Competition in artificial intelligence infrastructure

3. Market features in AI infrastructure

There are a range of different markets in the AI infrastructure chain, and each will have their own unique dynamics. While we are seeing constant innovation in both the AI systems being developed and launched and in the underpinning technology infrastructure, many layers of physical infrastructure underpinning AI have been relatively stable in relation to the fundamental organisation of the supply chain. This constancy makes it easier to assess potential competition issues and risks (Narechania and Sitaraman, 2023[45]).

Doing a full market-by-market analysis is beyond the scope of this paper. However, this chapter seeks to highlight some important features which appear to be present at multiple levels of the supply chain. These features may have important ramifications for competition and are therefore a good basis to start exploring potential competition policy responses in AI infrastructure. The intention is to survey market features that authorities and policy makers may wish to consider in more detail, noting that AI infrastructure covers a wide variety of different markets, with some common themes which this paper seeks to highlight.

3.1. High levels of innovation and intellectual property

The AI infrastructure supply chain is marked by a very high degree of research and development spend, innovation and intellectual property. At many layers of the supply chain such as chip design, lithography and fabrication firms must make exceptionally high investments in research and development. This is driven by the complexity and scale of the technologies involved. Competition in data centres may also increasingly focus on developing the optimal technologies for energy usage and cooling. Figure 5 highlights that firms involved in the chip ecosystem have significantly increased their research and development spending in the last ten years, with just three firms spending almost USD 20 billion on research and development in 2024.

Figure 5. Spend on research and development, example key firms in the chip ecosystem, 2015-2024

Notes: Converted to USD but not adjusted for inflation, R&D spending encompasses broader R&D investment beyond AI specific technology.

Source: OECD analysis based on annual reports.

In such sectors, intellectual property protections, including patents and trade secrets, can be essential to allow firms to recoup their R&D investments and maintain competitive advantages (Aguirre, 2024[42]). Licensing refers to the legal process by which the owner of IP rights grants permission to another party to use, produce, or commercialise the protected invention, work, or technology under agreed terms and conditions. Licensing is an important instrument for diffusing innovation, for allowing innovators to be rewarded for their efforts, and to promote co-operation and follow-on innovation during IP rights’ period of exclusivity. However, licensing agreements can also have anticompetitive effects, such as anticompetitive foreclosure (OECD, 2019[74]).

Firms with leading technological positions do not necessarily have to licence their technology to other suppliers but may do so for strategic reasons, for example generating additional revenue without taking on manufacturing risk. In some situations, the refusal to licence IP can infringe competition law. This is typically only under exceptional circumstances where a provider is in a dominant position and the IP is an essential input in another market. For example, there was previously a case against Microsoft for refusing to licence the specifications required to ensure interoperability of Windows to manufacturers of competitor server operating systems (European Union, 2007[75]). More recently there have been reported investigations into the licencing terms of the Microsoft cloud (Godoy, 2024[76]) (Reuters, 2025[77]), as well as complaints reportedly being raised by Qualcomm (a chip maker) about ARM’s (Chip IP developer) licensing model.1

There are broadly two types of licenses in the sector (Business Software Allliance, 2005[78]):

• proprietary or commercial licenses - which restrict usage to specific terms often involving fees and limited redistribution

• open-source licences – which aim to promote collaboration and innovation by allowing free use and modification.

The use of open-source licences has increasingly been used as a strategic tool by companies seeking to overcome the ecosystem advantages that a technology first mover has gained (see, for example, (Klotz, 2025[79])). This strategy encourages other players in the sector to contribute to systems and technologies around an open-source system which is not under proprietary control helping to reduce potential ecosystem barriers to entry and expansion.

Licensing can raise challenges for competition authorities in some circumstances, and they will need to carefully balance the need to protect intellectual property rights and ensure innovation and investment are not discouraged with the need to prevent dominant firms from leveraging the technology to exclude rivals from other markets. In doing so authorities may need to consider the likelihood to which rivals would be able to leverage open-source technologies to develop alternative solutions.

