## 1. AI Adoption Over Time and Across Countries

## Source details

**Canonical URL:** [1. AI Adoption Over Time and Across Countries](https://www.imf.org/-/media/files/publications/selected-issues-papers/2026/english/sipea2026072.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/selected-issues-papers/2026/english/sipea2026072.pdf.md)
- [Structured JSON version](/-/media/files/publications/selected-issues-papers/2026/english/sipea2026072.pdf.json)

---

### Executive summary and key findings
- AI technologies offer a significant opportunity to lift the UK's productivity growth; the UK is well positioned given its concentration in high-skill and knowledge-intensive services.
- Using a model of endogenous technology adoption calibrated to the UK, halving regulatory constraints and expanding the supply of AI-related skills could jointly increase the output gains from AI by two thirds over the baseline.
- The government's AI strategy covers the right areas but greater priority could be given to clarifying the regulatory framework and expanding the supply of AI-relevant skills, which the analysis identifies as the most pressing constraints.
- Investment in the domestic AI ecosystem has substantial insurance value: even moderate disruptions to foreign model access could generate large output losses.

### Potential of AI for UK productivity and growth
- Literature and empirical ranges:
  - Medium-term impact on aggregate TFP level ranges from below 1 to 35 percent over a decade across studies.
  - Misch and others (2025) estimate the UK’s TFP gains could range from 0.1 to 0.5 percent per year in the medium term.
  - Studies find AI could support the performance of between 23 to 80 percent of tasks.
  - Micro-economic studies suggest productivity could improve by between 15 and 55 percent on AI-exposed tasks.
- UK positioning and labor composition:
  - Roughly 36 percent of the workforce holds a college degree.
  - Nearly two-thirds of the workforce are employed in AI-exposed occupations.
  - More than half of AI-exposed workers are in high complementary occupations.
  - UK sectors with largest potential gains include business services, finance, IT, and pharma.
  - Various measures of labor costs place the UK above the OECD average.

### Bottlenecks to AI development and diffusion (five enabling conditions)
- Five enabling conditions examined:
  - Infrastructure (ICT and energy).
  - Financing of innovation (scale-up and investment for high-growth firms).
  - Supportive and predictable regulation.
  - Skills (supply of AI-relevant workers).
  - Resilience through trade openness (access to frontier inputs and market opportunities).
- Empirical drivers and survey evidence:
  - Econometric analysis and variance decomposition: AI-relevant skills are the most important driver of cross-country AI adoption; infrastructure is the second-most important; openness ranks next.
  - Business surveys emphasize regulatory framework and access to financing as key factors shaping adoption and innovation.
- UK-specific bottlenecks:
  - Barriers to ICT and energy infrastructure construction.
  - Constraints on business scale-up limiting financing of innovation.
  - Regulatory uncertainty slowing organizational and process transformations.
  - Shortages of AI-relevant skills.
  - Trade and geo-fragmentation risks to sourcing critical inputs.

### AI adoption levels, trends, and survey insights (Box 1 highlights)
- Adoption levels and trends:
  - As of mid-2025, business surveys indicate the AI adoption rate in the UK equaled 34 percent, close to the EU average of 37 percent and the US rate of 36 percent.
  - UK AI adoption rates have almost doubled from 2023 to 2025.
- Usage patterns and constraints:
  - Above-average adoption in data processing, image processing, and robotics; below-average rates in text generation using LLMs.
  - Leading reported obstacles (2024–25): skills shortages and unclear use case; costs & access to finance; lack of trust and technical challenges; regulation concerns.
  - Nearly 60 percent of respondents in a government-commissioned survey cited regulatory and policy obstacles as a barrier to AI adoption.
- ICT and energy infrastructure specifics:
  - Some sites face waits of up to 5 years for grid connection.
  - UK ranks highly on IMF global AI Preparedness Index for ICT infrastructure, near the middle of the G7, but lags in AI supercomputer capacity and faces the highest electricity prices in the G7.
  - Real investment in AI-related infrastructure has grown at a slower pace than peers; new construction orders have been stagnant since the COVID-19 pandemic.
- Financing for innovation specifics:
  - Gross domestic expenditure on R&D stood at 2.75 percent of GDP in 2022 (OECD average 2.72 percent).
  - UK VC investment represented 0.17 percent of GDP in 2024, supporting more than 9,100 businesses.
  - In 2024 in the AI sector, UK recipients raised only 40 percent of VC financing domestically, while 80 percent of UK’s VC financing was channeled abroad.
  - Seed-stage companies attract almost three-quarters of funding domestically; at the scale-up stage domestic share falls to less than one-third.
  - Restructuring costs estimated at 19 months of salary per employee in the UK, compared with 7.4 months in the United States and under 3 months in Denmark and Switzerland.
- Skills and labor market:
  - The UK workforce is highly educated and scores highly on the human capital dimension of the IMF AI Preparedness Index, but ranks below average on perceived skillset of graduates.
  - As of 2025, more than 70 percent of UK companies surveyed by the British Chamber of Commerce report recruitment difficulties; ICT industry is the second most affected.
  - One in three UK founders of Tech firms says that availability of talent is their biggest barrier to growth.
- Trade openness and concentration risk:
  - Digital trade represents more than half of the UK’s exports, twice the OECD and EU averages.
  - The UK depends heavily on US-headquartered hyperscalers for compute, cloud infrastructure, and foundation models; the UK sits in Tier 1 of the US AI Diffusion framework and faces no procurement restrictions on advanced GPUs.

### Model framework, baseline calibration, and quantified impacts (Box 2 and model summary)
- Model structure and baseline assumptions:
  - Endogenous technology adoption calibrated to the UK; occupations are task-based with varying AI exposure; firms adopt AI where AI capital is cheaper than labor.
  - Baseline anchors:
    - Long-run scale of adoption: AI raises the aggregate capital share by approximately 10 percentage points in the long run.
    - Pace of adoption: AI reaches the long-run scale at twice the pace of routine automation—steady state reached in 17 years rather than 34.
    - Productivity per automated task: cost-saving gains from AI set to 27 percent.
- Baseline model outcomes:
  - Cumulative TFP gains of 1.4 percent at the 10-year horizon.
  - Cumulative output gains of approximately 3 percent within the first 5 years, rising to around 8 percent over a decade.
- Commoditization and cost evidence:
  - Cost of using AI models cited as falling by roughly 90 percent per year; 99.7 percent decline in the cost efficiency frontier since March 2023 (Appenzeller 2024).
  - Open-source models now perform within 2 percent of frontier models on key benchmarks (Stanford HAI 2025).
  - Baseline assumes continued partial transmission of cost declines to users; final scenario relaxes this assumption.

