Artificial intelligence (AI) compute is a strategic asset. AI compute refers to physical hardware and software infrastructure critical for developing, scaling and accessing AI systems. Compute advances are expected to enable future AI breakthroughs, boosting economic productivity and transforming sectors.
Policy approaches to securing AI compute vary across countries, reflecting differing economic, geographic, technological and political contexts. Efforts to secure domestic AI compute are motivated by potential economic benefits, enabling customised and local AI deployment, bolstering supply chain security and lowering latency for AI applications. Policies to diversify, own and localise AI compute can reinforce economic security and resilience, safeguarding key inputs across the AI stack.
Governments are embedding AI compute commitments in national and regional AI strategies, but few have standalone AI compute strategies. Forty-four of the 46 member countries of the Global Partnership on AI (GPAI) and the European Union have formally adopted national or regional AI strategies, while four countries are currently updating or developing such strategies. Twenty-one jurisdictions treat compute as a dedicated pillar within such strategies, 10 include compute as an explicit AI strategy commitment, 13 reference AI compute broadly and three currently do not have AI strategies. Notably, Canada and the United Kingdom are the first to have dedicated national AI compute strategies distinct from their national AI strategies.
Targeted domestic AI compute plans should be informed by local needs assessments (OECD, 2023[1]). There is limited systematic understanding of effective policy levers to advance domestic AI compute and an overall lack of comparable data to help guide government actions. The OECD, including through co-operation with GPAI, can support countries with their national AI compute strategies, providing analytical frameworks, data and the sharing of good practices between its members and partners.
Securing AI compute
AI compute is an increasingly important economic and strategic asset for countries and regions around the world (OECD, 2023[1]). This infrastructure typically takes the form of AI accelerator chips deployed in data centres, coupled with suitable software tooling that enables the training and inference of AI systems. Governments are looking to secure this strategic asset, and associated capacities, within their borders. However, standardised approaches for assessing domestic AI compute and guiding investments are still not widely used. The OECD (2023[1]) Blueprint for Building National Capacity for Artificial Intelligence (hereafter ‘the Blueprint’) puts forward a framework for standardising such assessments and embedding AI compute considerations in national AI strategies, which has helped to inform AI compute reviews in the handful of jurisdictions that have completed such needs assessments today.
This policy brief examines government policy approaches to securing AI compute. It is structured according to the five recommendations to policymakers embedded in the OECD AI Principles, providing a framework for comparing government initiatives, including investment approaches, around the globe (OECD, 2024[2]). The Blueprint guides policymakers in assessing domestic AI compute needs by considering AI compute capacity, its effectiveness, and its resilience (OECD, 2023[1]). This work contributes to building comparable measures across these elements and leverages compute-specific data in the OECD.AI Policy Navigator.
AI compute resources are primarily stored within data centres, referring to purpose-built industrial facilities that house the core software and hardware inputs of compute technology such as server racks containing AI accelerators (i.e. AI chips/semiconductors), memory, storage and networking interfaces. Data centres and their specialised hardware and software stacks can be owned and operated by private firms, government or academia, or leased to third parties as “co-location” facilities to host AI compute. This infrastructure can be located domestically or abroad. For many AI model users and developers, AI compute is often accessed via cloud computing services offered by cloud providers through on-demand pricing models. Although compute can be used for both AI and non-AI workloads, much of recent compute-related investment is driven by private and public interest in bolstering AI-specific compute capacity especially within national borders. While AI-related productivity impacts also depend on effective AI adoption, compute remains a critical AI-enabling input in addition to data, skills and algorithms (OECD, 2023[1]).
The terms domestic, national and sovereign AI compute capacity are increasingly used to describe efforts to secure, access and build national/regional AI compute capacity. Domestic compute can include government-funded compute facilities (owned by government bodies that invest in supercomputers and accelerators), private compute clusters (owned and used by for-profit firms), and public cloud providers also understood as “compute for purchase” (owned and sold on-demand to users typically by hyperscalers and more recently neoclouds) (Lehdonvirta, Wu and Hawkins, 2024[3]; OECD, Forthcoming[4]). This report uses a framework based on the main layers of the AI stack (see Table 1) (OECD, 2026[5]), providing a structured lens through which the AI ecosystem can be analysed.
Compute is an important factor shaping a country's ability to develop specialised AI capabilities, harness advances in AI for breakthrough discovery and innovation, and accelerate trusted adoption of AI applications across the economy, and to participate in international discussions around AI compute. Enabling appropriate access, boosting domestic AI compute capacity and pursuing the most resilient solutions will depend on a country’s context and needs. A country’s unique economic, political, geographic and technological context impacts the relative strategic value of investing in domestic AI compute. While some may benefit from strategic partnerships that provide sufficient access to compute based in other countries (e.g. via regional and international voluntary compute-sharing programmes on mutually agreed terms), others may require investment within their borders to both build and operate AI compute. As governments consider sizable public investments, establishing clear national AI compute policy objectives helps to enable accountability in public-sector spending and leverage private-sector investments where possible and desirable.
Government initiatives on AI compute can advance a range of policy objectives including securing capacity, its effectiveness, human capital (e.g. skills, training), an enabling policy environment (e.g. appropriate laws), innovation (e.g. AI compute R&D) or initiatives to improve access (e.g. cost, usage rights). Resilience is another factor driving investments, including security and sovereignty considerations (e.g. AI compute location, ownership, supply chain security) and AI compute efficiency considerations (OECD, 2023[1]). This policy brief examines government policy initiatives across these areas, aligned with the OECD Recommendation on AI’s five actions for policymakers (OECD, 2024[2]).
Government initiatives in this regard can broadly be categorised into the policy instruments outlined in Table 2. These include the direct provision of AI compute, such as state ownership of AI compute capacity through a specialised public agency, legal actions such as domestic laws, market-based tools such as grants, tax incentives or facilitated permitting, or information-based instruments such as public strategies or needs assessments.