Beyond the complex interaction of intellectual property and competition, markets based on high levels of innovation may face other competition issues including high concentration and barriers to entry which are discussed further below.

3.2. High concentration and barriers to entry

While the innovation highlighted above is a positive outcome, in several layers of the supply chain there is also a high degree of concentration. Table 1 below illustrates the extent of concentration across several key segments of the AI infrastructure supply chain. In three segments, a single firm is reported to hold over 80% of the market, while in three others, the top three firms collectively hold over 60%.

Table 1. Examples of a concentrated supply chain

Key AI supply segment	Global share of largest provider reported over 80%	Global share of largest 3 players reported as having over 60% share
Advanced Lithography tools	Yes – ASML
Advanced AI chip fabrication	Yes - TSMC
GPUs for AI compute	Yes – Nvidia
HBM chips		Yes – SK Hynix, Samsung and Micron
Cloud provision		Yes – AWS, Google and Microsoft
EDA software		Yes – Cadence, Synopsis and Siemens

Note: The OECD has not done a market definition exercise, and these shares may not therefore relate to economic markets but are commonly understood to be key parts of the AI supply chain.

Sources: Authorité de la concurrence (2023[53]), Opinion 23-A-08 of 29 June 2023 on competition in the cloud sector, https://www.autoritedelaconcurrence.fr/en/opinion/competition-cloud-sector; Netherlands Authority for Consumers and Markets (2022[70]) Market Study Cloud services; Ofcom (2023[80]) Cloud services market study - Final report; Japan Fair Trade Commission (2022[81]) Report on Fact-Finding Survey on Trade Practices by Digital Platform Operators Report on Trade Practices in Cloud Services Sector Contents, https://www.jftc.go.jp/en/pressreleases/yearly-2022/June/221102EN.pdf; Korea Fair Trade Commission (Korea Fair Trade Commission, 2022[82]), Announcement of the results of the cloud field survey, https://www.ftc.go.kr/www/selectBbsNttView.do?pageUnit=10&pageIndex=1&searchCnd=all&searchKrwd=%ED%81%B4%EB%9D%BC%EC%9A%B0%EB%93%9C&key=12&bordCd=3&searchCtgry=01,02&nttSn=42705 ; OECD (2025[8]); Mapping the semiconductor value chain: Working towards identifying dependencies and vulnerabilities. https://doi.org/10.1787/4154cdbf-en; Narechania and Sitaraman (Narechania and Sitaraman, 2023[45]), An Antimonopoly Approach to Governing Artificial Intelligence, https://dx.doi.org/10.2139/ssrn.4597080; Center for Security and Emerging Technology (2020[22]), AI Chips: What They Are and Why They Matter, https://cset.georgetown.edu/publication/ai-chips-what-they-are-and-why-they-matter/; Aguirre (2024[42]) On Labs and Fabs: Mapping How Alliances, Acquisitions, and Antitrust are Shaping the Frontier AI Industry, https://doi.org/10.48550/arXiv.2406.01722 and Narechania and Sitaraman (2023[45]), An Antimonopoly Approach to Governing Artificial Intelligence, Sitaraman, Ganesh and Narechania, Tejas N., An Antimonopoly Approach to Governing Artificial Intelligence (January 17, 2024). 43 Yale Law and Policy Review 95 (2024), Vanderbilt Law Research Paper No. 24-8, Available at SSRN: https://ssrn.com/abstract=4597080 or http://dx.doi.org/10.2139/ssrn.4597080.

The firms listed in Table 1 are examples, focussing on some of the highest value areas of the supply chain. However, it is important to note that high concentration is also found throughout the semiconductor supply chain where the technologies often involve specialised suppliers which have very high market shares (Burja, 2024[83]). Other elements such as fibre network infrastructure and energy networks by their nature may also be potentially natural monopolies or oligopolies at a local level.