### Policy scenarios, calibrated experiments, and quantified impacts
- Regulatory uncertainty scenario:
  - Calibration: regulatory uncertainty holds back 20 percent of potential AI adoption on average; reform modeled as resolving half of this constraint.
  - Impact: regulatory reform would boost output by 1.24 percentage points at Year 5 and by 1.72 percentage points at Year 10; TFP: +0.22 at Year 5; +0.35 at Year 10.
- Skills investment scenario:
  - AI labor share of adoption costs calibrated to 30 percent; AI wage premium currently estimated at 14 percent in the UK.
  - Modeled congestion markup on adoption costs grows from around 1.15× at year 5 to nearly 1.3× by year 10.
  - Halving the AI wage premium (via training, vocational pathways, and immigration) implies increasing AI labor supply by approximately 27 percent at year 5 and 43 percent by year 10.
  - Impact: output gains +0.83 percentage points within 5 years; +1.58 percentage points by year 10. TFP gains: +0.15 percentage points at Year 5; +0.31 percentage points at Year 10.
- Comparison and joint effect:
  - Regulatory reform as percent of baseline output: +38% (Year 5), +22% (Year 10).
  - Skills investment as percent of baseline output: +25% (Year 5), +20% (Year 10).
  - Pursued together, regulatory and skills policies could jointly increase the output gains from AI by two thirds over the baseline.
- Concentration risk and domestic capacity scenarios:
  - Restricted access to foreign frontier AI modeled as a markup on adoption costs; domestic alternative assumed to deliver 90 percent of frontier benefits conditional on adoption.
  - Disruption markups considered: τ = 15 percent; τ = 25 percent; τ = 40 percent.
  - Central scenario (τ = 25 percent) at Year 5:
    - Baseline TFP gain 0.53pp reduced to 0.08pp under restricted access.
    - Baseline output gain 3.26pp reduced to 0.54pp under restricted access.
    - Equivalent to approximately £82 billion per year in lost output relative to the baseline.
    - A domestic alternative recovers roughly half of this loss, worth approximately £40 billion per year.
  - Severe disruption (τ = 40 percent): domestic option recovers approximately £57 billion per year.
  - Conclusion: modest investments in domestic resilience could recover a substantial share of output lost under disruption scenarios; insurance value exceeds the government's current AI commitment of over £2 billion.

### Econometric evidence on cross-country drivers (Annex II)
- Regression specification and data choices:
  - Cross-section regressions of AI adoption rate mid-2025 on ICT infrastructure, human capital, trade openness, electricity prices, log GDP per capita, and the World Uncertainty Index.
  - Timing: pre-adoption covariates used (2019 values for some variables; human capital and infrastructure indices based on 2023 or before).
- Key coefficient estimates (selected, exact values preserved):
  - Human capital: coefficients range from 4.80** to 5.69*** across specifications (standard errors reported in source).
  - ICT infrastructure: coefficients range from 6.59** to 8.69***.
  - Openness: coefficients range from 1.49 to 2.33**.
  - Log electricity price: -2.77* in one specification; other specifications vary.
  - Uncertainty (World Uncertainty Index): -2.01** in one specification.
- Fit and sample:
  - Observations: 30 in columns (1)–(3); 26 in columns (4)–(5).
  - R-squared ranges: 0.759 to 0.869 across specifications.
- Variance decomposition:
  - AI-relevant skills (human capital) are the most important unique driver of cross-country variation in AI adoption; ICT infrastructure is the next most important; trade openness follows.
  - Implication: policies addressing skills, infrastructure, and openness target the largest uniquely attributable components of adoption variation.

### Policy implications and prioritized recommendations
- Complementary policy mix:
  - Economy-wide reforms (planning, energy markets, scale-up financing) are fundamental prerequisites for AI development and adoption.
  - Targeted AI-specific interventions in regulation, skills, and supply-chain resilience are needed to address market failures specific to AI.
- Top priorities indicated by the analysis:
  - Clarify and reduce regulatory uncertainty to lower adoption costs and encourage organizational change.
  - Expand the supply of AI-related skills (near-term: streamline immigration; medium-term: structured employer co-investment and postgraduate/advanced training).
  - Ensure infrastructure (ICT and energy) projects are not impeded by construction barriers to support AI deployment at scale.
  - Improve access to scale-up financing for innovation to support domestic AI ecosystem development and provide insurance against foreign model access disruptions.
- Economy-wide actions and targeted interventions (selected, exact commitments and figures preserved):
  - Planning reforms: 2025 Planning and Infrastructure Act; datacenters designated as critical national infrastructure in September 2024; AI Growth Zones offer fast-track pathways.
  - Energy/grid reforms: Ofgem connection reform in April 2025; Clean Power 2030 Action Plan; final investment decision for Sizewell C.
  - Innovation finance instruments: National Wealth Fund (NWF) with £27.8 billion capital over nine years; £500m Sovereign AI Unit; £100m advance market commitment for UK AI hardware startups.
  - Government skills commitments: £200m to upskill workforce; £17.6m Spärck AI 100 scholarships; £160m TechFirst program; Global Talent Taskforce.
  - Compute and data investments: £1.75bn AI Research Resource; £100m National Data Library.
  - Regulatory and safety commitments: Data Use and Access Act (DUAA) passed in 2025; £240m for the AI Security Institute.
- Equitable and responsible deployment:
  - Complement supply-side interventions with redistributive policies to address displacement and inequality.
  - Government role in ensuring broad sharing of gains and responsible deployment.
- Central challenge:
  - Delivery, scale, and agile course-correction to translate strategy into implementation at the pace and ambition required.

*Source: IMF staff analysis as presented in the chapter "AI Adoption Over Time and Across Countries" in the United Kingdom Selected Issues paper (July 2026).*

### 1. AI  Adoption  Over  Time  and  Across  Countries  ______________________________________8

### 1. AI  Adoption  Over  Time  and  Across  Countries  ______________________________________8

### Executive summary and key findings
- AI technologies offer a significant opportunity to lift the UK's productivity growth; the UK is well positioned given its concentration in high-skill and knowledge-intensive services.
- Using a model of endogenous technology adoption calibrated to the UK, halving regulatory constraints and expanding the supply of AI-related skills could jointly increase the output gains from AI by two thirds over the baseline.
- The government's AI strategy covers the right areas but greater priority could be given to clarifying the regulatory framework and expanding the supply of AI-relevant skills, which the analysis identifies as the most pressing constraints.
- Investment in the domestic AI ecosystem has substantial insurance value: even moderate disruptions to foreign model access could generate large output losses.

### Potential of AI for UK productivity and growth
- Literature estimates of aggregate TFP gains from AI vary widely:
  - Medium-term impact on aggregate TFP level ranges from below 1 to 35 percent over a decade across studies.
  - Misch and others (2025) estimate the UK’s TFP gains could range from 0.1 to 0.5 percent per year in the medium term.
- Two principal determinants of aggregate AI gains (model framework):
  - Share of AI-exposed tasks (legal/technical ability to automate and regulatory permissibility).
  - Average AI productivity gains and cost savings from capital-labor (K-L) substitution.
- Empirical task- and firm-level evidence:
  - Studies find AI could support the performance of between 23 to 80 percent of tasks.
  - Micro-economic studies suggest productivity could improve by between 15 and 55 percent on AI-exposed tasks.
- Why the UK is well positioned:
  - Roughly 36 percent of the workforce holds a college degree.
  - Nearly two-thirds of the workforce are employed in AI-exposed occupations.
  - More than half of AI-exposed workers are in high complementary occupations (where AI is likely to enhance worker productivity rather than substitute).
  - UK occupations with the largest potential AI productivity gains (business services, finance, IT, pharma) are prevalent in the UK economy.
  - Various measures of labor costs place the UK above the OECD average, implying relatively more to gain from labor-saving AI adoption.