Note: This is a sample of tools; it is not exhaustive. Policy categories and instruments can be used in tandem and thus categories may not be discrete in practice, e.g. an AI Compute Strategy (“information-based”) including the creation of a new public supercomputer (“direct provision”).
Compute capacity is critical as a foundational input to undertaking AI research and development (R&D), including fundamental and applied research across the AI technology stack, development of open-source models and novel AI applications, and AI-enabled discovery and innovations across scientific domains. Conversely, AI R&D can also contribute to innovations in AI compute itself, for example research around new accelerator architectures or heat management systems.
Access to AI compute risks becoming a bottleneck on advanced and applied AI R&D. Reasons include increased demand for AI compute spurred by mass consumer adoption of AI systems and rapidly increasing training and inference costs. In the last five years, AI compute has grown in strategic relevance due to its dual-use nature and its role in economic competitiveness, economic security and scientific discovery, motivating recent government initiatives to secure compute capacity for AI R&D.
Governments are acting to secure AI compute for AI research and development
Governments across GPAI member countries have increasingly proposed and implemented initiatives to secure access to AI compute for AI R&D, reflecting possible concerns that market-led or private sector provided compute alone may not ensure sufficient or reasonable (e.g. cost effective) access. A prominent approach has been direct public provision of AI compute or co-ordination of large-scale compute infrastructure through national laboratories, supercomputers, or publicly funded computing facilities, including regional partnerships.
Governments are using subsidised access and allocation mechanisms to ensure that academic researchers, start-ups and small and medium-sized enterprises (SMEs) can access AI compute at reasonable costs. Launched in 2025, Canada’s Sovereign AI Compute Strategy includes the creation of the AI Sovereign Compute Infrastructure (SCIP) Program, a “Canadian-owned and located supercomputing system” supporting researchers and a cross-section of industry (ISED, 2025[8]). The SCIP programme aims to provide about CAD 890 million for design, construction and ongoing operation of the compute system, beginning in 2026-2027 (ISED, 2026[9]).
Germany’s Gauss Centre for Supercomputing has expanded its high-performance computing (HPC) infrastructure toward enabling AI applications for the academic research community, large companies and SMEs. They provide training for users on how best to access and use their facilities to accelerate AI research, with planned further expansion of new AI compute (Gauss Centre for Supercomputing, n.d.[10]). Türkiye’s HPC backbone is centred on the Scientific and Technological Research Council of Türkiye (TÜBİTAK) National Academic Network and Information Center’s National Science e-Infrastructure (TRUBA), which serves over 6 000 researchers in over 180 institutions countrywide. Furthermore, under Law No. 7545 on Cybersecurity, the Presidency of Cybersecurity of Türkiye is empowered to provide secure hosting infrastructure to public bodies and to identify and protect critical infrastructure, linking compute provision to compute security.
The United Kingdom’s AI Research Resource (AIRR) aims to address the shortage of publicly available compute resources, offering compute access through various academic institutions, providing specialised compute for SMEs and the AI R&D community. Applications to the programme offer “up to 10 000 GPU hours on AIRR services” (DSIT, 2025[11]). In the United States, America’s AI Action Plan acknowledges the need to improve the financial market for compute, ensuring adequate access for academics and start-ups. Amongst other proposed actions, the Action Plan recommends expanding the National AI Research Resource (NAIRR) pilot project to provide “world-class private sector computing” to the research community (White House, 2025[12]).
The European Union’s AI in Science Strategy provides EUR 600 million from Horizon Europe to expand access to compute for science applications, securing dedicated access to AI Gigafactories for European researchers and start-ups (European Commission, 2025[13]). The Strategy uses capacity from Europe’s AI Factories, an initiative leveraging the supercomputing capacity of EuroHPC Joint Undertaking (EuroHPC JU) to pool compute resources across member states and provide compute access that may otherwise be prohibitively costly for individual researchers or institutions (European Commission, n.d.[14]).
Governments also support R&D for AI compute innovation along the AI value chain
Beyond expanding access to AI compute for R&D purposes, governments are also supporting R&D for AI compute innovation itself along the AI value chain. This conveys a recognition of existing bottlenecks in AI compute, including technical limitations in memory, interconnection – the movement of data across the AI stack – power and heat management. Overcoming these limitations will hinge on continued innovation across the AI stack, and particularly for its compute layers, contributing to advances in system-level performance (OECD, Forthcoming[4]). By investing in R&D across the compute value chain, governments seek to reduce long‑term cost trajectories, alleviate future AI compute capacity constraints and reinforce domestic knowledge bases, thereby extending the scope of Principle 2.1 from access to compute for AI R&D toward the innovation of compute technologies themselves.
Support for semiconductor research and manufacturing represents a central example. In March 2026, Korea is reported to have committed about KRW 8.6 trillion to accelerate development in priority areas like semiconductors and AI, mobilising 23 ministries to co-ordinate research funding, policy tools and financial support, including increasing national strategic technology R&D by 30%. The fund aims to promote R&D commercialisation, supporting researchers to launch start-ups and engage in procurement processes (Gyu-lee, 2026[15]).
Governments have also focused on upstream and complementary innovations, including software optimisation, AI compute system integration and energy‑efficient compute. Efforts are often embedded in broader national AI or digital strategies, ensuring alignment between compute innovation and downstream AI applications. The United States NAIRR pilot provides federally funded researchers with access not only to compute hardware but also data, models and tools that equip scientists to tackle national science and technology challenges.
Governments are taking direct action to foster an enabling domestic ecosystem through which AI compute is built, accessed and used. Such initiatives include mobilising public and private capital for compute infrastructure, procuring AI compute capacity through domestic providers, and using subsidies and fiscal incentives to encourage private investment across compute infrastructure and the underlying semiconductor supply chain.