It is important to note that market concentration is not inherently problematic. Given the high levels of research and development, the high concentration and profits in these markets can likely, at least initially, be viewed as a reward for their investments and innovation. This is often referred to as the escape effect (Arrow, 1962[84]; OECD, 2023[85]; Schumpeter, 1934[86]) and creates incentives for further innovation.

At some levels of the supply chain, the competition on innovation can even lead to a situation of ‘competition for the market’ where firms compete through innovations and investment to become the sole or dominant provider for the next decade (OECD, 2019[87]). In the context of AI infrastructure, such competition can drive rapid advancements in compute capabilities, data management and model deployment efficiency, as firms seek to establish themselves as indispensable suppliers in the AI ecosystem.

There are also barriers to entry which make it challenging for firms to compete which can contribute to the high level of concentration and weaken competition. These barriers include:

• Very high capital requirements – many levels of the supply chain are very capital intensive, for example the tools for manufacturing chips can cost hundreds of millions of US dollars alone (Shilov, 2024[88]). While entry into chip design is possible in partnership with a manufacturer such as TSMC (or more recently Intel), as discussed above the high levels of innovation also create a very high degree of investment needed in research and development.

• Economies of scale – given the very high level of fixed costs at many levels of the supply chain a large degree of scale is needed to cover the fixed costs. For example, at the cloud level it has been found that there are substantial economies of scale (OECD, 2025[5]).

• Long lead times – building and connecting data centres can take years and developing chip manufacturing plants even longer. Given the highly dynamic markets, these long lead-times create additional risks which may restrict entry.

In addition, some elements of the supply chain such as chip design also have proprietary software ecosystems which add additional barriers to develop. For example, the EU’s case assessing the merger of Nvidia and OpenAI highlighted the high barriers to entry in discrete GPUs for use in datacentres including the need to have a well-developed software stack (European Commission, 2024[89]).

High concentration driven by innovation can still create a risk of anti-competitive conduct. Once a firm secures a dominant position, the incentives to innovate may shift, potentially leading firms seeking to protect their position. Potential behaviour includes the use of conditional rebates or other mechanisms to prevent rivals from reaching sufficient scale, or simply through targeting smaller innovative rivals through a series of killer acquisitions. Concentration can also increase the risk of collusion, particularly in markets with a small number of players with similar market shares and technology that has matured.

3.3. Vertical and conglomerate relations

At many levels of the AI infrastructure supply chain, the complexity historically led to a level of specialisation where no one company was able to produce everything in a fully vertically integrated way.2 Today, though there is an increasing amount of vertical integration in the value chain, as well as firms expanding into offering multiple services which may be complementary (conglomerates). For example, the hyperscalers have begun developing AI accelerator chips and investing in downstream modellers (Google, 2024[90]; Amazon, 2024[56]; Meta, 2024[91]; Warnock, 2025[92]), and Nvidia has expanded into networking solutions through the acquisition of Mellanox (European Commission, 2019[93]).

Figure 6 below shows two case studies of vertical integration across the AI infrastructure supply chain, which extends down to consumer facing AI models and services utilising AI.

Figure 6. Case studies of vertical integration in AI infrastructure supply chain

Note: This figure is illustrative of examples of vertical integration and is not intended as an exhaustive list of products, market segments and supply chain steps. Similar vertical integration figures could be generated for numerous firms in the AI infrastructure supply chain (e.g. Amazon, Microsoft and Intel).