### Bottlenecks to AI development and diffusion (five enabling conditions)
- The five enabling conditions examined:
  - Infrastructure (ICT and energy).
  - Financing of innovation (scale-up and investment for high-growth firms).
  - Supportive and predictable regulation.
  - Skills (supply of AI-relevant workers).
  - Resilience through trade openness (access to frontier inputs and market opportunities).
- Empirical and survey evidence on drivers of AI adoption:
  - Econometric analysis and variance decomposition of adoption rates indicate:
    - AI-relevant skills are the most important driver of cross-country AI adoption.
    - Infrastructure is the second-most important driver.
    - Openness (trade) ranks next.
  - Surveys of businesses emphasize:
    - Regulatory framework and access to financing are key factors shaping the pace of AI adoption and innovation.
- Identified UK-specific bottlenecks:
  - Barriers to the construction of ICT and energy infrastructure could impede deployment at scale.
  - Constraints on business scale-up limit financing of innovation for fast-growing AI firms.
  - Regulatory uncertainty may slow investment in organizational and process transformations enabled by AI.
  - Shortages of AI-relevant skills could limit effective adoption and realization of productivity potential.
  - Trade and geo-fragmentation risks to sourcing critical inputs could constrain AI diffusion in the future.

### Model insights and policy-relevant diagnostics
- Model determinants emphasized:
  - Share of AI-exposed tasks and average AI productivity gains and cost savings jointly determine economy-wide productivity gains.
- Feedbacks and dynamic considerations:
  - Current estimates of AI’s potential may be conservative because AI can accelerate the innovation process and positive feedbacks can expand AI capabilities and gains over time.
- Lessons from past technology revolutions:
  - The ICT revolution shows economy-wide productivity gains depend on effective adoption by AI-using sectors and complementary factors (see Annex I for details referenced in the source).

### Policy implications and priorities
- Complementary policy mix:
  - Economy-wide reforms (planning, energy markets, scale-up financing) are fundamental prerequisites for AI development and adoption.
  - Targeted AI-specific interventions in regulation, skills, and supply-chain resilience are needed to address market failures specific to AI.
- Top priorities indicated by the analysis:
  - Clarify and reduce regulatory uncertainty to lower adoption costs and encourage organizational change.
  - Expand the supply of AI-related skills to maximize the realization of AI-exposed tasks’ potential and to be the strongest driver of adoption.
  - Ensure infrastructure (ICT and energy) projects are not impeded by construction barriers to support AI deployment at scale.
  - Improve access to scale-up financing for innovation to support domestic AI ecosystem development and provide insurance against foreign model access disruptions.

*Source: IMF staff analysis as presented in the chapter "AI Adoption Over Time and Across Countries" in the United Kingdom Selected Issues paper (July 2026).*

### Box 1. United Kingdom: AI Adoption Over Time and Across Countries

### Box 1. United Kingdom: AI Adoption Over Time and Across Countries

### AI Adoption: levels and trends
- As of mid-2025, business surveys indicate that the AI adoption rate in the UK equaled 34 percent, close to the EU average of 37 percent and the US rate of 36 percent.
- UK AI adoption rates have been growing fast, having almost doubled from 2023 to 2025.
- Cross-country survey insights on AI usage purpose:
  - Above-average adoption in data processing, image processing, and robotics.
  - Average rates in visual content creation and autonomous vehicles.
  - Below-average rates in text generation using LLMs.
- Survey caveats:
  - Business surveys yield only rough proxies for effective AI adoption; cross-survey differences arise from question formulation, timing, and sector/firm-size coverage.
  - Worker statistics and surveys yield similar results.

### Obstacles to adoption (survey evidence, 2024–25)
- In the ONS and other surveys, leading reported obstacles include:
  - Skills shortages and unclear use case.
  - Costs & access to finance.
  - Lack of trust and technical challenges.
  - Regulation concerns.
- In a government-commissioned survey, nearly 60 percent of respondents cited regulatory and policy obstacles as a barrier to AI adoption.
- Datacenter and power constraints:
  - Some sites face waits of up to 5 years for grid connection.
  - Businesses can outsource computing to circumvent some issues; recent planning reforms have begun to address acute bottlenecks.

### ICT and Energy Infrastructure: strengths and headwinds
- UK strengths:
  - Robust ICT infrastructure and sizeable datacenter capacity relative to comparably sized European peers.
  - Ranks highly on the IMF global AI Preparedness Index for ICT infrastructure, near the middle of the G7.
- UK weaknesses and constraints:
  - Many existing datacenter facilities are designed for general-purpose enterprise workloads and lack the power density, energy integration, and technical specifications needed for AI training and inference.
  - The UK lags in AI supercomputer capacity.
  - The UK faces the highest electricity prices in the G7.
  - Real investment in AI-related physical and digital infrastructure has grown at a slower pace than peers.
  - New orders for construction of public and private infrastructure have been stagnant since the COVID-19 pandemic, with signs of more recent activity concentrated in the public sector.
- Cross-country analysis:
  - Electricity prices are a significant determinant of AI adoption rates, suggesting elevated UK prices are a headwind for AI development.

### Financing for innovation: strengths and structural gaps
- UK R&D and VC landscape:
  - Gross domestic expenditure on R&D stood at 2.75 percent of GDP in 2022, marginally above the OECD average of 2.72 percent.
  - The UK is Europe's largest VC market, ranking third globally and growing rapidly.
  - UK VC investment represented 0.17 percent of GDP in 2024, supporting more than 9,100 businesses.
  - Most recent VC funding has been channeled into AI businesses; the number of newly funded AI startups and private investment in AI startups has increased rapidly.
- Cross-border flows and the "missing middle":
  - In 2024 in the AI sector, UK recipients of VC financing raised only 40 percent of such financing domestically, while 80 percent of UK’s VC financing was channeled abroad, mostly to US startups.
  - Seed-stage companies attract almost three-quarters of funding domestically; at the scale-up stage domestic share falls to less than one-third.
  - Share of investment received from domestic sources: UK, EU, US examples summarized in source data (preserved as reported).
  - US and EU startups’ VC financing is more likely to come from domestic sources (70 and 50 percent respectively); US and EU VC funds are much less likely to be invested abroad (15 and 55 percent respectively).
- Consequences and frictions:
  - Financing constraints are among the top barriers to digital technology adoption; most binding for SMEs and young innovative ICT firms.
  - Three in four founders in the ICT sector identify access to growth capital as their primary obstacle.
  - Structural issues also include restrictive university spinout rules and lack of exit opportunities via listings.
  - Restructuring costs estimated at 19 months of salary per employee in the UK, compared with 7.4 months in the United States and under 3 months in Denmark and Switzerland—compounding investor reluctance at the growth stage.