Governments are mobilising capital for AI compute at unprecedented scale
Governments have significantly expanded direct investment in national AI compute capacity and national AI strategies are emerging as a way to mobilise substantial private-sector capital for domestic compute infrastructure. Australia's National AI Capability Plan positions the country as a leading data centre investment destination in the Indo-Pacific, with private-sector commitments announced between 2023 and 2025 reaching over USD 100 billion. Under its forthcoming 2026–2030 AI Action Plan, Brazil's National AI Infrastructure Program commits approximately USD 1 billion to overhaul the country's compute infrastructure through a six-pillar strategy, which includes the acquisition of a national AI supercomputer and shared access arrangements with other countries in Latin America, the Caribbean and Africa (MCTI, 2025[16]).
In March 2024 under the France 2030 Plan, France awarded approximately EUR 40 million to extend the Jean Zay supercomputer's infrastructure, increasing GPU capacity to meet growing AI research demands (CNRS, 2024[17]). France was selected in March 2025 as one of the EuroHPC JU AI Factories, with AI Factory France (AI2F) receiving a budget of EUR 30 million over three years (Garreau, 2025[18]). Japan has similarly increased public investment in AI supercomputing and has committed USD 1 billion through the Ministry of Economy, Trade and Industry (METI) to expand national computing resources (Harris, n.d.[19]). A key component is the AI Bridging Cloud Infrastructure (ABCI) 3.0 supercomputer, operated by the National Institute of Advanced Industrial Science and Technology (AIST). ABCI launched in August 2018 and upgraded to ABCI 3.0 in January 2025, providing large-scale open AI infrastructure for research and industry collaboration (Harris, n.d.[19]; ABCI, 2025[20]).
Governments are also turning to public-private partnerships to mobilise capital for national AI compute infrastructure. Denmark launched the Gefion supercomputer in 2024, through a partnership between the Novo Nordisk Foundation and the Export and Investment Fund of Denmark (EIFO) (Novo Nordisk Foundation, 2024[21]). The two organisations financed the Danish Centre for AI Innovation A/S, which owns and operates Gefion. The Novo Nordisk Foundation committed approximately DKK 600 million to the centre’s initial costs, while EIFO contributed DKK 100 million (Novo Nordisk Foundation, 2024[21]). Türkiye aims to reach at least 1 gigawatt of installed data centre capacity by 2030 and mobilise at least USD 10 billion, mainly from the private sector.
The largest recent commitments are being channelled through new strategic investment vehicles designed to mobilise both public and private capital at unprecedented scale. Saudi Arabia launched HUMAIN in May 2025 as a flagship AI company owned by the Public Investment Fund (PIF) (PIF, 2025[22]). Backed by PIF's USD 900 billion in assets, HUMAIN operates as the Kingdom's national platform for AI infrastructure development, covering data centres, cloud services and AI model development. The PIF has allocated over USD 40 billion to AI ventures, with HUMAIN planning major investments in large-scale data centre campuses and capacity expansion through 2030, including partnerships with notable chip manufacturers and cloud hyperscalers (Soliman, 2025[23]).
In the United States, Stargate was announced in January 2025 as a joint venture between OpenAI, SoftBank and Oracle committing USD 100 billion initially, with plans to scale investments to USD 500 billion by 2029 (OpenAI, 2025[24]). Announced by the White House, this privately-led initiative aims to build the world's largest AI supercomputing infrastructure. In September 2025, participants announced five US AI data centre sites under Stargate, bringing planned capacity to nearly 7 gigawatts and over USD 400 billion in cumulative investment (OpenAI, 2025[25]).
Together, these initiatives illustrate how AI compute is increasingly being treated as strategic national infrastructure, requiring sustained, large-scale capital mobilisation. While announced commitments already amount to significant investment globally, their long-term impact will depend on how effectively they translate into operational infrastructure, AI capabilities and AI adoption over time. Execution capacity, semiconductor supply constraints, energy generation and evolving policy priorities will shape whether these investments achieve their intended objectives.
Governments are also procuring AI compute hardware and software at scale
Alongside direct investment in national AI infrastructure, governments are increasingly using public procurement to secure access to AI compute hardware, cloud services and supporting software. Procurement strategies allow governments to rapidly expand domestic compute availability while supporting the development of domestic cloud and AI ecosystems.
India launched the IndiaAI Mission in March 2024, under which the IndiaAI Compute Pillar aims to deploy more than 18 000 GPUs through public-private partnerships to support AI start-ups, researchers and public-interest applications (IndiaAI, n.d.[26]). Under Korea’s National AI GPU Infrastructure Initiative, the Korean Ministry of Science and ICT (MSIT) announced in July 2025 a KRW 1.46 trillion government investment to procure 13 000 GPUs (Park, 2025[27]). The procurement is distributed across three domestic cloud providers – Naver Cloud, NHN Cloud and Kakao – reflecting an effort to strengthen local AI compute capacity through domestic providers (OECD.AI Policy Navigator, 2026[28]).
Governments are using direct subsidies and grant programmes to encourage private investment in AI compute and the underlying semiconductor supply chain
Governments are also using direct subsidies and grant programmes to stimulate private investment in AI compute infrastructure and its enabling systems, including data centres, compute clusters, energy supply and cooling infrastructure. Canada's AI Compute Challenge, part of the CAD 2 billion Sovereign AI Compute Strategy (December 2024), commits CAD 700 million in investments to support private companies building AI data centre infrastructure in Canada (ISED, 2025[29]). Japan has allocated JPY 72.5 billion through METI to fund private sector cloud and AI compute capacity expansion projects, under the Economic Security Promotion Act. The initiative designates cloud programmes as specified critical products, aiming to strengthen domestic AI infrastructure, reduce reliance on foreign cloud providers and ensure resilient generative AI service provision in Japan (METI, 2024[30]).
Governments are shaping an enabling governance and policy environment for AI compute through a range of initiatives. These include embedding compute commitments in national AI strategies, adopting dedicated compute strategies with ringfenced budgets or commissioning national reviews and needs assessments to shape future compute policy. Countries are also delivering compute capability through regional frameworks and major national programmes, funding institutions to host national compute resources and through reformed permitting and spatial planning to enable data centre construction.
