By Dr. Andrea Coscelli, Emily Chissell, Dr. Nitika Bagaria and Tega Akati-Udi

Introduction

Recent generative AI (“gen AI”) tools can perform a range of tasks, and financial market trends indicate that these tools are set to disrupt industries such as data, advertising, and software. Following Anthropic’s launch of its powerful new AI model, Opus 4.6 and the release of marketing, legal and finance features for its Claude Cowork tool,^[1] the share prices of several existing firms in these sectors have recently dropped sharply.^[2] Among others, LSEG’s share price fell by 12% within a week,^[3] Thomson Reuters’ share price declined 20% within seven days,^[4] and Publicis Groupe experienced a loss of about 10%.^[5]

Although the long-term impact of gen AI tools on these industries remains uncertain, it is expected that such tools will continue to reshape competitive dynamics as investment in gen AI initiatives grows across various sectors. These dynamics are also highly relevant to the CMA’s ongoing review of how merger efficiencies are assessed.^[6]

This article sets out an economic framework for understanding the impact of gen AI tools on competition in different sectors and how merger review practices might adapt to these new dynamics. First, we consider the growing use of gen AI tools across non-digital industries (investments in AI by tech firms have already been widely covered^[7]). Next, we draw on economic theory to identify conditions under which gen AI tools are likely to increase competition (e.g., lowering entry or expansion barriers). Last, we discuss how to practically reflect these dynamics in a merger assessment.  

The increasing use of gen AI tools in non-digital industries

ISG’s 2025 report on the state of enterprise AI adoption finds that AI adoption is accelerating across a diverse set of use cases.^[8]

The use cases for gen AI vary by industry. For example, the healthcare industry is currently focusing on clinical support (e.g. clinical note generation). Whereas manufacturing and retail industries are using gen AI to improve demand forecasting. Beyond AI chat assistants, which many customer-facing sectors are deploying, other gen AI use cases include: mortgage agents in the real estate industry (Rightmove);^[9] document automation for regulatory submissions and drug discovery in pharmaceuticals (Novo Nordisk);^[10] volume prediction and warehouse coordination in the supply chain and logistics industry (DHL);^[11] assistants in the banking industry (UBS, Dave, Bunq);^[12] and problem debugging on shop floors in the manufacturing industry (BMW).^[13]

In addition, the use of gen AI is transforming how consumers find and access products, and the emergence of new substitutes.^[14]

How gen AI tools are likely to change competitive dynamics – an economic framework

Gen AI tools will affect markets and competition in different ways, and understanding the economic factors driving these changes can help evaluate potential effects. If the adoption of gen AI tools is widespread, it could drive market-wide gains. In other cases, it may strengthen incumbents, particularly those with complementary assets. Alternatively, it could also disrupt incumbents by lowering entry and expansion costs.

Economically, the competitive impact of gen AI tools will depend on its effects on business cost structures and how it alters consumer preferences and behaviour. On the cost side, gen AI use cases have the potential to affect three key elements:

1.     Fixed operating costs (relevant for entry and expansion analysis): gen AI tools can automate a variety of administrative, support, or analytical tasks, reducing the minimum output scale required for viable operation.

2.     Marginal costs: gen AI tools may lower variable production costs through optimisation, logistics, or pricing algorithms.

3.     Strategic (endogenous) fixed costs: gen AI tools may alter the cost of discretionary investments (i.e., endogenous sunk costs) such as R&D, design, and marketing.

By lowering fixed operating costs, adoption of gen AI tools may increase entry or enable expansion

With gen AI tools, fixed overheads required to enter an industry, such as accounting, are likely to decline. In economic terms, this lowers the minimum efficient scale needed to enter and operate profitably, ceteris paribus.^[15] This encourages market entry (and/or expansion), intensifying contestability – a merger in this competitive environment will likely be less concerning.^[16]

That said, a critical consideration is whether these cost reductions are accessible to all firms. If every firm in the industry can adopt and enjoy similar cost savings, the playing field levels.^[17] However, if leveraging gen AI depends on specific pre-existing capabilities (such as proprietary data), then the benefits might be asymmetric – reinforcing existing market leaders rather than enabling new challengers (or in some cases favouring challengers from adjacent markets). This dynamic could drive AI-motivated merger efficiencies, as firms combine complementary assets to compete effectively with players already able to exploit gains from gen AI tools.