Source: https://www.gov.uk/government/publications/cma-ai-strategic-update/cma-ai-strategic-update; https://www.generativevalue.com/p/the-ai-semiconductor-landscape ; https://www.nvidia.com/en-us/ai/ ; https://www.nvidia.com/en-us/data-center/h200/ ; https://www.nvidia.com/en-us/data-center/; https://www.nvidia.com/en-us/data-center/colocation-partners/; https://developer.nvidia.com/ai-models/; https://www.nvidia.com/en-gb/geforce/technologies/dlss/ ; https://www.nvidia.com/en-us/geforce-now/; Authorité de la concurrence (2023[53]) Opinion 23-A-08 of 29 June 2023 on competition in the cloud sector , https://www.autoritedelaconcurrence.fr/en/opinion/competition-cloud-sector; Netherlands Authority for Consumers and Markets (2022[70]) Market Study Cloud services. 

Vertical integration in the AI supply chain (as in other markets), has the potential for several benefits to consumer welfare. This includes enhancing investment incentives for producers, improving efficiency through enhanced co-ordination and removing double marginalisation (Beck and Scott Morton, 2021[94]). Given the very high barriers to entry, large providers moving to compete at other levels of the supply chain may also provide a route for potential competition which may otherwise be absent.

On the other hand, vertical integration involving firms with high market shares and power, can increase barriers to entry and raise the risk that dominant firms utilise their position at different parts of the value chain to tilt competition in their favour and make it harder for others to compete. In dynamic markets such as AI infrastructure, vertical foreclosure can be an optimal strategy because, although it can lead to short term profit loss, it affects future market structure, increasing the integrated firm's future profits (Fumagalli and Motta, 2020[95]). Such foreclosure could occur through different mechanisms including for example bundling/tying or even refusal to supply a key input.

These concerns may also extend beyond straightforward vertical integration between suppliers and producers to conglomerate integration at the same level where firms may employ similar tactics.

3.4. Partnerships and minority shareholdings

Agreements between firms can enhance efficiency, co-ordination and investment incentives. For instance, exclusive supply or distribution arrangements can ensure stable demand, enabling upstream firms to invest in specialised production or research and development. They can also help reduce transaction costs and improve logistical integration, which is important in sectors with high fixed costs and fast innovation cycles (OECD, 2021[96]). In the AI stack, partnerships can potentially help to accelerate innovation, optimise infrastructure use and grow the market (Ansari, 2025[97]). For example, they can help to ensure that companies developing AI models have access to the compute resources they need without undertaking huge capital investments (Auer and Zúñiga, 2025[98]).

It has become increasingly common for firms in the AI infrastructure supply chain to make investments and seek partnerships with other parties at different levels of the supply chain (Groza and Wierzbicka, 2024[99]). Often such arrangements appear to fall short of full mergers or acquisitions, will be for minority equity stakes and are structured together with a broader set of arrangements (Federal Trade Commission, 2025[100]).

Figure 7 highlights some of the recent deals which have been agreed in six weeks (September- October 2025). It demonstrates both the huge amounts of capital at stake, how quickly the industry is developing and the increasing number of cross holdings and partnerships. While some are simply long-term supply agreements others involve equity stakes and other arrangements. Others have been in place much longer such as the Microsoft - OpenAI partnership which has been in place since 2019 and recently been updated (OpenAI, 2025[101]).

While agreements can enhance efficiency and innovation, they can also pose potential risks to competition (Groza and Wierzbicka, 2024[99]) especially when involving competitors or potential competitors. As highlighted above, one major concern in relation to firms with significant market share in AI infrastructure is the risk that firms will seek to foreclose rivals. The partnership arrangements with firms at other levels of the supply chain may change the incentives or ability of firms in the market to carry out such foreclosure. For example, the CMA highlights the risk that partnerships could reinforce or extend existing positions of market power through the value chain (Competition and Markets Authority, 2024[102]).

Figure 7. Reported agreements in AI infrastructure between September and October 2025

Note: The above agreements are a selection of the most recent agreements at the time of writing, thy are demonstrative and not exhaustive.