Governments are increasingly embedding AI compute commitments in national/regional AI strategies
Governments are increasingly treating AI compute as a distinct policy priority. Where AI compute was often included within non-AI-specific programmes such as HPC initiatives, governments are now adopting national AI strategies with substantive compute commitments and, in a few cases, dedicated AI compute strategies. The OECD's 2023 Blueprint calls on governments to develop national AI compute plans, and a growing number of governments have done so (OECD, 2023[1]).
Analysis of formally adopted government policy documents across 46 GPAI member countries and the European Union shows that 44 have a formally adopted national or regional AI strategy, of which four are being updated. The treatment of AI compute within these strategies varies considerably (see Figure 1). To capture this variation, the analysis applies a four-category classification: strategies where compute is addressed as a dedicated pillar with named instruments, capacity targets or ringfenced budget (Tier 1); strategies where compute appears as an explicit commitment in the form of a transversal enabler (Tier 2); strategies where compute is referenced at the level of principles or generic actions (Tier 3); and the absence of a national AI strategy (Tier 4). Twenty-one GPAI member countries treat compute as a dedicated pillar (Tier 1), 10 as an explicit commitment (Tier 2), 13 at the level of references and generic actions (Tier 3), and three members have no national AI strategy (see Figure 1).
Governments adopting or updating their national AI strategies in 2025–2026 are increasingly treating compute as a dedicated pillar, with recent strategies naming compute instruments, setting capacity targets and ringfencing budgets – features that were the exception in earlier strategy documents (see Figure 2). Of the 18 AI strategies adopted or updated in 2025–2026, 13 (approximately 72%) address compute as a dedicated pillar (Tier 1). By contrast, only eight of 26 strategies adopted between 2019 and 2024 are classified as Tier 1.
Several GPAI member countries operate substantive AI compute capability outside their national AI strategies, and comprehensive AI compute policy is often delivered through a combination of strategy text and adjacent instruments. For EU countries, substantive AI compute capability is increasingly delivered through EuroHPC JU channels. Italy's Leonardo supercomputer and IT4LIA AI Factory infrastructure, for example, operate outside the Italian Strategy for AI 2024–2026, but fall within the broader EuroHPC JU programme (AGID, 2024[31]).
Source: OECD analysis based on national AI strategies of 46 GPAI member countries and the EU as of May 2026. Full analysis available in the supporting material at https://oecd.ai/en/supporting-materials-policy-brief-securing-ai-compute.
Source: OECD analysis based on national AI strategies of 46 GPAI member countries and the EU as of May 2026. Full analysis available in the supporting material at https://oecd.ai/en/supporting-materials-policy-brief-securing-ai-compute.
Dedicated AI compute strategies are emerging as a policy option for governments
Some governments articulate AI compute commitments through a dedicated AI compute strategy with dedicated budgets and delivery timelines. Of all 46 GPAI member countries, only two have adopted such approaches: Canada (December 2024) and the United Kingdom (July 2025), making dedicated AI compute strategies a recent but emerging policy option. Both strategies were preceded by pre-adoption processes commissioned by their respective governments: the UK's Independent Review of the Future of Compute (the Ghahramani Review, March 2023) assessed domestic AI compute capacity and identified investment priorities (Department for Science, Innovation and Technology, 2023[32]) and Canada's public consultation in summer 2024 engaged over 1 000 stakeholders from research, industry and civil society, with outcomes summarised in the What We Heard report (Government of Canada, 2024[33]).
Canada’s Sovereign AI Compute Strategy commits CAD 1 billion allocated specifically for building sovereign compute capacity, complemented by a CAD 300 million Access Fund that provides direct subsidies and credits to SMEs and research institutions (Government of Canada, 2025[34]). The programme extends CAD 700 million in investments through the AI Compute Challenge designed to encourage and support private companies to build AI data centre infrastructure (Government of Canada, 2025[34]).
The United Kingdom’s Compute Roadmap establishes a framework for national AI infrastructure development. The country has committed GBP 2 billion through 2030, including GBP 1 billion to expand the AIRR programme and GBP 750 million to build a new supercomputer based in Edinburgh, joining flagship supercomputing systems Isambard-AI and Dawn (GOV.UK, 2025[35]). The strategy emphasises both capacity expansion and democratised access through the AIRR programme, which provides pathways for SMEs and researchers to access advanced compute (GOV.UK, 2025[36]). Central to this approach are AI Growth Zones, designated areas for delivering large-scale AI infrastructure development (GOV.UK, 2025[37]).
Governments are commissioning national reviews and needs assessments to shape future AI compute policy
Governments are commissioning national reviews and needs assessments on AI compute as a standalone government initiative. These reviews produce substantive recommendations to government on national compute policy, which may be followed by formal dedicated strategies or policy actions.
In 2025, the Norwegian government asked the Research Council to conduct a concept study examining the country's computing power needs in the coming years. The resulting report recommends an investment of at least NOK 3.4 billion over 2026–2030 for national HPC infrastructure (Forskningsrådet, 2025[38]). It builds on the Council's 2024 report on heavy computing needs for research and AI, which recommended substantial Norwegian co-investment in the next-generation AI-optimised LUMI supercomputer through EuroHPC (Forskningsrådet, 2025[38]; Forskningsrådet, 2024[39]).
A further joint study, commissioned by Norway's Ministry of Digitalisation and Public Governance and Ministry of Education and Research, extends this work to the public-sector. Led by the Norwegian Digitalisation Agency (Digdir) with Sigma2 (Norway’s national HPC and data storage provider), the study assesses demand for AI compute capacity, external expertise, AI readiness, access models and financing practices across public entities, drawing on surveys and interviews with Norwegian public sector organisations and with other governments and international organisations. It evaluates how Norway should organise and fund AI compute for public-sector use in the short- and long-term – covering delivery models, financing options, governance roles and requirements for digital sovereignty and operational resilience.