### Regulatory framework: principles-based approach and uncertainty
- UK approach:
  - Since 2023, the UK has relied on existing laws and regulatory guidance based on five non-statutory cross-sector principles—safety, transparency, fairness, accountability, and contestability—implemented by existing regulators (FCA, ICO, CMA, Ofcom, and others).
  - This approach has avoided the compliance costs associated with the EU AI Act's tiered risk-classification system.
- Regulatory uncertainty:
  - Regulation features prominently as a barrier, especially for larger firms and in regulated sectors.
  - Copyright rules and ambiguity on permissible use of data for training and deploying models are salient issues.
  - Fragmentation across sectoral regulators creates an information friction that dampens adoption at scale.

### Skills: human capital strengths and specific shortages
- General human capital:
  - The UK workforce is highly educated and scores highly on the human capital dimension of the IMF AI Preparedness Index.
  - Strengths include high average years of schooling and good digital skills among the active population.
- Skill gaps and mismatches:
  - The UK ranks below average on the perceived skillset of graduates.
  - Prevalence of workers with AI engineering skills in key sectors (financial services, professional services, ICT) lags the US substantially.
  - Persistent shortages in intermediate and technical occupations, particularly in construction and digital roles.
  - As of 2025, more than 70 percent of UK companies surveyed by the British Chamber of Commerce report recruitment difficulties; ICT industry is the second most affected.
  - Recruitment difficulties most pronounced for professional and managerial positions (engineers, lawyers, finance professionals, HR managers).
  - One in three UK founders of Tech firms says that availability of talent is their biggest barrier to growth.
  - A study found the wage premium for AI-user skills is much higher in the UK than in the US.
- Supply risk:
  - Skill shortages have increased over the past decade and are at historically elevated levels.
  - Net migration inflows of cross-country workers with AI skills have declined sharply in recent years.

### Trade openness and resilience
- Digital trade strengths:
  - Digital trade represents more than half of the UK’s exports, twice the OECD and EU averages, with strong comparative advantage in digitally deliverable services, particularly financial and professional services.
  - Digital trade exports, including computer services, have grown at nearly three times the rate of other UK exports.
- Concentration risk:
  - The UK’s AI ecosystem depends heavily on US-headquartered hyperscalers for compute, cloud infrastructure, and foundation models, creating exposure to supply decisions and policy changes outside UK control.
  - The UK sits in Tier 1 of the US AI Diffusion framework and therefore faces no procurement restrictions on advanced GPUs.

### Policies for lifting bottlenecks: economy-wide vs targeted
- High-level proposition:
  - Both economy-wide policies (as part of the authorities' “Growth Mission”) and targeted AI interventions are needed.
  - Economy-wide reforms can address planning restrictions, elevated energy costs, and financing ecosystem issues.
  - AI-specific market failures (regulatory uncertainty, AI-specific skills shortages, concentration risk on frontier models) may require targeted interventions.
- Timing:
  - Competitor economies are investing heavily in AI infrastructure and skills; the window for shaping the UK’s competitive position is narrow, supporting near-term temporary interventions alongside longer-term reforms.

### Economy-wide policy actions and considerations
- Infrastructure: planning reform
  - 2025 Planning and Infrastructure Act aims to speed delivery of nationally significant infrastructure, including digital and energy infrastructure.
  - Datacenters were designated as critical national infrastructure in September 2024; the Act allows them to be opted into the Nationally Significant Infrastructure Projects regime.
  - AI Growth Zones under the AI Opportunities Action Plan offer fast-track pathways with pre-cleared planning approval and access to clean power.
  - Implementation caveats:
    - Delivery depends on scaling capacity of institutions (e.g., local planning authorities).
    - Stronger coordination between planning authorities, network operators, and investors is needed for sequencing consenting timelines.
    - Systematic monitoring and reporting of approval times and project completions is important.
- Infrastructure: energy and grid connection reforms
  - Ofgem reformed the grid connection regime in April 2025, replacing “first‑come‑first‑served” with “first ready and needed, first connected”; connections queue reduced substantially.
  - Clean Power 2030 Action Plan and establishment of Great British Energy on a statutory footing are intended to accelerate domestic clean generation and improve long‑term price predictability.
  - Government steps to expand nuclear capacity include final investment decision for Sizewell C.
  - Main challenge: speed of reforms vs rapid expansion of AI electricity demand; nuclear additions are 2030s-timed and cannot resolve near-term cost pressures.
  - Recommended focus: accelerate permitting of new renewables, connections reform, network buildout, and electricity market design.
- Innovation finance: mobilizing scale-up capital
  - Recent reforms:
    - Pension fund consolidation under the Mansion House Accord.
    - Listing and market reforms to lower frictions and improve exit opportunities.
    - British Business Bank’s expanded Growth Guarantee Scheme.
    - National Wealth Fund (NWF) with £27.8 billion capital over nine years.
    - AI-specific instruments: £500m Sovereign AI Unit and £100m advance market commitment for UK AI hardware startups.
  - Assessment of scale:
    - The Growth Guarantee Scheme remains several times smaller, relative to GDP, than comparable schemes in the United States and major European economies.
    - Reaching the scale of institutions such as Bpifrance or KfW would imply annual capital of roughly £21 billion.
  - Potential additional reforms:
    - Consolidate public support schemes and improve coordination to cover larger financing rounds.
    - Scale up public-finance institutions (BBB and NWF) over time to peer-comparable footprints while preserving co-investment discipline.
    - Address tax distortions: make R&D tax credits payable in cash for loss-making startups; reform capital gains tax treatment of capital losses and reinvestment to reduce lock-in effects.

### Targeted AI interventions and regulatory clarity
- Rationale for targeted interventions:
  - Market failures specific to AI (information frictions, skills shortages, concentration risk) justify AI-specific policies in addition to economy-wide reforms.
  - Near-term temporary interventions can be important while longer-term structural reforms take effect.
- Regulation: addressing information asymmetry and uncertainty
  - Problem:
    - AI applications can cut across several regulatory boundaries (e.g., employment law, data protection, equality legislation, sector-specific governance) with no single regulator clarifying permissibility.
    - This fragmentation creates an information friction that discourages early deployment and thus delays the evidence base needed for informed regulation.
  - Implication:
    - Providing clarity on liability, intellectual property, and sectoral expectations could reduce uncertainty and dampened adoption at scale.

*Source: IMF staff summary of "Box 1. United Kingdom: AI Adoption Over Time and Across Countries" (sipea2026072).*

### Box 2. United Kingdom: The Government’s AI Strategy

### Box 2. United Kingdom: The Government’s AI Strategy

### Strategic orientation and progress
- AI is a strategic priority in the UK's 10-year industrial strategy (DBT and DSIT, 2025) and one of three "Big Choices" highlighted by the Chancellor in March 2026.
- Core anchor: January 2025 AI Opportunities Action Plan, supplemented by the UK Compute Roadmap, the Digital & Technologies Sector Plan, the AI Growth zones program, and the AI for Science Strategy.
- Complementary to the Growth Mission (easing input constraints via higher public investment, planning reform, mobilizing patient capital, expanding labor supply).
- As of early 2026, the government reports having met 38 of the 50 actions in the Action Plan, with progress tracked via an online dashboard (DSIT 2026).