By lowering marginal costs, adoption of gen AI tools can lead to consumer welfare gains

In sectors with repetitive or data-intensive production processes, marginal costs – the incremental cost of producing one more unit – may be reduced by gen AI efficiencies. As with fixed-cost reductions, if gen AI tools lower marginal costs symmetrically across firms in a market, the competitive landscape doesn’t change, and consumers benefit from lower prices.^[18] However, if such cost reductions are asymmetric, firms may strategically lower prices to deter entry.^[19]Therefore, it will be important to consider whether the benefits of gen AI tools are broadly accessible and whether the market structure makes such pricing profitable.^[20]

Adoption of gen AI tools could spark a strategic R&D investment race – the shape and nature of this race determines the impact on competition

Gen AI tools could significantly reduce R&D costs – for example, by automating certain parts of product design, drug discovery, or software testing.

R&D expenditures are discretionary fixed costs, meaning companies choose how much to invest.^[21] These costs are also often sunk – once spent, they cannot be recovered on exit. John Sutton's (2006) work on endogenous sunk costs provides a helpful framework for evaluating how industries evolve when R&D costs decline.^[22]

Sutton's theory predicts that in industries where product quality strongly drives demand, when the cost of R&D declines, firms have an incentive to increase innovation spend to capture a larger market share. An innovation arms race ensues: creating the need for even larger investments (and scale) to enter the market. This means new firms can enter only if they make a comparably significant R&D investment – raising barriers to entry.

The competitive outcome will ultimately depend on the nature of the industry and customer behaviour: the extent to which gen AI tools improve product quality, how much consumers value those gains, and whether there are diminishing returns to R&D beyond a certain point. If returns from increasing R&D spending start to level off, the arms race abates, and entry costs are limited.

In the context of a merger, if a market is prone to an AI-fuelled arms race, authorities might be concerned that a merger could further increase barriers to expansion. But there could also be benefits from two firms merging to achieve sufficient scale to invest in the required R&D, thereby increasing competitive pressure on rivals, particularly if those rivals already have the required scale to invest.^[23]

The use of gen AI tools could have important implications for merger assessments

While the use of gen AI tools remains nascent and uncertain, their adoption by businesses and customers is growing, and it is likely to increasingly shape competition across many industries.^[24] Despite its potential impact, there has been relatively limited consideration of gen AI use cases in merger assessments outside of digital industries.^[25]

Foreseeing the effects of gen AI tools on firms – particularly within the confines of a merger investigation – will be challenging, and there will be uncertainties. Nonetheless, it is an area that is likely to be increasingly important for future merger assessments. It is highly relevant to the ongoing debate on innovation and investment, as seen in the recent European Commission Consultation on the merger guidelines.

The areas of merger assessment that could be particularly impacted are: market definition;^[26] closeness of competition and loss of competition, barriers to entry/expansion^[27] and merger efficiencies.

Dynamic competition (driven by gen AI tools) may be demonstrated by event studies such as the analysis of changes in share prices when new gen AI tools are launched, indicators such as startup entry, venture capital investment, declining fixed costs, significant AI investment by incumbents (signalling an expectation of future competition), or customer willingness to adopt AI-enabled alternatives.

While agencies are increasingly aware of issues related to investment, innovation, and dynamic competition, it is likely that AI-related arguments presented by merging parties will be given careful scrutiny. This makes it critical that there is compelling, detailed, industry-informed evidence that provides a nuanced narrative on the adoption of gen AI tools grounded in a robust economic framework.

______________________________________________________________________________

‍

^[1] https://www.anthropic.com/news/claude-opus-4-6

^[2] https://www.ft.com/content/fd134065-c2c6-4a99-99df-404d658127e6#comments-anchor; 2026.06: SaaSmageddon and the Super Bowl

^[3] Calculated as the change in the daily closing price of 29^th Jan and 4^th February (the same period as the estimate for the other companies). See also: https://www.telegraph.co.uk/business/2026/02/07/londons-stock-market-bet-big-on-data-its-now-unravelling/?icid=return_to_article

^[4] https://simplywall.st/stocks/ca/commercial-services/tsx-tri/thomson-reuters-shares/news/is-it-time-to-reconsider-thomson-reuters-tsxtri-after-the-sh

^[5] https://www.ft.com/content/fd134065-c2c6-4a99-99df-404d658127e6#comments-anchor

^[6] https://connect.cma.gov.uk/call-for-evidence-merger-efficiencies-review

^[7] See, for example, https://www.ft.com/content/d503afd5-1012-40f0-8f9d-620dcb39a9a2

^[8] ISG, State of Enterprise AI Adoption, September 2025.