In 2026, the Swiss Science Council published the Synergise. Strategise. Realise. report, which assesses the country's AI compute infrastructure and its readiness for future compute-intensive research. The report recommends the development of a long-term national AI infrastructure strategy, anchored in a tiered and interoperable computing infrastructure system spanning three levels: international supercomputing (building on Swiss participation in EuroHPC initiatives such as LUMI and the HEARTS AI Factory Antenna), national HPC (building on the existing Alps supercomputer), and compute centres at Swiss higher education institutions. It further recommends that the system be governed by an independent strategic board appointed by the Federal Council and supported by long-term public funding rather than grant-based financing (Swiss Science Council, 2026[40]).
Beyond strategy documents, governments are securing AI compute through regional frameworks and major national programmes
Governments are also securing AI compute through regionally co-ordinated frameworks and major national programmes, without taking the form of a dedicated AI compute strategy document. The European Union’s approach to AI compute is co-ordinated through multiple interconnected initiatives. EuroHPC JU was established in 2018 as a general supercomputing programme to co-ordinate investment in pan-European HPC infrastructure. EuroHPC JU was amended in 2024 to include AI Factories, ecosystems built around AI-optimised supercomputers that combine computing power, data and talent to develop cutting-edge AI models and to support European start-ups, SMEs and researchers. Over 2021–2027, overall EU investments in supercomputing infrastructure and AI Factories are projected to reach EUR 10 billion through the EuroHPC JU, with AI Factories typically funded through cost-sharing between EuroHPC JU and participating member countries (EuroHPC, n.d.[41]). The network comprises 19 AI Factories across 16 EU countries, complemented by 13 AI Factory Antennas that extend access to countries not directly hosting a Factory (EuroHPC JU, n.d.[42]).
Building on the AI Factories, the EU’s InvestAI initiative, launched in February 2025, mobilises EUR 200 billion in capital with a dedicated EUR 20 billion fund to develop three to five AI Gigafactories (European Commission, n.d.[43]). AI Gigafactories are designed as a next-generation extension of the AI Factories model, with each facility hosting around four times the capacity of an AI Factory and capable of training next-generation AI models (European Commission, 2025[44]). Building on the InvestAI initiative, the 2025 AI Continent Action Plan aims to triple EU data centre capacity within five to seven years and establish digital innovation hubs across member countries (European Commission, n.d.[45]).
Several governments deliver AI compute capability through programmes or initiatives that operate at scale but are not framed as dedicated AI compute strategy documents. India's IndiaAI Mission, Cabinet-approved in March 2024, includes the IndiaAI Compute Capacity pillar which foresees the establishment of a state-of-the-art AI compute infrastructure featuring over 18 000 GPUs, through public-private partnerships (IndiaAI, n.d.[46]). Israel's National Program for AI R&D Infrastructure (Telem Program) launched the National AI Supercomputer, which will allocate about 1 000 GPUs to high-tech companies and researchers through a public-private partnership with Nebius (a neocloud) as operator (Israel Innovation Authority, 2026[47]).
Spatial planning, permitting reforms and energy efficiency requirements are shaping AI data centre expansion
Governments are using zoning frameworks, permitting reforms, and infrastructure designation to enable AI data centre construction at scale. Israel adopted Government Resolution No. 3907 in February 2026, recognising qualifying AI data centres as national infrastructure and establishing a faster, more co-ordinated planning pathway for eligible projects (The Knesset, 2026[48]). In the United Kingdom, data centres were designated as Critical National Infrastructure in September 2024, alongside energy and water systems, allowing the government to support the sector in the event of critical incidents and minimise impacts on the economy (GOV.UK, 2024[49]). The designation signals government prioritisation of the sector and improves conditions for investment, in turn supporting the expansion of national data centre capacity. The July 2025 United States Executive Order 14318, Accelerating Federal Permitting of Data Center Infrastructure includes provisions for loans, loan guarantees, grants and tax incentives, and opens federal lands for data centre and associated energy infrastructure development (The White House, 2025[50]).
Sustainability and energy efficiency considerations are being embedded in national policies and regulatory frameworks governing data centres. Chile's National Data Center Plan sets out a nationwide strategy to foster the sustainable growth of the data centre industry in Chile. The Plan calls for a decentralised model that leverages regional capabilities while prioritising reductions in the energy and water consumption of the data centre industry (MinCiencia, 2024[51]). Germany’s 2023 Energy Efficiency Act outlines efficiency requirements for computing infrastructure as per the European Union’s 2023 Energy Efficiency Directive (EED). The EED requires data centres to report on their energy use (European Commission, n.d.[52]). Taken together, these examples illustrate a broader trend of treating efficiency as an integral dimension of data centre policy.
AI compute hardware alone is not sufficient to enable the development and deployment of AI; a skilled workforce – often with very specific hardware and software skills – is also needed. Equipping people with the skills for AI and supporting workers in an AI-driven labour market are important dimensions of AI compute policy. Government policy initiatives include both making thorough use of AI compute infrastructure (i.e. by having the specialised workforce to use it effectively and efficiently) and preparing for new labour market productivity driven by increased data centre growth and the emergence of new skills from AI compute innovations. Although national strategies often include dimensions to support human capacity development for AI, dedicated AI compute workforce needs assessments or targeted programmes are rare.
Some governments offer AI compute access and training programmes, while also recognising domestic AI compute as key to attracting and retaining top AI talent
The 2026 India AI Impact Summit included the priority of “Expanding Access to AI Resources, Skills, and Technology Nationwide” including by “Democratising AI in India”. This involved initiatives under the IndiaAI Mission to provide access to GPUs for “nearly one third of the global average cost” (AI Impact Summit, 2026[53]). Further up the AI stack, the India Semiconductor Mission 2.0 provides support for industry-led research and training centres to help “build a future-ready skilled workforce, strengthening the long-term foundations for accessible and resilient AI compute in India” (AI Impact Summit, 2026[53]).
