The recent success of generative artificial intelligence (GenAI) tools capable of generating human-like content has shaken digital markets and started a new technological race. GenAI is powered by foundation models: large and complex AI models trained on vast amounts of data. Its impact is estimated to boost global GDP by 7% in the next decade according to a recent study [1]. Competition authorities across Europe have taken a forward-looking approach to understanding the market in its nascent stage and ensuring competition and innovation are preserved both presently and in the long run.
This white paper provides a preliminary review of current developments in the GenAI space and the emerging implications for competition. Further research is needed to shed additional light on the evolving competitive situation in GenAI, as well as any case-specific matters, particularly as this is a fast-evolving sector.
We have reviewed the research literature and identified three potential sources of competition concerns affecting GenAI markets:
Our assessment of current market developments suggests that there are no evident signs of immediate competition concerns, with a number of new entrants present with diversified products and business models. It remains important to monitor and pre-empt future foreclosure concerns, which can be assessed under Article 102 or the Digital Markets Act in Europe.
At this stage, the market appears dynamic, with no or little sign of insurmountable barriers to entry
The number of foundation models launched is high and continues to increase with one third of more than 250 models available launched since August 2023 according to a database kept by Stanford University [2]. There are many open solutions available to non-integrated GenAI developers. New firms with different business models and degrees of specialisation compete head-to-head or even outperform established firms according to well-known rankings. More than 13 firms in GenAI have already achieved the coveted unicorn status.[3] Among those, OpenAI leads the GenAI race, while the recent European startup Mistral already established itself as a key competitor. A large influx of venture capital investments, with an almost fivefold annual increase to EUR 20 bn in 2023, allows startups to access costly inputs (thus reducing any barriers to entry) and signals that investors are confident in GenAI markets’ competitiveness. Models are becoming increasingly smarter and less computationally demanding. However, there is uncertainty about whether inputs may become less accessible in the future. Some European GenAI startups face challenges in growing, and highlight regulatory costs (e.g., the AI Act).
Depending on their design, partnerships between large cloud providers and AI startups may give rise to competition concerns
Vertical collaborations between large digital players and GenAI startups are common and generate efficiencies, as they allow AI startups to access highly specialised hardware and computing power, additional support and investments.
However, if a larger partner uses its market power to exercise decisive control over a startup or gain privileged or exclusive access to its technology, this may harm competition across the value chain or may ultimately remove or dampen potential competition from the startup itself in the Gen AI or related markets. Any competitive implications will depend on the nature of the assets and capabilities brought together by the agreement, and its specific design. We find that partnerships are less likely to create competition concerns if there are a) no/limited exclusivity conditions, either in supply or distribution, and b) limited privileged access to the startup’s valuable technological assets.
Integration of GenAI solutions into the existing services of large digital players can give rise to foreclosure
GenAI applications are most valuable when used together with existing services or products. This integration of GenAI into existing services can boost innovation in adjacent markets and potentially increase the competition faced by established players. However, this integration can also give rise to potential anti-competitive foreclosure via practices such as tying, bundling, or self-preferencing. Such practices should be monitored and assessed under Article 102 or the Digital Markets Act in Europe.
[1] Goldman Sachs (2023a). Generative AI could raise global GDP by 7%.
[2] Standford University (2024a). Ecosystem Graphs for Foundation Models.
[3] Unicorn status refers to a milestone in a startup’s development when it reaches a market valuation of at least USD one billion.
This white paper was funded by the Computer & Communications Industry Association (CCIA Europe).
Download