### Specific commitments across enabling conditions
- Skills
  - Committed £200m to upskill the workforce through AI courses.
  - Expanded the Growth and Skills Levy for employer-sponsored short AI and digital courses.
  - New AI scholarship programs including the £17.6m Spärck AI 100 scholarships.
  - Launched the £160m TechFirst program to support over 4,000 graduates and researchers.
  - Established a Global Talent Taskforce to attract top researchers.
- Infrastructure and compute
  - Plans to invest £1.75bn to build public compute capacity via the AI Research Resource program (in partnership with the University of Edinburgh).
  - Develop AI Growth Zones (AIGZs) for accelerated data center buildout with streamlined planning approval and access to clean power.
  - Additional investment of £100m for a National Data Library to house high-value public sector datasets and make them AI-ready.
- Innovation financing
  - A £500m Sovereign AI Unit set up to allocate access to public compute, provide early-stage financing alongside the British Business Bank, and finance AI infrastructure buildup.
  - An "advance market commitment" of £100m to act as a first customer for UK startups building AI hardware products.
- Regulation and safety
  - Reform of the text and data mining regime.
  - Data Use and Access Act (DUAA) passed in 2025.
  - Increased funding for regulators and £240m for the AI Security Institute to expand frontier model testing, foundational safety, and societal resilience.
- Public sector adoption
  - Government commitment to make the public sector a leader in AI adoption to support diffusion of AI tools across central government and public services.

### Regulatory uncertainty: current gaps and implications
- The DUAA simplifies compliance and enables responsible data sharing but leaves key questions open:
  - Application of UK copyright law to AI model training under review; DUAA required an impact assessment and report by March 2026 (DUAA, ss135–137).
  - DUAA introduces changes to the UK's data protection framework that diverge from EU norms; EU renewed the UK's data adequacy status to 2031, but longer-term implications for firms serving EU clients depend on the divergence.
- Industry reports demand greater clarity on regulation; some regulated sectors report regulators may lack resources to keep pace with AI developments, possibly slowing deployment.
- Policy options to enhance clarity and support adoption:
  - Provide cross-cutting, operational assurance tools demonstrating compliance with AI governance principles.
  - Develop concrete guidance iteratively with industry and stakeholders (model suggested: Singapore).
  - Accelerate resolution of outstanding legal questions on copyright, liability for AI-generated outputs, and conditions for AI-assisted decision-making in healthcare, financial services, and education.
  - Improve coordination and accountability across decentralized sector regulators; expand resources and technical capacity for regulators.
  - Maintain compatibility with EU and US regulatory frameworks where possible to reduce trade frictions and preserve data adequacy.

### Skills: market failures and policy focus
- Two reinforcing externalities constrain AI-specific skills supply:
  - Congestion: as more firms adopt AI, competition for specialists bids up wages, raising effective adoption costs and slowing further adoption.
  - Poaching: trained workers can be hired away by competitors, reducing firms' incentives to invest in training and slowing aggregate specialist supply growth.
- Current policy targeting and gaps
  - Short introductory courses aimed at a broad population are useful but may not suffice to build professional and managerial specialist capabilities at scale (AI engineers, data scientists, implementation specialists).
  - Programs to attract world-class researchers target a narrow cohort.
  - Recent Growth and Skills Levy reform expands employer-sponsored training but simultaneously defunded Master's-level apprenticeships for learners aged 22 and over.
  - Need for wider access to programs that deepen specialist skills at postgraduate and advanced levels.
- Complementary approaches recommended
  - Near-term: expand access to global AI talent via streamlined immigration paths for AI-relevant occupations (inflows have sharply declined).
  - Medium-term: structured employer co-investment in AI training to mitigate poaching externalities by better aligning private and social returns to training.

### Resilience: imports concentration and coordination failures
- Concentration of frontier AI development in a small number of foreign firms creates a speculative vulnerability if access is disrupted (export restrictions, supply chain bottlenecks, market power).
- Current UK position: access to foreign leading models and cloud infrastructure is broadly competitive, so the constraint is not yet binding.
- Government steps toward resilience:
  - Investments in sovereign AI capacity including the £1.75bn AI Research Resource and AI Growth Zones for accelerated data center buildout.
- Policy emphasis: build conditions for rapid domestic scaling if needed rather than directly replicating frontier capabilities:
  - Continue investing in compute infrastructure and energy capacity.
  - Diversify chip supply chains.
  - Ensure access to high-quality training data via the National Data Library.
  - Maintain regulatory compatibility with like-minded partners to reduce fragmentation risk.
- Modelling indicates modest investments in domestic resilience could recover a substantial share of output lost under disruption scenarios, and the insurance value of such investments significantly exceeds the government's current AI commitment of over £2 billion.

### Quantifying benefits: model framework and key assumptions
- Model: endogenous technology adoption calibrated to the UK economy (Rockall and others, 2025); occupations are task-based with varying AI exposure; firms adopt AI where AI capital is cheaper than labor for tasks.
- Baseline calibration anchors
  - Long-run scale of adoption: AI raises the aggregate capital share by approximately 10 percentage points in the long run (mirrors increase from routine automation 1980–2014).
  - Pace of adoption: AI reaches the long-run scale at twice the pace of routine automation—steady state reached in 17 years rather than 34.
  - Productivity per automated task: cost-saving gains from AI set to 27 percent (average of micro-empirical estimates).
- Under baseline assumptions:
  - Model produces cumulative TFP gains of 1.4 percent at the 10-year horizon.
  - Model projects cumulative output gains of approximately 3 percent within the first 5 years, rising to around 8 percent over a decade.
- Commoditization assumption and evidence:
  - Central assumption: cost of AI technologies will continue to fall rapidly over the next decade.
  - Evidence cited: cost of using AI models falling by roughly 90 percent per year; 99.7 percent decline in the cost efficiency frontier since March 2023 (Appenzeller 2024).
  - Open-source models now perform within 2 percent of frontier models on key benchmarks (Stanford HAI 2025).
  - Uncertainty: degree to which cost reductions pass through to end users depends on market structure, pricing strategies, and recouping of upfront investments.
  - Baseline assumes continued partial transmission of cost declines to users; final scenario relaxes this assumption.

### Policy scenarios and quantified impacts
- Focused policy experiments isolate AI-specific frictions while taking broader constraints as addressed.
- Regulatory uncertainty scenario
  - Calibration: regulatory uncertainty holds back 20 percent of potential AI adoption on average (consistent with share of executives reporting significant regulatory limitations).
  - Regulatory exposure index constructed from O*NET work context measures captures sectors most affected.
  - Reform modeled as resolving half of this constraint (clarifying permissible uses and strengthening legal protections), with the remaining half reflecting substantive safety and accountability requirements.
  - Impact: regulatory reform would boost output by 1.24 percentage points (additional TFP and adoption effects accrue over time as technology costs fall and regulatory uncertainty becomes a dominant constraint).