Singapore’s 2026 AI Strategy includes the Enterprise Compute Initiative facilitating access for Singapore-based companies to AI tools, cloud compute credits, training and consultancy services. The short-term programme contributes to the broader National AI Strategy 2.0 goal of “anchoring AI teams across Singapore” (Digital Industry Singapore, 2026[54]). The United Kingdom’s Future of Compute Review identified investing in domestic skills and attracting and retaining talent as a top priority (Department for Science, Innovation and Technology, 2023[32]). The UKRI provides grants for AI compute projects encouraging in particular projects involving early career researchers and training to onboard new users of HPC for research purposes (UKRI, 2025[55]).
The United States AI Action Plan strives to “train a skilled workforce for AI infrastructure” through the identification of priority roles underpinning the build-out of AI infrastructure, the development of modern skills frameworks, the promotion of industry-driven training programmes and early pipelines for critical AI infrastructure roles through apprenticeships and hands-on education and training. Recommended policy actions include creating a national initiative to develop skill frameworks and competency models for AI infrastructure occupations, partnering with local organisations and industry to develop targeted education and training programmes aligned with local employer needs, and harnessing the talent and facilities at national laboratories for hands-on research training and development (White House, 2025[12]).
International co-operation can help governments learn from good practices and leverage joint resources to help build both national and regional resilience for AI compute. Governments can co-operate across borders and sectors voluntarily to share information on mutually agreed terms, develop standards and work together towards responsible stewardship of AI. This is especially critical for AI compute where investments are high in capital expenditure and supply chains are complex, making international co-operation with trusted partners often strategic to secure AI compute in a timely manner on mutually agreed terms.
Several countries have established bilateral cross-border AI compute initiatives to drive mutual domestic AI compute access, often centring around research and development
In 2024 Canada and the United Kingdom signalled intent to collaborate on AI compute access for researchers and industry, actions to reduce environmental impacts of compute and developing AI talent (DSIT, 2024[56]). In 2026, Canada and Norway released a joint statement on Sovereign Technology and AI Cooperation, exploring expanded access to AI compute capacity and “promoting sustainable, resilient and environmentally responsible digital infrastructure” (Government of Canada, 2026[57]).
Beyond the EU countries, several countries have chosen to strengthen AI compute capacity through EuroHPC JU, such as Norway, Switzerland, Türkiye and the United Kingdom. This initiative includes access through AI Factories as central AI compute hubs (EuroHPC, n.d.[58]). For example, the LUMI AI Factory consortium formed by Czechia, Denmark, Estonia, Finland, Norway and Poland facilitates shared AI compute access across Nordic and other European countries (LUMI AI Factory, n.d.[59]).
EuroHPC also facilitates collaboration between countries that are geographically further apart. For example, Türkiye is a founding consortium partner in the Barcelona Supercomputing Center AI Factory (BSC-AIF), one of the European AI Factories established under EuroHPC JU. The initiative is a joint effort of Portugal (FCT), Romania (ICI București), Spain (BSC-CNS) and Türkiye (TÜBİTAK) and aims to support strategic sectors across countries, including public administration, healthcare, agriculture and energy (TÜBİTAK, 2024[60]).
There are different ways to boost domestic AI compute capacity; the most effective and resilient approach will depend on a country’s context and needs. Policymakers should reflect on the extent to which domestic AI compute capacity is an economic and strategic priority for their country, with respect to their unique policy and economic contexts including available alternatives. In some instances, cross-border collaborations on AI compute capacity, agnostic to physical location, can be beneficial and sufficient for enabling domestic AI ecosystems. As governments consider sizable public investments, policymakers should firmly ground plans for AI compute in national policy objectives, enabling accountability in public-sector budget allocations and leveraging private-sector investments where possible and desirable.
Governments should commission domestic AI compute needs assessments to inform AI compute policy approaches. The OECD calls for the development of targeted domestic AI compute plans informed by local needs assessments. However, only a few countries have undertaken such reviews. The OECD, including through co-operation with GPAI, can support countries with their national AI compute strategies by providing analytical frameworks, qualitative and quantitative data on a variety of AI compute indicators (e.g. venture capital investment in IT infrastructure and hosting, public cloud compute availability, national AI compute policies) and facilitating the sharing of good practices between members and partners.
Policymakers should consider avenues for dialogue, collaboration and voluntary knowledge sharing on mutually agreed terms with countries at different levels of AI compute development, including in developing and emerging economies. This may require specific support for countries where broader infrastructure, such as connectivity and ecosystem considerations are systemic barriers. Initiatives and tools such as the OECD AI Policy Toolkit and AI Policy Navigator can support such efforts. Governments are encouraged to report initiatives aimed at enhancing domestic AI compute capacity via the OECD.AI Policy Navigator to facilitate the dissemination of good practices.
OECD (2023), “A blueprint for building national compute capacity for artificial intelligence”, OECD Digital Economy Papers, No. 350, OECD Publishing, Paris, https://doi.org/10.1787/876367e3-en.
OECD.AI (2026), OECD.AI Policy Observatory, https://oecd.ai/en/.
OECD.AI (2026), OECD.AI Policy Navigator, https://oecd.ai/en/dashboards/overview.
Contact
Celine Caira and Tana McCauley, gpai@oecd.org
References
[20] ABCI (2025), About ABCI, https://abci.ai/en/about_abci/ (accessed on 4 September 2025).
[31] AGID (2024), Italian Strategy For Artificial Intelligence 2024-2026, https://www.agid.gov.it/sites/agid/files/2024-07/Italian_strategy_for_artificial_intelligence_2024-2026.pdf (accessed on 27 May 2026).
[53] AI Impact Summit (2026), Democratising AI in India, https://www.pib.gov.in/PressReleasePage.aspx?PRID=2225781®=3&lang=2 (accessed on 24 May 2026).
