Competition Dynamics
Equilibrium Prices and Market Power
A key component of competition and monopoly analysis is that they rely on the benefits expected from effective competition. More specifically, certain benefits are expected in markets where there is effective competition. These are generally understood to be: allocative efficiency (as resources are directed towards the goods and services that consumers value most); productive efficiency (as firms are unable to charge prices above their costs and must therefore produce in the most cost-efficient way or risk losing customers); lower prices for consumers (as a result of allocative and productive efficiencies), innovation and more choice for consumers (as firms compete for consumers' businesses).
Editorial Note: Yacine: great framing to weave through the paper
When there is no effective competition, and especially in scenarios of monopoly or collusion, the market tends to lead to a restriction in output and therefore in higher prices. This means that the monopolistic firm (i.e. the firm with significant market power not related to its own merits), or the cartel of multiple companies that together control the market will not be constrained by the competition of other firms that could provide the same product at a lower price and can keep prices artificially high or use strategies to make switching more difficult or impossible.
Applying these concepts to the AI value chain is particularly challenging because the chain is long, complex, and spans both hardware and software. Each layer of the value chain constitutes a distinct market, yet it is closely connected to the others, creating strong inter-dependencies between firms. As discussed in section 2, the AI value chain can be broadly divided into main components of infrastructure, data and products. Most of the value is concentrated in the infrastructure layer, which is also characterised by higher levels of market concentration.
The range of products, possible pricing schemes, and the fact that most alternatives to a provider correspond to a slightly different service or development model make it even more complex. Another major complexity is that companies pay in many ways that are not direct monetary compensation - most notably by providing licensed access to valuable training data and intellectual property and data produced by users.
Undue market power is broadly defined as any situation that maintains prices higher than their equilibrium price, for example in cases where providers collude to keep prices artificially high or use strategies to make switching more difficult or impossible...
High Priority TODO: Yacine: we should also articulate concerns about divergence of costs between self-managed and centralized AI solutions. Even if centralized solutions are only marginally expensive to begin with, vendor lock in in the longer run increases market power and allows providers to maintain higher prices; in addition to centralizing flow of the data resource. This could go in either the "undue market power" or the "vertical integration" subsections, or both.
High Priority TODO: Articulate concern about locking in customers and sunk costs, short-term vs. long-term difference in costs when relying on external products with lock in potential, and what happens when switching costs increase.
In this work, we are particularly interested in self-development as an alternative to purchasing a vended service from an AI provider offering a service based on a computationally intensive model.