridge’s Centre for the Study of Existential Risk. In that time the Centre tripled in size, and he advised the UK, US, and Singaporean governments; the EU, UN and OECD; and leading technology companies.
Tom Milton
Tom Milton is Technology Policy Lead at UK DayOne. Tom is an engineer and startup founder with experience across medtech, biotech, and deep tech. He has worked for a leading UK medtech startup, was the first employee at a biotech startup, and most recently founded his own engineering design company, Amodo Design. His main interests are in encouraging impactful science and technology inventions, differential technological development, and supporting the UK startup ecosystem.