*Source: Box 2. United Kingdom: The Government’s AI Strategy (sipea2026072).*

### 0.22 percentage points within 5 years, growing to 1.72 and 0.35 percentage points respectively by

### sipea2026072 - 0.22 percentage points within 5 years, growing to 1.72 and 0.35 percentage points respectively by

### Productivity impacts of regulatory reforms
- Regulatory reform removes 50 percent of estimated task-level regulatory constraints on AI use, calibrated so that regulation holds back 20 percent of potential AI adoption.
- Productivity gains from regulatory reform (percentage points):
  - TFP: +0.22 at Year 5; +0.35 at Year 10.
  - Output: +1.24 at Year 5; +1.72 at Year 10.
- Policy interpretation:
  - As technology costs fall and adoption expands, regulatory uncertainty becomes the dominant barrier to adoption.
  - Targeted reforms resolving regulatory uncertainty are likely more effective over the medium and long run than cost-reducing interventions (e.g., adoption subsidies or tax breaks).
  - A policy that lowers adoption costs through subsidies delivering comparable near-term productivity gains would deliver substantially smaller gains at longer horizons because the value of cost-based interventions is eroded as AI matures.

### Skill shortages and externalities for AI talent
- Market failures identified:
  - Congestion externality and poaching externality in the market for AI specialists.
  - Without intervention, AI specialist supply is unlikely to expand fast enough, widening the gap between technological potential and realized adoption.
- Decomposition of adoption costs:
  - Adoption costs = technology component (falls over time) + labor component (rises over time).
  - AI labor share of adoption costs calibrated to 30 percent, consistent with estimates that 29–49 percent of frontier AI costs are attributable to labor.
  - AI wage premium currently estimated at 14 percent in the UK (PwC UK AI Jobs Barometer, 2024).
- Modeled congestion and labor dynamics:
  - Congestion markup on adoption costs grows from around 1.15× at year 5 to nearly 1.3× by year 10.
  - AI labor's share of total costs rises from around 39 percent to 46 percent between year 5 and year 10.
- Policy simulation: halving the AI wage premium through expanded training, vocational pathways, and immigration:
  - Output gains: +0.83 percentage points within 5 years; +1.58 percentage points by year 10.
  - TFP gains: +0.15 percentage points at Year 5; +0.31 percentage points at Year 10.
  - Implied increase in AI labor supply required: approximately 27 percent at year 5; 43 percent by year 10.
- Transitional considerations:
  - Analysis does not capture transitional labor market disruptions and potential displacement effects.
  - Work using same model framework suggests higher output will lead to higher wages overall but many workers may see lower wages due to displacement.
  - Wealth inequality likely to increase through higher capital returns; complementary redistributive policies are important.

### Comparison between regulatory reform and skills investment
- Both interventions deliver similar headline gains but operate through different channels and are complementary.
  - Regulatory reform: expands tasks for which firms are willing to deploy AI, unlocking productivity gains not achieved by lowering costs.
  - Skills investment: addresses a self-reinforcing bottleneck that grows over time (talent availability).
- Joint effect:
  - Pursued together, neither regulation nor talent availability becomes the dominant constraint.
  - Combined policies could jointly increase the output gains from AI by two thirds over the baseline.
- Table summary (cumulative gains relative to year 0; policy gains over baseline in percentage points and as share of baseline gains):
  - Baseline: TFP 0.53 (Year 5), 1.40 (Year 10); Output 3.26 (Year 5), 7.96 (Year 10).
  - Regulatory reform: TFP +0.22 (Year 5), +0.35 (Year 10); Output +1.24 (Year 5), +1.72 (Year 10).
  - Skills investment: TFP +0.15 (Year 5), +0.31 (Year 10); Output +0.83 (Year 5), +1.58 (Year 10).
  - Regulatory reform as percent of baseline: TFP +41% (Year 5), +25% (Year 10); Output +38% (Year 5), +22% (Year 10).
  - Skills investment as percent of baseline: TFP +28% (Year 5), +22% (Year 10); Output +25% (Year 5), +20% (Year 10).

### Concentration risk, access barriers, and the value of domestic AI capacity
- Risk scenario setup:
  - Restricted access to foreign frontier AI modeled as a markup on adoption costs.
  - Three-way firm choice modeled: do not adopt; adopt a home-grown AI technology (not subject to disruption cost but less effective); adopt foreign frontier model at higher cost.
  - Assumed domestic alternative delivers 90 percent of the benefits of the frontier model conditional on adoption.
- Disruption markups considered: τ = 15 percent; τ = 25 percent; τ = 40 percent.
- Central scenario (τ = 25 percent) impacts at Year 5:
  - Baseline TFP gain of 0.53pp reduced to 0.08pp under restricted access.
  - Baseline output gain of 3.26pp reduced to 0.54pp under restricted access.
  - Equivalent to approximately £82 billion per year in lost output relative to the baseline.
  - A domestic alternative recovers roughly half of this loss, worth approximately £40 billion per year.
- Severe disruption (τ = 40 percent):
  - Near-term productivity gains from AI are effectively lost under restriction, but domestic option recovers approximately £57 billion per year.
- Policy implications:
  - Investment in UK domestic AI capacity acts as insurance against supply-side disruptions.
  - Strengthening ecosystem requires enabling conditions: data center capacity and energy infrastructure, diversified chip supply chains, subsidized compute for model development, facilitated access to training data.
  - Current government AI investment commitment is over £2 billion; this is modest relative to potential benefits and the output recovered under moderate disruption.
  - Policy focus should be on enabling conditions rather than attempting to replicate frontier capabilities.

### Conclusion and policy recommendations
- Context:
  - The UK's growth performance has weakened markedly over the past two decades; slowdown driven primarily by sluggish TFP growth.
  - UK economy and workforce concentrated in sectors and occupations where generative AI shows largest productivity gains and cost savings.
- Five enabling conditions that need to be addressed: skills, financing for innovation, regulation, infrastructure and energy, and trade openness.
- Priority economy-wide reforms under the authorities' Growth Mission:
  - Ensure recent planning reforms translate into materially shorter approval timelines for critical infrastructure.
  - Continue reforms in renewables permitting, grid connection processes, and electricity market design to narrow the UK's energy cost gap with peers.
  - Mobilize domestic capital for high-growth, innovation-intensive firms, including better coordination and consolidation of public support schemes and fiscal incentives tailored to financially constrained startups.
- Targeted AI-specific interventions:
  - Expand supply of AI-relevant skills: streamline immigration pathways in near term; ramp up education and training programs.
  - Provide regulatory clarity on copyright issues and facilitate AI use in regulated sectors.
  - Invest in domestic AI ecosystem foundations to insure against supply disruptions.
- Equitable and responsible deployment:
  - Complement supply-side interventions with redistributive policies to address displacement and inequality.
  - Government role in ensuring broad sharing of gains and responsible deployment.
- Central challenge: delivery, scale, and agile course-correction to translate strategy into implementation at the pace and ambition required.