[17] CNRS (2024), « Il s’agira probablement du plus grand supercalculateur dédié à l’IA en France », https://www.cnrs.fr/fr/actualite/il-sagira-probablement-du-plus-grand-supercalculateur-dedie-lia-en-france (accessed on 1 October 2025).
[32] Department for Science, Innovation and Technology (2023), Independent Review of The Future of Compute: Final report and recommendations, https://www.gov.uk/government/publications/future-of-compute-review/the-future-of-compute-report-of-the-review-of-independent-panel-of-experts (accessed on 27 May 2026).
[54] Digital Industry Singapore (2026), Enterprise Compute Initiative, https://www.disg.gov.sg/enterprise-compute-initiative/ (accessed on 28 May 2026).
[11] DSIT (2025), AIRR advanced supercomputers for the UK, https://www.gov.uk/government/publications/ai-research-resource/airr-advanced-supercomputers-for-the-uk (accessed on 25 May 2026).
[56] DSIT (2024), UK-Canada cooperation in AI compute: memorandum of understanding, https://www.gov.uk/government/publications/uk-canada-cooperation-in-ai-compute-memorandum-of-understanding (accessed on 28 May 2026).
[41] EuroHPC (n.d.), Discover EuroHPC JU, https://www.eurohpc-ju.europa.eu/about/discover-eurohpc-ju_en (accessed on 4 September 2025).
[58] EuroHPC (n.d.), The European High Performance Computing Joint Undertaking (EuroHPC JU), https://www.eurohpc-ju.europa.eu/ai-factories/ai-factories-access-modes_en (accessed on 24 May 2026).
[42] EuroHPC JU (n.d.), AI Factories, https://www.eurohpc-ju.europa.eu/ai-factories_en (accessed on 28 May 2026).
[44] European Commission (2025), AI Continent Action Plan, https://digital-strategy.ec.europa.eu/en/factpages/ai-continent-action-plan (accessed on 28 May 2026).
[13] European Commission (2025), European AI in Science Strategy, https://research-and-innovation.ec.europa.eu/strategy/strategy-research-and-innovation/our-digital-future/european-ai-science-strategy_en (accessed on 24 May 2026).
[14] European Commission (n.d.), AI Factories, https://digital-strategy.ec.europa.eu/en/policies/ai-factories (accessed on 24 May 2026).
[52] European Commission (n.d.), Energy performance of data centres, https://energy.ec.europa.eu/topics/energy-efficiency/energy-efficiency-targets-directive-and-rules/energy-efficiency-directive/energy-performance-data-centres_en (accessed on 28 May 2026).
[43] European Commission (n.d.), EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence*, https://ec.europa.eu/commission/presscorner/detail/en/ip_25_467 (accessed on 4 September 2025).
[45] European Commission (n.d.), Shaping Europe’s leadership in artificial intelligence with the AI continent action plan, https://commission.europa.eu/topics/eu-competitiveness/ai-continent_en? (accessed on 4 September 2025).
[38] Forskningsrådet (2025), Slik bør Norge investere 3,4 milliarder til infrastruktur for tungregning, https://www.forskningsradet.no/nyheter/2025/investere-milliarder-infrastruktur-tungregning/ (accessed on 28 May 2026).
[39] Forskningsrådet (2024), Kraftig økning i behovet for tungregnekraft, https://www.forskningsradet.no/nyheter/2024/kraftig-okning-i-behovet-for-tungregnekraft/ (accessed on 27 May 2026).
[18] Garreau, M. (2025), A quoi va servir l’AI factory de la France, soutenue par l’Europe ?, https://www.usinenouvelle.com/editorial/a-quoi-va-servir-l-ai-factory-de-la-france-soutenue-par-l-europe.N2230196 (accessed on 24 September 2025).
[10] Gauss Centre for Supercomputing (n.d.), Supercomputing and Artificial Intelligence, https://www.gauss-centre.eu/for-users/gcs-supercomputing-und-artificial-intelligence (accessed on 24 May 2026).
[37] GOV.UK (2025), AI Growth Zones, https://www.gov.uk/government/collections/ai-growth-zones (accessed on 25 May 2026).
[36] GOV.UK (2025), AIRR advanced supercomputers for the UK, https://www.gov.uk/government/publications/ai-research-resource/airr-advanced-supercomputers-for-the-uk (accessed on 4 September 2025).
[35] GOV.UK (2025), UK Compute Roadmap, https://www.gov.uk/government/publications/uk-compute-roadmap/uk-compute-roadmap#conclusion-and-next-steps (accessed on 4 September 2025).
[49] GOV.UK (2024), Data centres to be given massive boost and protections from cyber criminals and IT blackouts, https://www.gov.uk/government/news/data-centres-to-be-given-massive-boost-and-protections-from-cyber-criminals-and-it-blackouts (accessed on 28 May 2026).
[57] Government of Canada (2026), Canada-Norway Joint Statement on Sovereign Technology and AI Cooperation, https://www.canada.ca/en/innovation-science-economic-development/news/2026/03/canada-norway-joint-statement-on-sovereign-technology-and-ai-cooperation.html (accessed on 24 May 2026).
[34] Government of Canada (2025), Canadian Sovereign AI Compute Strategy, https://ised-isde.canada.ca/site/ised/en/canadian-sovereign-ai-compute-strategy (accessed on 4 September 2025).
[33] Government of Canada (2024), What We Heard Report: Consultations on AI Compute, https://ised-isde.canada.ca/site/ised/en/what-we-heard-report-consultations-ai-compute (accessed on 27 May 2026).
[15] Gyu-lee, L. (2026), Korea to invest $5.8 bil. in semiconductor, AI, battery technologies, https://www.koreatimes.co.kr/southkorea/society/20260313/korea-to-invest-58-bil-in-semiconductor-ai-battery-technologies (accessed on 4 May 2026).
[19] Harris, D. (n.d.), Japan Enhances AI Sovereignty With Advanced ABCI 3.0 Supercomputer, https://blogs.nvidia.com/blog/abci-aist/ (accessed on 4 September 2025).