^[9] Rightmove noted it is currently working on 27 different AI projects in order to build a “leading digital ecosystem for the whole moving experience.”

^[10] https://www.novonordisk.com/content/dam/nncorp/global/en/investors/irmaterial/cmd/2024/P10-Data-Science-and-AI.pdf; Exscientia Uses Generative AI to Reimagine Drug Discovery; How Recursion Pharmaceuticals is Using AI to Revolutionize Drug Discovery | by Devansh | Medium; In AI-enabled drug discovery, there might be more than one winner | TechCrunch

^[11] https://www.dhl.com/global-en/delivered/innovation/questions-about-artificial-intelligence-in-the-workplace.html; https://group.dhl.com/en/media-relations/press-releases/2025/dhl-boosts-operational-efficiency-and-customer-communications-with-happyrobots-ai-agents.html

^[12] For example, UBS recently began using Open AI and Synthesia models to create AI-generated scripts and avatars of its analysts for video content. Source: https://fortune.com/europe/2025/05/20/ubs-bank-ai-generated-video-avatars-analysts/. See also: Neobank Dave’s new chatbot achieves 89% resolution rate, CEO says | Banking Dive; bunq becomes the first AI-powered bank in Europe as it unveils its own GenAI platform | bunq Newsroom; Meet Finn — bunq’s new GenAI chatbot – TechReviewers.net.

^[13] BMW created a digital AI assistant, “Factory Genius” to help employees with equipment problems on the manufacturing floor. Source: https://www.press.bmwgroup.com/global/article/detail/T0451072EN/%E2%80%9Cjust-ask-factory-genius-%E2%80%9D%3A-how-ai-helps-maintain-manufacturing-equipment

^[14] For example, Booking.com noted in their response to the European Commission’s consultation on merger guidelines, gen AI tools are transforming how customers search for, plan and book travel, increasing the options available for customers. Source: https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/14596-Merger-guidelines-review/public-consultation_en

^[15] In other words, other things being equal.

^[16] This possibility is echoed in many responses to the European Commission's consultation on merger guidelines. Source: https://ec.europa.eu/info/law/better-regulation/have-your-say/initiatives/14596-Merger-guidelines-review/public-consultation_en

^[17] In this scenario, it is also possible that smaller and/or new ‘AI-first’ firms are better equipped to adapt and adjust to gen AI use cases than larger incumbents, weighed down by legacy infrastructure (such as existing contracts or physical assets).

^[18] This depends, however, on whether cost savings are passed on to customers.

^[19] This is likely in markets with imperfect competition where prices may be used strategically. In perfectly competitive markets, firms are price takers.

^[20] The conditions required for predatory pricing are substantial market power for the incumbent, ability to sustain short-term losses, high barriers to entry for competitors, and the likelihood of recoupment. Source:
“Exclusionary Practices”, The Economics of Monopolisation and Abuse of Dominance, pp. 14 – 125.

^[21] Discretionary fixed costs are unlike exogenous fixed costs which are technologically determined setup costs required for industry entry.

^[22] John Sutton, “Sunk Costs and Market Structure”, MIT Press, 1991. Sutton’s work suggests that the effects of sunk costs in promoting concentration may be greatest in differentiated product industries.

^[23] Equally, a vertical merger may create positive efficiencies and help downstream firms to overcome the costs of R&D, as long as there are no incentives and the ability to foreclose rivals.

^[24] One caveat is that the effect of gen AI tools could run up against the typically short merger assessment timeframe (2-3 years). This makes it even more critical that merging parties present compelling and detailed industry-informed evidence.

^[25] Beyond tech company mergers/partnerships, at the time of writing, the most recent and only merger assessment in the CMA which mentions gen AI use cases (albeit briefly) is Getty/Shutterstock. Source: https://www.gov.uk/cma-cases/getty-images-slash-shutterstock-merger-inquiry

^[26] Market definition could become increasingly important and complex, as AI-driven products blur the boundaries between existing markets or create entirely new submarkets.

^[27] Here it will also be necessary to show that any entry/expansion benefits will be market-wide, long-term (i.e., sustained) and will not be deterred by restrictive access to inputs for gen AI tools.

‍

‍