### Annex I — Lessons from the ICT revolution (summary)
- General findings:
  - General-purpose technologies lift growth by diffusing across sectors and require complementary investment in intangible capital (organizational change, new business processes, reorganization of tasks and workers).
  - ICT emergence in the 1990s contributed significantly to labor productivity growth; ICT-related gains and capital accumulation explained about 30–50 percent of aggregate labor productivity growth in the 1990s and early 2000s.
  - Productivity gains can materialize long after the technological breakthrough; effective use of AI may be felt only now despite earlier inventions.
  - Most ICT-driven productivity gains came from adoption in ICT-using sectors (market services) rather than from ICT sectors themselves.
  - Faster ICT-led growth in the US versus Europe explained by larger ICT sectors and more effective adoption in ICT-using sectors, not by higher investment rates.
- UK experience:
  - Over 1995–2006, the UK’s growth gains differed from European peers and approached those in the US.
  - Main difference in overall growth rates between US and UK estimated to come from capital deepening.
  - UK TFP grew almost as rapidly as in the US during the adoption phase of ICT, and faster than most other European countries, though some studies find UK TFP fell somewhat short of the US in earlier ICT phases due to adoption lags and frictions.
- Enablers of effective ICT adoption:
  - Flexible markets, complementary investments in business practices, high-skilled workforce, openness to trade and ideas.
  - Proximity to the innovation frontier, trade and capital openness, a significant share of high-skilled workers, and supportive institutions facilitate diffusion and effective adoption.

*Source: International Monetary Fund — "UNITED KINGDOM" selected issues paper content (sipea2026072).*

### Annex II. Drivers of Business Adoption of AI Across Countries

### Annex II. Drivers of Business Adoption of AI Across Countries

### Summary of analysis
- Purpose: To identify structural drivers of AI adoption rates across countries using cross-section regressions of AI adoption rate mid-2025 on potential enabling conditions and controls.
- Key enabling conditions studied: ICT infrastructure, human capital stock (from the AI preparedness index in Cazzaniga and others, 2024), trade openness (exports plus imports over GDP, World Bank WDI), electricity prices (UK Department for Energy Security & Net Zero derived from IEA Energy Prices and Taxes), log GDP per capita, and the World Uncertainty Index.
- Timing choices to avoid feedback: 2019 values used for electricity variables, GDP per capita, openness, and the uncertainty index; human capital and infrastructure indices based on year 2023 or before.

### Regression specification
- Structural cross-section regression:
  y_i = x_i β + z_i γ + e_i
  - y_i: AI adoption rate mid-2025 in country i.
  - x_i: potential drivers (ICT infrastructure, human capital, trade openness, electricity prices).
  - z_i: controls (log GDP per capita, World Uncertainty Index) in some specifications.
- Estimation focus: fundamental, structural drivers (not transitory deviations).
- Note: All variables are standardized except for GDP per capita.

### Key regression results (Table II.1)
- Estimation sample sizes and fit:
  - Observations: 30 in columns (1)–(3); 26 in columns (4)–(5).
  - R-squared: 0.759 (1), 0.764 (2), 0.788 (3), 0.869 (4), 0.869 (5).
- Coefficient estimates (standard errors in parentheses; significance: * p<0.10, ** p<0.05, *** p<0.01):
  - Human capital:
    - (1): 5.69*** (1.54)
    - (2): 4.80** (1.95)
    - (3): 5.25*** (1.49)
    - (4): 5.15*** (1.78)
    - (5): 5.16** (1.83)
  - ICT infrastructure:
    - (1): 7.30*** (2.35)
    - (2): 6.59** (2.54)
    - (3): 7.00*** (2.25)
    - (4): 8.69*** (2.46)
    - (5): 8.64*** (2.56)
  - Openness:
    - (1): 1.86** (0.86)
    - (2): 1.49 (0.99)
    - (3): 2.05** (0.83)
    - (4): 2.33** (1.03)
    - (5): 2.31** (1.08)
  - Log GDP per capita:
    - (2): 2.33 (3.07)
    - (3): 0.98 (3.37)
    - (5): 1.08 (3.58)
  - Log electricity price:
    - (3): -2.77* (1.49)
    - (5): 0.19 (1.73)
  - Uncertainty (World Uncertainty Index):
    - (4): -2.01** (0.85)
    - (5): -2.06* (0.99)
  - Constant terms:
    - (1): 32.64*** (1.23)
    - (2): 9.16 (30.98)
    - (3): 32.67*** (1.18)
    - (4): 24.15 (33.83)
    - (5): 23.16 (35.85)

### Empirical findings and interpretations
- Robust positive associations:
  - Higher human capital, stronger ICT infrastructure, and greater trade openness are associated with higher AI adoption rates across countries (columns (1) and (2)).
- Energy prices:
  - Higher log electricity prices are associated with significantly lower AI adoption in column (3) (coefficient -2.77*, standard error 1.49), suggesting energy costs can deter deployment.
- Uncertainty:
  - Higher country-level uncertainty is significantly associated with lower adoption (column (4): -2.01**, (0.85)), consistent with uncertainty deterring investment in new technology.
- Robustness:
  - Results remain largely unchanged and mostly significant after controlling for log GDP per capita (column (2)) and when including all variables jointly (column (5)).

### Variance decomposition of adoption rates
- Method: Squared semi-partial correlations computed from the specification in column (1) to measure the unique contribution of each predictor to total explained variance (R²).
- Main quantitative ranking (as reported in text narrative):
  - AI-relevant skills (human capital) are the most important unique driver of cross-country variation in AI adoption.
  - ICT infrastructure is the next most important unique driver.
  - Trade openness follows.
  - The residual between R² and the sum of unique contributions captures variance jointly explained by correlated predictors and is not attributed to any single driver.
- Implication: Policy attention to skills, infrastructure, and openness addresses the largest uniquely attributable components of cross-country differences in AI adoption.

### Additional notes
- Use of pre-adoption period covariates (2019 for some variables) is intended to avoid capturing reverse-causation where AI adoption affects drivers like electricity demand.
- The analysis complements the broader discussion in Section C and the third panel of Figure 3 on enabling conditions for AI adoption.

*Source: Annex II. Drivers of Business Adoption of AI Across Countries, sipea2026072.*

### 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."

### sipea2026072 - 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."

### Overview of referenced literature and reports
- Citations focus on generative AI, AI adoption, productivity impacts, technology diffusion, and policy responses across academic, policy, and institutional sources.
- Material spans academic journals, IMF working papers and selected issues papers, government policy papers, think-tank reports, and industry analyses.
- Temporal coverage in the references includes works dated from 2001 through 2026, reflecting both foundational technology-economics literature and recent AI-specific analyses.