[7] Hood, C. (1983), The tools of government, Macmillan Education UK.
[26] IndiaAI (n.d.), IndiaAI Compute Capacity, https://indiaai.gov.in/hub/indiaai-compute-capacity (accessed on 4 September 2025).
[46] IndiaAI (n.d.), IndiaAI Compute Capacity, https://indiaai.gov.in/hub/indiaai-compute-capacity (accessed on 28 May 2026).
[9] ISED (2026), AI Sovereign Compute Infrastructure Program, https://ised-isde.canada.ca/site/ised/en/ai-sovereign-compute-infrastructure-program (accessed on 24 May 2026).
[29] ISED (2025), AI Compute Challenge, https://ised-isde.canada.ca/site/ised/en/canadian-sovereign-ai-compute-strategy/ai-compute-challenge (accessed on 24 May 2026).
[8] ISED (2025), Canadian Sovereign AI Compute Strategy, https://ised-isde.canada.ca/site/ised/en/canadian-sovereign-ai-compute-strategy (accessed on 24 May 2026).
[47] Israel Innovation Authority (2026), Access to National Supercomputer is Open, Offering Discounted Nvidia B200 GPUs Accelerators to High-Tech Companies and Researchers, https://innovationisrael.org.il/en/press_release/supercomputer-access-2026/ (accessed on 28 May 2026).
[3] Lehdonvirta, V., B. Wu and Z. Hawkins (2024), Compute North vs. Compute South: The Uneven Possibilities of Compute Governance Around the Globe, AAAI/ACM Conference on AI, Ethics, and Society.
[59] LUMI AI Factory (n.d.), At the forefront of European HPC and AI development, https://lumi-ai-factory.eu/about-us/lumi-aif-consortium/ (accessed on 24 May 2026).
[16] MCTI (2025), AI for the Good of All: Brazilian Artificial Intelligence Plan, https://www.gov.br/mcti/pt-br/acompanhe-o-mcti/transformacaodigital/ai-for-the-good-of-all-2025.pdf (accessed on 7 July 2026).
[30] METI (2024), Approval of Plans for Ensuring a Stable Supply of Cloud Programs under the Economic Security Promotion Act, https://www.meti.go.jp/english/press/2024/0419_001.html (accessed on 27 May 2026).
[51] MinCiencia (2024), National Data Centers Plan 2024-2030, https://api.oecdai.org/storage//policy-initiatives/May2026/117ojp2admosrnx09-national_data_centers_plan.pdf (accessed on 28 May 2026).
[21] Novo Nordisk Foundation (2024), Denmark’s first AI supercomputer is now operational, https://novonordiskfonden.dk/en/news/denmarks-first-ai-supercomputer-is-now-operational/ (accessed on 26 May 2026).
[5] OECD (2026), Benefits of AI openness: OECD discussion paper for the G7, OECD Publishing, https://doi.org/10.1787/876367e3-en.
[2] OECD (2024), Recommendation of the Council on Artificial intelligence, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449.
[1] OECD (2023), A blueprint for building national compute capacity for artificial intelligence, OECD Publishing, https://doi.org/10.1787/876367e3-en.
[4] OECD (Forthcoming), “Powering AI Progress: The Future of Compute”, OECD Digital Economy and Society Outlook.
[28] OECD.AI Policy Navigator (2026), South Korea National AI GPU Infrastructure Initiative, https://oecd.ai/en/dashboards/policy-initiatives/south-korea-national-ai-gpu-infrastructure-initiative-1 (accessed on 27 May 2026).
[24] OpenAI (2025), Announcing The Stargate Project, https://openai.com/index/announcing-the-stargate-project/ (accessed on 4 September 2025).
[25] OpenAI (2025), OpenAI, Oracle, and SoftBank expand Stargate with five new AI data center sites, https://openai.com/index/five-new-stargate-sites/ (accessed on 10 May 2026).
[27] Park, M. (2025), South Korea Invests $1.1B to Build National AI GPU Infrastructure, https://koreatechtoday.com/south-korea-invests-1-1b-to-build-national-ai-gpu-infrastructure/ (accessed on 4 September 2025).
[22] PIF (2025), HRH Crown Prince launches HUMAIN as global AI powerhouse, https://www.pif.gov.sa/en/news-and-insights/press-releases/2025/hrh-crown-prince-launches-humain-as-global-ai-powerhouse/ (accessed on 4 September 2025).
[6] Salamon, L. (2002), The tools of government: A guide to the new governance, Oxford University Press.
[23] Soliman, M. (2025), Saudi Arabia’s AI ambition, and what it means for the United States, https://www.mei.edu/publications/saudi-arabias-ai-ambition-and-what-it-means-united-states (accessed on 4 September 2025).
[40] Swiss Science Council (2026), Swiss Science Council recommends national AI infrastructure strategy, https://www.wbf.admin.ch/en/newnsb/TuWsKZqat-kRc99SWg_nB (accessed on 28 May 2026).
[48] The Knesset (2026), Internal Affairs and Environment Committee approves for final readings bill to advance establishment of AI server farms as national infrastructure, https://main.knesset.gov.il/en/news/pressreleases/pages/press672026y.aspx (accessed on 16 July 2026).
[50] The White House (2025), ACCELERATING FEDERAL PERMITTING OF DATA CENTER INFRASTRUCTURE, https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/ (accessed on 28 May 2026).
[60] TÜBİTAK (2024), National Artificial Intelligence Factory is Established!, https://tubitak.gov.tr/en/news/national-artificial-intelligence-factory-established (accessed on 28 July 2026).
[55] UKRI (2025), Access to high performance computing facilities: autumn 2025, https://www.ukri.org/opportunity/access-to-high-performance-computing-facilities-autumn-2025/ (accessed on 27 May 2026).
[12] White House (2025), America’s AI Action Plan, https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf (accessed on 27 May 2026).