All regulators, including the competition authorities, have been grappling with what to do about AI for some time now. However, the speed at which AI is developing poses constantly new challenges and the complex stack of activities involved means that its regulation is a moving target. There is also added pressure as, after a decade of experience in regulating digital activities, regulators are keen to be on the front foot with regards to AI by taking advantage of the new ex-ante powers granted under new digital markets legislation. In practice, this means that businesses in the AI ecosystem need to be vigilant and alert to the risks that regulatory interventions present.

Competition authorities' lighter-touch fact-finding exercises to understand AI markets better and be ahead of the curve have begun to see more targeted action being taken. For example, in the US, the Department of Justice (DOJ) signalled that it is investigating Nvidia, whose high-end chips are a critical input for computing power required for the development of AI foundation models (FMs) – large AI models trained to be used as the basis for different types of tasks and operations – while the Federal Trade Commission (FTC) is probing Microsoft's investment in OpenAI. The European Commission (EC) is also scrutinising AI partnerships between companies such as Nvidia/Run:ai that it has recently cleared. Google's agreement with mobile device manufacturers such as Samsung, for the pre-installation of its AI model, Gemini Nano, on mobile devices is also on the EC's radar. The UK Competition and Markets Authority (CMA), which signed the joint statement on competition in generative AI FMs and AI products alongside the DOJ, FTC and EC in July last year,¹ is also trying to keep pace with the rapid progression of AI and has initiated a number of investigations in this sector. Although more targeted, these inquiries do not necessarily signal "stronger intervention", in particular considering how swiftly some of them, such as Amazon/Anthropic and Microsoft/Inflection, have been closed.

The results of fact-finding exercises, along with public statements and investigations to date, indicate that the primary concern of regulators is whether incumbent big tech companies could gain a significant advantage in AI markets. They are also considering how to address this potential issue without stifling innovation. The CMA, given how active it is in the AI field, is likely to be one of the regulators leading the way in regulating AI and influencing other jurisdictions. In this context, this article sets out the CMA's approach to AI considering its recent AI projects and interventions to help businesses understand its current priorities and what to expect in the near future.

CMA's review of AI FMs

The CMA published its initial paper on AI FMs in September 2023 and followed it up with an update in April 2024 (Update Paper). As defined above, FMs are like foundations of buildings on which various AI applications (from search engines to social media or productivity software) can be built. For example, GPT, an FM developed by OpenAI, is the basis of the blockbuster generative AI chatbot (software application) ChatGPT. OpenAI's GPT has also been incorporated into other software applications, such as Copilot, which Microsoft developed.

The CMA has looked at the AI value chain holistically and analysed the upstream market of the development of FMs as well as downstream markets where FMs are incorporated into consumer-facing applications.² This analysis showed that effective competition in the upstream market is critical for protecting competition in downstream markets as these markets are likely to produce positive outcomes if downstream players can easily switch between FMs. So, the question is, what factors may affect FM developers' ability to compete effectively in this market? The CMA answered this question by determining key inputs that any FM developer needs to access (either as their internal resources or via partnerships with third-party providers) to enter and sustain its position in the FM market:

• Computing power: FM developers need access to significant computing power to train FMs. This encompasses the availability of AI accelerator chips, which are in high demand and supplied by a few chipmakers, including Nvidia, controlling the vast majority of the market share.
• Data: FM developers need to access data, as FMs are trained on vast quantities of data to build foundational knowledge. Also, a high volume of data is required to refine the FM for a specific application.
• Technical expertise: FM developers need to attract highly skilled research scientists and engineers in short supply.
• Capital: Required to access the three inputs above.

Although large tech companies typically have greater access to these "key inputs", this does not mean their market power in digital markets will automatically give them an advantage in AI markets. For example, it is unclear how much proprietary data smaller FM developers need to access or whether "synthetic data", which is data that is artificially "generated from original data and a model that is trained to reproduce the characteristics and structure of the original data",² can be a substitute. Also, FM developers lacking substantial in-house computing resources typically contract with incumbent cloud service providers as an alternative way to access the essential input. However, if cloud providers have existing market power over computing, that may be a concern if the incumbent service provider discriminates between customers without objective justification, or tries to influence its AI partner with a view to eliminating it as a possible source of future competition. Unsurprisingly, the CMA is vigilant about a knock-on effect of any ability to control an essential input in the AI ecosystem.

Tools available to the CMA

Traditional powers such as the prohibition of anti-competitive agreements, abuse of dominance under the Competition Act 1998 and merger control are available to the CMA to monitor the AI markets. For instance, in an ongoing market investigation, the CMA is examining competition in the provision of public cloud service infrastructure services, which will include the assessment of the potential impact of FMs in this market.⁴ It is also examining AI arrangements, such as ongoing Microsoft/OpenAI and recently closed Alphabet (Google LLC)/Anthropic, under the merger control rules to understand how these can affect competition in the AI ecosystem.

The CMA is now also equipped with new powers entrusted by the UK's Digital Markets, Competition and Consumers Act (DMCCA) that came into force on 1 January 2025, which will likely be game-changing in their ability to tackle AI-related concerns. In the Update Paper, the CMA signalled that, in terms of the designation of companies with significant market status (SMS), they may prioritise digital activities that are critical inputs for developing FMs. This means we may have firms controlling key inputs for developing FMs designated as SMS as early as 2025. Indeed, the CMA has signalled that it considers the digital markets powers an efficient and effective way to address digital markets issues by closing parallel investigations into Google's Play Store and Apple's App Store (under existing competition powers) ahead of the DMCCA rollout, which will enable them to address competition concerns in these cases "more holistically".

The EU's Digital Market Act (DMA), which came into force two years ago, and the DMCCA aim to achieve the same purpose but differ in design. On paper, although their application in practice is yet to be seen, DMCCA rules are more flexible than the DMA, giving more leeway to the CMA to bring big AI players under the DMCCA digital markets regime. You can find more about the new regime and its comparison with the DMA here. Furthermore, SMS will be subject to new mandatory and suspensory merger reporting requirements to prevent so-called "killer acquisitions", which can directly affect strategic investments in the AI ecosystem.

What to expect

The CMA has stated in the Update Paper that it is "concerned that the FM sector is developing in ways that risks negative market outcomes", so those involved in AI markets should expect stronger intervention to come, particularly given that the DMCCA is now in force. That said, historically, the UK's competition watchdog has adopted a collaborative and participative approach involving not only those under investigation but also stakeholders in its enforcement in the digital markets – this has also been reflected in the DMCCA.⁵ It is therefore likely that the CMA will continue to hear the concerns and needs of businesses on the ground to come up with solutions to foster resilient AI markets without disrupting market dynamics and interfering with what already works well. Accordingly, in the UK "more invasive" AI interventions will probably be kept to a minimum where they are absolutely necessary to ensure that the markets are not developing in ways that risk negative market outcomes.

1. Also, see the EC, CMA, DOJ and FTC's joint statement on competition in generative AI FMs and AI products published on 23 July 2024 here.↩
2. See the Update Paper Figure 2 on page 9 for the illustration of the big tech firms in the FM value chain.↩
3. "Synthetic Data" accessed 3 October 2024.↩
4.  CMA, "Cloud services market investigation"accessed 3 October 2024.↩
5.  See CMA, "How the UK's digital markets competition regime works" accessed 10 January 2024, where it is confirmed that the process for investigations and interventions under the DMCCA will be participative and transparent.↩