Source: OECD summary of news based on: FT (2025) Microsoft taps Nebius to supply up to USD 20 billion of AI computing power, https://www.ft.com/content/a7aadbfe-1d6c-4159-890a-2f91d97a8d4b ; WSJ (2025) Exclusive | Oracle, OpenAI Sign USD 300 Billion Cloud Deal – WSJ https://www.wsj.com/business/openai-oracle-sign-300-billion-computing-deal-among-biggest-in-history-ff27c8fe ; ASML (2025) asml-mistral-ai-enter-strategic-partnership, https://www.asml.com/en/news/press-releases/2025/asml-mistral-ai-enter-strategic-partnership

Reuters (2025), CoreWeave, Nvidia sign USD 6.3 billion cloud computing capacity order https://www.reuters.com/business/coreweave-nvidia-sign-63-billion-cloud-computing-capacity-order-2025-09-15/ Nvidia (2025) , NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products, https://nvidianews.nvidia.com/news/nvidia-and-intel-to-develop-ai-infrastructure-and-personal-computing-products

TomsHardware (2025) 900 million on enfabrica tech, https://www.tomshardware.com/tech-industry/nvidia-drops-a-cool-usd900-million-on-enfabrica-tech-and-hiring-its-ceo-report-claims-ai-networking-chip-company-boasts-capacity-to-connect-100-000-gpus-together#xenforo-comments-3886527 Nvidia (2025) OpenAI and Nvidia announce strategic partnership, https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems ; Reuters (2025), Alibaba shares leap on Nvidia partnership, data center plans https://www.reuters.com/world/china/alibaba-launches-qwen3-max-ai-model-with-more-than-trillion-parameters-2025-09-24/.

Another aspect of the agreements which could cause competition risk is that even if equity stakes do not grant control communications facilitated by corporate governance rights can create opportunities for collusion between rivals (Ha and Law, 2025[103]). Recently for example the European Commission fined companies in the delivery market relating to the exchange of sensitive information (Fortuna, 2025[104]). As such minority shareholdings grow in AI infrastructure the risk of collusion may also grow.

The competition implications stemming from each agreement are difficult to assess and would need to carefully be examined. The possibility of examining under the merger regime and enforcement are discussed in section 4.

3.5. Switching barriers

Adding to entry barriers and limiting competition even between firms in the market there are several areas of the supply chain which have potentially high switching barriers. These barriers include:

• Contracts and pricing: Large customers often enter multi-year contracts with chip suppliers that include volume commitments, exclusivity clauses, or favoured pricing tiers to ensure their investments are worthwhile. Such commitments will often be driven by fears of scarcity but may mean that customers’ ability to switch is limited in the short term. At the cloud level of the supply chain, concerns have been highlighted about the predicting contract spend, and potential barrier created by egress fees (charges for data leaving a cloud network) that may amount to penalising customers for switching providers. (OECD, 2025[5]).

• Lack of standards and interoperability: Given the fast pace of developments and the high levels of innovation, many areas of the supply chain do not have agreed standards and interoperable parts. This means, for example, that switching suppliers can often entail reconfiguring data centres, updating firmware, and modifying cooling and power systems. These physical and operational changes can be prohibitively expensive, especially enterprises with large-scale deployments. There has recently been an initiative launched by the Open Compute Project to seek to change this, with the support of Google, Meta and Microsoft (Open Compute Project, 2025[105]). Integration with existing cloud services may also be optimised for specific hardware, creating additional friction, and at the cloud level the ability to integrate services across multiple providers is limited by the use of proprietary interfaces and vendor specific solutions (OECD, 2025[5])

• Feedback loops and ecosystems: Like the previous waves of digital developments, key parts of the AI supply chain rely on integration between hardware and software. These integrated ecosystems can have benefits including seamless integration, enhanced user experience, convenience and efficiency, and enhanced security. Such ecosystems can also have strong network effects. These arise when the value of a system increases as more users, developers or data contributors participate (OECD, 2017[106]). For example, as more developers and enterprises adopt certain software ecosystems, the software become more valuable to other users as the number of tools and use cases increase. In AI infrastructure at different layers such software ecosystems or platforms have been made on a proprietary basis and tied to hardware, for example Nvidia’s CUDA software. This then creates barriers to switch to alternative hardware, as a customer may need to make a complete ecosystem switch at significant expense. This is why challenger firms often seek to lead open-source initiatives to try and overcome the network effects by pooling resources with other challengers.

