Defining AI Markets
As a starting point, the Bureau’s paper identifies three categories of AI markets involved in the production of an AI product or service:
- Markets for AI infrastructure: the supply of inputs for AI tools and development, such as compute (or the computational resources required to develop AI technologies), AI chips, supercomputers, data centres and data;
- Markets for AI development: the supply of AI models, algorithms or architecture that can be used for or integrated into a final product or service; and
- Markets for AI deployment: the supply of final AI products or services.
All three categories of AI markets are closely related; indeed, the discussion paper notes that vertical relationships “play a significant role in AI markets” with vertically-integrated firms operating across two or more categories.
That being said, even within the same broader category of market, different competitive dynamics may emerge between individual markets. For example, the discussion paper notes that compute is commonly accessed through a market participant’s own data centre, a publicly available supercomputer, or through a cloud compute provider. Data centres are costly to build and operate, and a limited number of large technology firms are the main suppliers of cloud compute services, suggesting the supply of compute is limited. On the other hand, data (another key input for AI development) has generally been available to AI developers in sufficient volumes from public sources, suggesting it is readily accessible at relatively low cost. However, the discussion paper notes that some research indicates that public data sets available for training large language models (LLMs) may be exhausted in the coming years, increasing the demand for proprietary data sets and suggesting that the supply of data may be more limited in the near future.
Competitive Impacts of AI
Having distinguished categories of markets at each stage of the AI supply chain, the Bureau turns to the potential competitive impacts of AI. The potential concerns identified in the paper can be broadly grouped into three categories:
i. Traditional competition concerns arising in new AI markets (e.g., barriers to entry caused by limited access to compute or data)
For example, the paper describes several scenarios in which the structure of AI markets may raise traditional competition concerns regarding anti-competitive conduct. The paper observes that AI infrastructure markets for cloud compute and AI chips exhibit high concentration and significant upfront costs. As large AI firms may operate across multiple AI markets (e.g., supplying inputs and developing AI models and algorithms), start-ups in the AI development market may be required to access necessary inputs from their competitors. This structure potentially creates opportunities for exclusionary conduct. Within the AI development market, large AI firms may also have an advantage in achieving economies of scale and scope, limiting entry for new participants. Finally, the paper describes potential data network effects in the AI deployment market, whereby the more user data that is collected by an AI system, the more its technology improves. Taken together, these structural considerations could lead to concentration and entrenched market power in each of the categories of AI markets identified in the paper.
ii. The use of AI as a new tool to facilitate traditional anti-competitive conduct (e.g., using AI tools to engage in misleading telecommunications campaigns)
The paper also discusses the potential use of AI tools to identify opportunities for, and to implement, predatory conduct. For example, the paper suggests AI tools could be used to identify customers at risk for switching and target them with below-cost pricing. In theory, this strategy would minimize the loss associated with the predatory conduct, potentially increasing the incentives to engage in such conduct. The paper raises a similar scenario in the context of tying and bundling, whereby AI could potentially be leveraged to target bundling and tying strategies to customers identified as at risk of switching.
In turning to the potential use of AI to implement or sustain cartel agreements, the paper suggests that AI may lead to increased market transparency (e.g., due to the ability to gather and assess pricing data) and increased frequency of competitor interactions (e.g., due to greater ease in detecting and responding to pricing changes), potentially increasing the ability and incentive for cartels to form. The paper again suggests that AI technology could be used to facilitate anti-competitive conduct, in this case by automatically implementing agreed upon prices or strategies, monitoring competitive behaviour and detecting deviations.
iii. Arguably novel concerns arising from the unique capabilities of AI technologies (e.g., the ability of AI decision-makers to autonomously enter prohibited agreements with competitors)
While more traditional competition concerns define the majority of the scenarios raised in the discussion paper, the Bureau does briefly reference the potential for AI tools to be employed as decision-makers, and to autonomously employ predatory, exclusionary or discriminatory conduct or to facilitate tacit collusion. To mitigate this risk, the paper suggests that proper human oversight – to identify and correct potentially anti-competitive decisions – may be required for firms implementing AI tools.
* * *
The Bureau’s discussion paper solicits feedback on, among other things, “relevant or informative market study topics, specific to the AI sector or a portion of the AI sector”. This question could signal further analysis of the sector in the near future, particularly in light of the Bureau’s recently expanded powers to pursue formal market studies (including the ability to seek a court order to compel market participants to produce documents and/or written responses to advance the market study).
In any event, we can expect the Bureau to be alert to potential anti-competitive conduct as AI-related markets continue to evolve. In a speech to the Canadian competition bar in October 2023, the Commissioner of Competition (the head of the Bureau) offered a call to action in which he highlighted the need to “keep pace” with emerging technologies, including AI and machine learning; the “important questions that need to be asked about the entrenchment of dominant players who control the critical inputs that fuel AI”; and the “very real risk posed by anti-competitive mergers or deliberately unethical conduct of these advanced tools to deceive consumers”. The discussion paper seems to fit squarely into this call to action, offering a roadmap to where the Bureau may increase its engagement with AI markets.
