After making a splash in 2023, generative AI is set face greater scrutiny this year from competition authorities.
Antitrust watchdogs in the US and Europe have learnt from a body of cases on digital markets and have acquired new powers – such as those conferred by the EU’s Digital Markets Act (DMA) – to examine the tech sector. But are these powers well suited to the new challenges posed by generative AI? And given that regulators are convinced they were slow to corral digital markets, do they now risk over-correcting with early interventions in AI?
Competition authorities have been quick to comment on developments in generative AI. Last March, less than six months after the launch of ChatGPT catapulted the technology into the headlines, the UK’s Competition and Markets Authority (CMA) launched a review of AI Foundation Models; a detailed report was published later in the year. By June the US Federal Trade Commission (FTC) was voicing “competition concerns” in relation to generative AI. Some regulators in Europe, such as the Portuguese Competition Authority, followed suit.
These debates look set to move centre stage in 2024, with antitrust watchdogs investigating both the role of competition in the supply chain that supports generative AI and partnerships between businesses that operate at different levels of this supply chain.
1. The AI supply chain: heads in the clouds?
As the CMA set out in its initial review, the supply chain for providing generative AI services consists of three key levels:
- the AI infrastructure level, which covers three important inputs – computing power, data and the expertise required to build and run AI models;
- the AI model development level, which uses the inputs from (i) to develop AI models; and
- the AI deployment level, which turns the AI models into services that consumers use or serve as inputs by businesses in the services they provide.
Much remains uncertain about how the supply chain will develop in the next year and, as a result, which competition issues will become most relevant. But one area that we can expect competition authorities to keep a close eye on is the assessment of cloud computing as a critical input for AI development.
Large amounts of specialised computing power – typically drawing on multiple graphics processing units (GPUs) – are an essential requirement for training leading AI models. Rather than buying this computing power outright, AI developers often ‘rent’ it from providers of cloud services. Some competition authorities have observed that these markets for cloud services are concentrated, with firms such as Amazon and Microsoft accounting for a high share of supply. These providers of cloud services are also active elsewhere in the AI supply chain, both in developing AI models (either on their own or through large investments in leading AI developers; more on such partnerships below) and in deploying AI services (e.g. in online search or productivity software).
The European Commission (EC) may also assess cloud services using its powers under the DMA, which considers cloud a ‘core platform service’. The EC, in its first designation last September, identified firms such as Microsoft and Amazon as ‘gatekeepers’ – but not for the provision of cloud services. It will be interesting to see whether the EC moves to revise this when it next updates its designation. Any firm named as a gatekeeper for cloud services would then need to comply with the DMA’s obligations, many of which aim to address areas of concern similar to those singled out by Ofcom and the CMA. These include vertical issues such as fair access and prevention of self-preferencing.
However, the focus on cloud infrastructure may be just the opening shot in the watchdogs’ campaign. For all the attention being paid to cloud computing services as a possible pinch point in the AI supply chain, it is the human talent that may prove decisive in the development of these markets. Some of the leading generative AI firms that are fast becoming household names have a remarkably small headcount, typically in the hundreds rather than thousands, compared to the leading tech firms that competition authorities are used to dealing with. Tellingly, OpenAI’s board fired CEO Sam Altman following a rift in November only to reinstate him after most of the firm’s workforce threatened to quit. Microsoft’s public backing of Sam Altman and its offer to hire OpenAI’s entire staff also underscored the close relationship between top-notch AI developers and the more established tech firms that digital market regulators have been scrutinising recently. This interdependence is leading competition authorities to take note.
2. The role of partnerships in the development of AI
There have been a number of transactions involving AI developers in recent years, including acquisitions by leading tech groups such as Google, Meta, Microsoft, Apple and Amazon. Some of these relationships are longstanding. For example, since finding fame for writing programs that could beat professional players of strategy board games, DeepMind has played a leading role in the development of Google’s AI products – including its flagship competitor to ChatGPT, Gemini, which launched in December. While mergers of businesses with complementary skill sets can spur innovation, authorities are nonetheless likely to be vigilant of further AI deals in the year ahead.
More recently, partnership models have emerged between AI development companies and large technology firms. Notable examples include:
- Microsoft’s investment of over $10bn in its partnership with OpenAI, the developer of ChatGPT. This makes Microsoft the default provider of computing power to Open AI and allows it to integrate OpenAI’s capabilities into its consumer products, e.g. Bing search, Microsoft Office.
- Amazon’s investment of about $6bn in a partnership with Anthropic, which it is providing with cloud services. In doing so, Amazon has replaced Google, which had previously made its own multi-billion-dollar investment in Anthropic.
These partnerships may not be out-and-out mergers, but competition authorities have already made it clear that they are keeping close tabs on how they evolve. Even before the OpenAI drama piqued widespread regulatory interest, Germany’s Federal Cartel Office (FCO) was already looking at Microsoft’s partnership with OpenAI, while the FTC, FCO and CMA were all in contact with Amazon about its relationship with Anthropic.
As the CMA’s 2020 investigation into Amazon’s investment in Deliveroo illustrated, competition authorities can intervene even if the deal only involves the acquisition of a minority stake. Against this background, and given the hopes and fears that generative AI is fanning among politicians as well as the general public, further scrutiny of AI partnerships seems all but guaranteed in 2024.
This article forms part of a wider forthcoming Frontier think-piece on themes for the year ahead in competition economics. Keep an eye out for the full article coming soon.