Switching barriers such as these are not necessarily be the result of illegal conduct but can add friction to competition by making it harder for companies to win on the merits and for customers to pick the optimal product or service for them. In section 4 there is a broader discussion of potential competition authority responses including through market studies and regulation which could potential seek to lower switching barriers.

3.6. Supply shortages/bottlenecks

The chip supply chain has in the past seen supply shortages (Jackson, 2024[107]). With the current forecast for demand for AI indicating massive demand, there could be a risk of supply shortages or bottlenecks emerging again, especially for cutting-edge chips and advanced lithography systems which are produced by a small number of companies in a limited number of locations.

The cost and scarcity of the specialised AI accelerator chips, such as GPUs and ASICs, may represent an important bottleneck for the sector (European Commission, 2024[108]). As chip designers begin to expand their offerings downstream into the cloud, and cloud providers begin to develop their own chips, a risk could grow that there are firms that could try to exclude rivals from upstream or downstream inputs or services. For example, a CMA market investigation highlighted that the supply of AI accelerator chips (such as GPUs) has been unable to keep pace with increasing demand associated with AI over the past three years. A delay in Nvidia chip delivery in 2024 had been a key factor in one cloud computing provider experiencing lower estimated revenue from AI compute for 2025 (Competition and Markets Authority, 2025[64]).

Shortages may also emerge in relation to local or national markets in relation to key energy, networking and water resources. With firms which are unable to gain access potentially being locked out of the market.

Shortages could both slow innovation in AI developments downstream and concentrate power, creating a risk of resource hoarding and preferential allocation. In such a context, firms with privileged access, often through exclusive agreements or via vertical integration, can accelerate their AI development while others are left behind. Control over who gets supplied becomes a strategic lever in this environment. Dominant cloud providers and chip manufacturers may, for example, prioritise internal projects or preferred partners, effectively shaping the competitive landscape.

On the other hand, demand outstripping supply can create the potential for alternative competitors to gain share, for example, in the operation of data centres several smaller providers appeared to have recently received large orders with AI modellers and cloud providers looking for as much supply as possible. The amount of capital being driven into the sector has recently raised concerns of a potential bubble, (Thornhill, 2025[109]) to the extent that there is any crash in confidence or demand for AI companies, the competition dynamics could change dramatically and make it harder for new entrants to raise funds.

3.7. Increasing state intervention

Given the importance of AI to all sectors and especially high technology industries, there is increasing government intervention in relation to the AI supply chain. Governments around the world are investing heavily in AI infrastructure because it is increasingly viewed as a strategic asset which is central to economic competitiveness, national security and technological sovereignty (Nakazawa and Pisa, 2025[110]).

Compute power is a key building block of modern AI systems, and control over these resources can shape innovation, industrial capabilities, and geopolitical influence. Public investment can also potentially help to address potential market failures, such as underinvestment in open-access compute. For example, the French (Autorité de la concurrence, 2024[111]) highlighted that certain technical and organisational developments, as well as public policies, may limit the barriers to entry including providing access to public supercomputers. Access to key infrastructure such as supercomputers can therefore be procompetitive as it can help to prevent potential exclusion or exploitation by dominant providers.

However, public investment itself can also raise competition policy challenges. If not carefully designed, public-private partnerships may entrench incumbents, distort markets or create dependencies on dominant platforms (OECD, 2022[112]). Preferential access to state-funded compute or data, for example, could reinforce existing power asymmetries. Moreover, governments may inadvertently favour national champions or larger firms, undermining cross-border competition and innovation.

