On January 25, 2024, the Federal Trade Commission (FTC) hosted a summit focused on the impact of artificial intelligence (AI) on consumers and competition in the technology sector. Comprising three panel discussions and related commentary from the Commissioners, the Summit focused on the need to promote an open, competitive landscape while protecting consumer safety and data privacy. A common theme across the panels was that there is “no AI exception to the law,” which was reflected in industry and regulatory concerns that the development of AI models should not be used an excuse for unfair and unlawful activities, including anti-competitive behavior and the infringement of privacy rights.
Panelists expressed a diversity of opinions from the perspective from academia, industry, and government to examine the challenges new market entrants, consumers, and regulators currently face in the industry. Specifically, the first panel discussed the current state of the supply chain through the various layers of technologies used in AI applications. The second panel looked at data acquisition methods for training models and how we should think about personal data rights. Finally, the third panel highlighted the importance of transparency and accountability in training models and collecting data for business and legal purposes. This article is the first in a three-part series summarizing each of the three panel discussions at last month’s FTC AI summit.
Stacked Up: How Concentration Can Affect Competition
Training and deploying AI applications like ChatGPT can be an extremely resource-intensive process. The layers of hardware and software that are required are sometimes referred to as the “AI stack.” Each layer of the AI stack builds on the layers below. At the base of the AI stack is the silicon—the individual semiconductors that, when combined, form hardware accelerators. These accelerators, known as Graphics Processing Units (GPUs), supply the raw processing power that is needed to train AI models on millions of gigabytes of data, and are primarily manufactured by Nvidia. Because of this, AI models require a significant GPU infrastructure on a network of servers and data storage hardware to host them. Sitting atop this AI stack are applications, such as ChatGPT, that allow users to interface with the models.
According to the panelists, one primary obstacle to AI marketplace competition is the high concentration of a small number of established companies that create and support the silicon and cloud layers (e.g., chip designers, manufacturers, and lithographers) for hundreds of companies building consumer applications. This poses significant barriers to entry for potential new competitors in the market, which can have the effect of stifling competition and innovation. As summarized by panelist and Chief Cloud Economist at the Duckbill Group, Corey Quinn, “all roads lead to Nvidia.”
Concentrated Challenges
The panelists identified several challenges that the concentration of market power among AI tech companies creates for innovation and competition:
Volatility and Surveillance: Panelists expressed concern with the small number of players in the semiconductor and cloud infrastructure markets, which they indicated could lead to volatile pricing and quality issues. Panelists also expressed concerns that a few dominant companies are vertically integrated across multiple layers of the AI stack—i.e., “hyperscalers”—and these companies are well positioned to leverage their unique access to sensitive customer information about manufacturing hardware and/or data. For example, a fabless semiconductor startup might be forced to transfer highly confidential specifications to a manufacturer that designs and fabricates its own chips and may use that information to compete with the startup.
Barriers to Entry: Panelists also remarked that a substantial investment is required to train AI models, which, coupled with the scarcity of essential resources like GPUs, creates a high barrier to entry for startups and new market entrants. One panelist also indicated that the lack of transparency around how such resources are distributed complicates market access and stifles potentially innovative solutions in favor of companies that have connections to the few dominant firms.
Customer Mobility: The panelists also suggested that the lack of competition in the semiconductor and cloud layers of the AI stack limits customer mobility and may hinder the development of competing technologies. Some noted that established players can impose onerous contract terms to discourage customers from moving to a competitor (if any) or starting a competing business, and that the significant costs of switching providers, plus the benefit of an integrated product offering, can exacerbate the issue of portability and deter the adoption of alternative products.
Regulatory Perspectives and Solutions
The second part of the discussion focused on potential regulatory intervention and solutions to encourage competition and innovation throughout the AI stack.
Increased Transparency: Panelists called for greater transparency around the distribution of scarce computing resources, the mitigation of preferential treatment, and the promotion of fair access. They suggested implementing non-discrimination regulations that require service providers to treat customers equally as one way to accomplish greater transparency and equity.
Break Up Vertical Integration: Panelists expressed concerns that the same few players would keep a stranglehold on AI innovation so long as they maintained the advantages of vertical integration, and that the FTC could use traditional tactics like structural separation and merger guidelines to force separation of existing monopolies while restricting future mergers and/or vertical integration within the AI industry. Some panelists also claimed this would give startups more confidence in their suppliers.
Coming Up
The first panel made it clear that they believed more competition is needed to encourage healthy growth and a positive direction for the AI industry. But not all panelists agreed. Some rejected the idea that there is insufficient competition and proposed ways they thought would better help AI companies foster innovation and mitigate risks for companies incorporating AI into their business. Beyond the challenges present in the AI stack supply chain, the other panels focused on data privacy concerns, consumer safety concerns around marketing, and the specific needs of AI consumer application startups and other businesses looking to leverage AI technology. Finally, regulators from the FTC, Consumer Financial Protection Bureau (CFPB), and Consumer Protection Bureau (CPB) identified actual and prospective processes and solutions intended to promote fairness and competition in the market for AI technologies.