### Major thematic clusters in the references
- Generative AI and productivity effects
  - Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work." Quarterly Journal of Economics, 140(2), 889–942.
  - Noy, S. and Zhang, W., 2023. "Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence." Science, 381, 187–192.
  - Peng, S., Kalliamvakou, E., Cihon, P. and Demirer, M., 2023. "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot." arXiv preprint arXiv:2302.06590.
  - Misch, F., Park, B., Pizzinelli, C. and Sher, G., 2025. "AI and Productivity in Europe." IMF Working Paper 2025/067.
- Macroeconomic and distributional implications of AI
  - Cerutti, E.M., Pascual, A.I.G., Kido, Y., Li, L., Melina, G., Tavares, M.M. and Wingender, P., 2025. "The Global Impact of AI: Mind the Gap." IMF Working Paper 2025/076.
  - Filippucci, F., Gal, P. and Schief, M., 2024. "Miracle or Myth? Assessing the Macroeconomic Productivity Gains from Artificial Intelligence." OECD Artificial Intelligence Papers No. 29.
  - Filippucci, F., Gal, P., Laengle, K. and Schief, M., 2025. "Macroeconomic Productivity Gains from Artificial Intelligence in G7 Economies." OECD Artificial Intelligence Papers No. 41.
  - Moll, B., Rachel, Ł. and Restrepo, P., 2022. "Uneven Growth: Automation's Impact on Income and Wealth Inequality." Econometrica, 90(6), 2645–2683.
  - Rockall, E.J., Tavares, M.M. and Pizzinelli, C., 2025. "AI Adoption and Inequality." IMF Working Paper 2025/068.
- Technology diffusion and historical productivity literature
  - Comin, D. and Hobijn, B., 2004. "Cross-Country Technology Adoption: Making the Theories Face the Facts." Journal of Monetary Economics, 51(1), 39–83.
  - Comin, D. and Hobijn, B., 2010. "An Exploration of Technology Diffusion." American Economic Review, 100(5), 2031–2059.
  - Jorgenson, D.W., 2001. "Information Technology and the U.S. Economy." American Economic Review, 91(1), 1–32.
  - van Ark, B., Inklaar, R. and McGuckin, R.H., 2003. "ICT and Productivity in Europe and the United States: Where Do the Differences Come From?" CESifo Economic Studies, 49(3), 295–318.

### United Kingdom–focused policy, data, and analysis sources
- Government strategy, policy papers, and surveys
  - Department for Business and Trade and Department for Science, Innovation and Technology (DBT and DSIT), 2025. Digital and Technologies Sector Plan. Policy Paper.
  - Department for Science, Innovation and Technology (DSIT), 2025. Artificial Intelligence Sector Study 2024.
  - Department for Science, Innovation and Technology (DSIT), 2026. AI Opportunities Action Plan: One Year On. Policy Paper.
  - Department for Science, Innovation and Technology and UK Research and Innovation (DSIT and UKRI), 2025. UK Compute Roadmap.
  - Department for Education (DfE), 2025. Skills England: Sector Evidence on the Growth and Skills Offer.
  - Ministry of Housing, Communities & Local Government (MHCLG), 2025. Local Authority Planning Capacity and Capability Survey 2025.
- Industry, think-tank, and survey evidence on UK AI adoption and growth
  - Confederation of British Industry (CBI), 2025. "AI Adoption among UK Firms." Unpublished CBI presentation.
  - Microsoft and Public First, 2025. Unlocking the UK's AI Potential: Harnessing AI for Economic Growth.
  - TechNation, 2025. Unlocking the UK's Growth Potential: The Tech Nation Report 2025.
  - Resolution Foundation, 2025. Mountain Climbing: Making Progress on the UK's Growth Policy Challenge.
  - Tony Blair Institute (TBI) for Global Change, 2025. From Startup to Scaleup: Turning UK Innovation into Prosperity and Power.
  - NewMind AI, 2025. "AI Policy and Regulations of United Kingdom." NewMind AI Journal Country Report.
  - Stanford Institute for Human-Centered Artificial Intelligence (HAI), 2025. Artificial Intelligence Index Report 2025.
  - Microsoft and Public First, 2025. Unlocking the UK's AI Potential: Harnessing AI for Economic Growth.
- UK research on startup, scale-up, and productivity dynamics
  - Nguyen, N. and Coyle, D., 2025. "Start-ups and Scale-ups in UK Technology Sectors." Bennett School for Public Policy, University of Cambridge.
  - Indraccolo, L., 2025. "Bridging the Gap: Understanding the UK-US Productivity Decoupling." IMF Selected Issues Paper SIP/2025/112.
  - Martin, J., 2025. "The UK Productivity Slowdown: A Review of Timing, Magnitude, and Drivers." International Productivity Monitor, 48 (Spring), 29–62.
  - McLean, A. and Smith, D., 2025. Technology Adoption Review 2025.
  - Willemyns, J., 2025. “Copyright & AI: the case for a pro-growth approach.” Centre for British Progress.

### Data sources, indices, and empirical contributions cited
- IMF and OECD datasets and analyses
  - International Monetary Fund (IMF), 2024. "Slowdown in Global Medium-Term Growth: What Will It Take to Turn the Tide?" Chapter 3 in World Economic Outlook: Steady but Slow — Resilience amid Divergence, April 2024.
  - OECD, 2025. Taxing Wages 2025: Decomposition of Personal Income Taxes and the Role of Tax Reliefs.
  - OECD, 2026. "OECD Overall R&D Growth Stable; Government R&D Budgets Decline and Reorient towards Defence." OECD Statistical Release, March 2026.
- Indexes and working paper series
  - Stanford HAI, 2025. Artificial Intelligence Index Report 2025.
  - Yotzov, I., Barrero, J.M., Bloom, N., Davis, S.J. et al., 2026. "Firm Data on AI." NBER Working Paper No. 34836.
  - Jaumotte, F., Kim, J., Koll, D., Li, E., Li, L., Melina, G., Song, A. and Tavares, M.M., 2026. "Bridging Skill Gaps for the Future: New Jobs Creation in the AI Age." IMF Staff Discussion Note SDN/2026/001.

### Cross-cutting methodological and conceptual references
- Technology adoption and trade theory
  - Eaton, J. and Kortum, S., 2002. "Technology, Geography, and Trade." Econometrica, 70(5), 1741–1779.
  - Vandenbussche, J., Aghion, P. and Meghir, C., 2006. "Growth, Distance to Frontier and Composition of Human Capital." Journal of Economic Growth, 11(2), 97–127.
- Historical perspectives on ICT and productivity
  - Jorgenson, D.W., Ho, M.S. and Stiroh, K.J., 2008. "A Retrospective Look at the U.S. Productivity Growth Resurgence." Journal of Economic Perspectives, 22(1), 3–24.
  - Oliner, S.D., Sichel, D.E. and Stiroh, K.J., 2008. "Explaining a Productive Decade." Brookings Papers on Economic Activity, 2007(1), 81–152.

*Compiled from the reference list in sipea2026072 - 31161. Published as Brynjolfsson, E., Li, D. and Raymond, L., 2025. "Generative AI at Work."*

---


_Source: https://www.imf.org/-/media/files/publications/selected-issues-papers/2026/english/sipea2026072.pdf_