Box 3 highlights some of the examples of government intervention from the first developments of the semiconductor to more recent government moves to develop AI infrastructure.

Box 3. Examples of OECD government investments and AI infrastructure partnerships

The origins of the semiconductor industry are intertwined with government programmes. In the 1950s and 1960s, the US Department of Defence and NASA were among the first major customers for integrated circuits, providing critical demand and funding that enabled firms like Fairchild Semiconductor and Texas Instruments to scale production. Defence contracts helped subsidise R&D, while government labs and procurement policies accelerated the commercialisation of transistor and chip technologies. This early public investment laid the foundation for Silicon Valley’s rise and established a model of state-supported innovation in strategic technologies.

Governments in East Asia in the 1970s and 1980s were also highly involved in the development of their own semiconductor industries. For example, Japan provided extensive funding for research and development in semi-conductor manufacturing and South Korea also used targeted R&D subsidies, export incentives and infrastructure development. Chinese Taipei, meanwhile, leveraged its semiconductor leadership anchored by TSMC, with government-backed industrial parks and research institutes.

In recent years, governments have intensified efforts to secure strategic control over AI infrastructure for example:

• The US government launched a series of initiatives to bolster domestic AI infrastructure. The CHIPS and Science Act (2022) allocated USD 52 billion to support semiconductor manufacturing and R&D, with a focus on AI-relevant technologies. In 2025 the US government announced a deal with Intel to purchase 10% of its equity for a USD 10 billion investment to expand US manufacturing facilities.

• The European Union launched a public-private investment strategies in AI infrastructure, centred around the InvestAI initiative and the broader AI Continent Action Plan. InvestAI aims to mobilise EUR 200 billion in AI investment, including EUR 20 billion specifically earmarked for the development of AI gigafactories, large-scale computing facilities designed to support the training of frontier models. EU member states have also recently been investing in AI infrastructure. The EU Commission recently approved a state aid package from Germany to support Infineon in setting up a new semiconductor manufacturing facility. France has also invested in national compute capacity and supporting startups through Bpifrance and the Grand Défi initiative.

• The UK government’s GBP 2 billion AI Opportunities Action Plan includes sovereign compute investments and regional AI Growth Zones.

• South Korea in 2025 announced a support package of over USD 23 billion to support its semiconductor industry and Japan was reported to have invested USD 27 billion in support of the semiconductor industry between 2000 and 2024 with plans for USD 65 billion or more by 2030.

Note: This box includes a sample of examples and is by no means exhaustive.

Sources: Miller (2022[113]) Chip war: the fight for the world’s most critical technology; Simon & Schuster; OECD (2019[63]) Measuring distortions in international markets: The semiconductor value chain, https://doi.org/10.1787/8fe4491d-en; OECD (2025[8]), Mapping the semiconductor value chain: Working towards identifying dependencies and vulnerabilities, https://doi.org/10.1787/4154cdbf-en; European Commission (2025[114]) Commission approves EUR 920 million German State aid measure to support Infineon in setting up a new semiconductor manufacturing facility, https://ec.europa.eu/commission/presscorner/detail/en/ip_25_557; European Commission (2025[115]), EU launches InvestAI initiative to mobilise EUR 200 billion of investment in artificial intelligence, https://ec.europa.eu/commission/presscorner/detail/en/ip_25_467 ; Intel Corp (2025[116]) Intel and Trump Administration Reach Historic Agreement to Accelerate American Technology and Manufacturing Leadership, https://www.intc.com/news-events/press-releases/detail/1748/intel-and-trump-administration-reach-historic-agreement-to; (UK Department for Science, Innovation & Technology (2025[117])AI Opportunities Action Plan,https://www.gov.uk/government/publications/ai-opportunities-action-plan/ai-opportunities-action-plan#lay-the-foundations; OECD (2025[118]) Policy Brief - Recent trends in semiconductor subsidies, https://www.oecd.org/en/publications/recent-trends-in-semiconductor-subsidies_5e91af33-en.html; European Commission (2022[119]), A Chip Act for Europe, https://www.decision.eu/wp-content/uploads/2022/09/A-Chips-Act-for-Europe-Commission-Staff-Working-Document.pdf; Reuters (2025[120]), South Korea unveils USD 23 billion support package for chips amid US tariff uncertainty, https://www.reuters.com/technology/south-korea-unveils-23-billion-support-package-chips-amid-us-tariff-uncertainty-2025-04-14/; Tomoshige (2022[121]) Japan's Semiconductor Industrial Policy from the 1970s to Today, https://www.csis.org/blogs/perspectives-innovation/japans-semiconductor-industrial-policy-1970s-today; Yamaguchi and Kihara (2024[122]) Japan unveils USD 65 billion plan to aid domestic chip industry, https://www.reuters.com/world/japan/japan-propose-65-billion-plan-aid-domestic-chip-industry-draft-shows-2024-11-11 /; The Economist (2025[123]) Japan storms back into the chip wars , https://www.economist.com/asia/2025/08/21/japan-storms-back-into-the-chip-wars; France (2025[124]) Make France an AI powerhouse, https://www.elysee.fr/en/emmanuel-macron/2025/02/11/make-france-an-ai-powerhouse.

In addition to government investments another form of intervention which has increasingly featured in markets are trade barriers including tariffs and export controls. For example, there has been an evolving set of restrictions in advanced chips and manufacturing equipment such as those imposed by the US on China.3 Recently China has also placed restrictions on firms abilities to use Nvidia’s chips in an effort to boost its domestic industry (Leng, Wu and Bradshaw, 2025[125]).

The interventions, while often justified by national security or industrial policy goals, can create trade barriers that can distort markets, fragment supply chains and entrench dominant players. The implications on competition are uncertain. By preventing companies from competing in certain jurisdictions, export restrictions can reduce global economies of scale and efficiency as well as reducing options in the affected jurisdiction. However, such actions are implemented by governments balancing other geopolitical considerations. And in the long-term in jurisdictions with limitations on imports, domestic industries may begin to develop in the market gap (if the affected market is sufficiently large). This could result in additional global competitors or simply distinct supply chains in separate geographic markets.

As well as direct investments and trade policies, government regulations and legislation more broadly will impact the way competition develops in the AI supply chain. This is likely to be particularly the case at the data centre level, as they require large volumes of land, construction, energy and water. these inputs may be subject to a range of local and national legislation for other policy reasons. Such regulations may impact the growth of data centres and the provision of accelerated compute, benefitting the companies which have already made investments in these assets over new entrants (Meyers and Bourreau, 2025[126]). In addition, complex regulation can provide other advantages to the biggest providers as they may be better able to manage the regulation due to their scale and involvement in the establishment of the regulations.

Notes

← 1.  The reported dispute focuses on Arm’s instruction set architecture, the fundamental code enabling software communication with processors. The dispute is driven by Arm’s shift from an open licensing model, under which chipmakers like Qualcomm could develop custom processors based on Arm’s designs, to a more restrictive approach favouring its own chip products. Qualcomm argues that this move threatens competition in the semiconductor industry, which has relied on Arm’s technology for over two decades (Swain, 2025[168]).

← 2.  The extent of vertical integration is much more limited upstream production processes beyond chip design, with firms not tending to get into wafer fabrication or chip packaging. In addition, even for chip design the entry is typically conducted with established chip design companies to help them design AI accelerators, primarily Broadcom and Marvell.

← 3.  For example, Nvidia's AI chips have been a key focus of US export controls, with US officials working to keep the most advanced chips from being sold to China as the US tries to keep ahead in the AI race. After those controls were implemented, Nvidia began designing chips that would come as close as possible to US limits (Reuters, 2025[176]).