KFTC's Generative AI and Competition Report: A Quick Overview
January 20, 2025
On December 17, 2024, the Korea Fair Trade Commission (KFTC), the Korean competition authority, released its report on generative AI, titled "Generative AI and Competition".
This report marks the very first time the agency has transparently articulated its perspective and stance on a specific competition issue (generative AI, in this case). As I was interviewed by ai fray, the KFTC’s approach taken this time on AI issues—focusing on identifying the current situation and potential concerns—is commendable. The report systematically examines market conditions, identifies potential problem areas, and evaluates whether existing tools are sufficient to address these issues or if there are gaps that need to be filled.
The report is quite lengthy, spanning nearly 100 pages, but the key messages can be distilled into the following six points:
1. Vigilance over NVIDIA's Dominance in AI Chip Market: The report finds that NVIDIA now runs as the undisputed front-runner in the market for AI accelerator chips (para 142), where it benefits from preemption effects through its freely distributed AI programming support toolset, CUDA (paras 141, 213). Citing the French competition authority's investigation into NVIDIA’s practices, the KFTC emphasizes the need for close monitoring of actions by other agencies, particularly regarding supply restrictions, allocations via partnerships, and unfair contract terms (para 217-219). However, noting the development of in-house AI chips by major cloud service providers, it does not argue for immediate actions at this stage (paras 136, 144).
2. Caution Regarding AWS's Cloud Computing Market Leadership: In the cloud computing segment, the report observes that Amazon’s AWS (followed by Microsoft’s Azure) leads the local market (paras 156-157), benefiting from the preemption effects (paras 160, 213), while it also notes that current market dynamics show positive signs, such as the parallel use of multiple cloud services (paras 163-164), warranting no immediate enforcement action. It just very briefly mentions the potential for predatory practices through cloud credits (para 239).
3. Concerns Over Access to Data and Data Exploitation: Identifying high-quality data as an essential input (para 84) and a resource that can be depleted (paras 85-86), and recognizing that difficulties in data access may raise entry barriers (para 196), the report points out the theoretical possibility of abusive data access refusal (paras 223-224). Additionally, it highlights the need for further discussion and research, and potential institutional improvements regarding MRFTA responses to exploitative and non-consensual data collection and use, citing Germany’s Facebook case (paras 261-267, 283).
4. Expertise Poaching, Expertise Concentration, and 'Acqui-hiring': Seeing expertise as a critical input for AI development (paras 90, 197), the report raises competition concerns regarding the poaching of skilled professionals to hinder competitors' businesses (unfair competition) (paras 252-253), the concentration of expertise through mergers (para 254), and the “Acqui-hiring” practice, citing the Microsoft-Inflection case (paras 257-259). The report sees that the last concern necessitates reviewing the current merger review regime (paras 260, 283).
5. Concerns in the Foundation Models Market: The foundation models are also identified as essential components for generative AI (para 97). While recognizing the entry barriers to the foundation model market (paras 99, 198), and the competitive advantages held by some foundation model developers through partnerships with Big Tech cloud service providers (paras 105, 175-176, 180), the report also suggests that the competitive landscape could continue to evolve (paras 173, 179, 182). Meanwhile, agreements for the exclusive access and distribution of foundation models between cloud service providers and foundation model providers are noted as potentially raising competition concerns (paras 244-245).
6. Vigilance Over Vertical Expansion by Big Tech: Particularly, the report shows heightened vigilance regarding the trend of vertical expansion by major Big Tech companies, such as Amazon, Microsoft, and Google (and Meta), as they enter (almost) all segments of the Generative AI value chain, ranging from AI chips and cloud computing to foundation models and AI service sectors (paras 119-127), establishing a dual position as both essential input providers and AI service providers (para 122). It then warns of the possibility of tipping effects favoring the Big Techs (para 148). Specifically, the report notes several practices, such as bundling strategies that combine cloud computing services with access to foundation models (para 226), and integrating AI services into its existing products, as seen in Google's arrangement with Samsung for Gemini nano to be pre-installed on Samsung devices (para 232) or in Germany's Google Automotive Services case (para 234), which require close monitoring. Additionally, it emphasizes the need for merger control policy responses regarding expansions through large-scale investments or partnerships, such as the Microsoft-OpenAI deal (paras 255-260, 283), including the previously mentioned Acqui-hiring.
All in all, this report appears to be, by and large, influenced by similar reports from other agencies, as mentioned in the report itself (paras 31-38), and it does a comprehensive job of addressing the various issues under discussion.
However, I found it somewhat questionable that the report does not address collusion issues, and that it does not address access to investment (para 57), despite that it repeatedly emphasizes the enormous financial resources required for generative AI development (e.g., para 99). (The latter stood out to me, especially considering the discussions that arose in October following reports that OpenAI had asked its investors not to invest in competitors.) Additionally, I found it noteworthy that copyright issues (para 50 and footnote 18 on p.16) and personal information (data privacy) concerns are excluded from the scope, seemingly trying to delineate the boundaries of authority (see paras 267, 272). Nevertheless, while updates will likely be needed, I found this version to be a well-organized report.
One last interesting point is that I got the impression the KFTC is walking a fine line between competition policy and protectionism throughout the document. While not explicitly stated, there is a clear distinction made between global and domestic companies based on their nationality (if any).
For example, in the AI chip market, domestic companies like Rebellions (an NPU developer, based in Korea), which actually have little impact as competitors to NVIDIA, are described in detail (pp.43-44). Similarly, in the cloud computing market, the report explains the status of domestic players such as Naver (pp.50-52), alongside global players like Amazon, Microsoft, and Google (pp.53-54). Although specific figures are not provided, I believe the three global firms' combined market share likely covers more than 90% of the local market (see para 159 and Table 48). Particularly, when discussing the foundation model market, the concept of "Sovereign AI" is mentioned as one of the key points in the agency's survey results, highlighting the importance of foundation models (see para 97, Table 25, and footnote 48) (for reference, a key domestic player is Naver's HyperCLOVA X (as is stated at para 126)).
I would guess (although uncertain) that the dichotomous approach, distinguishing between global and domestic players (which are actually not that significant from a competition policy perspective), might be intended to soften any potential backlash when introducing AI competition regulation in the future (justifying the need for regulation with a narrow regulatory scope, mainly targeting foreign companies). But, as there are no specific statements, it remains purely speculative.
To sum up, the report provides a good overview of key competition issues in the generative AI sector, though it raises some questions about areas left unexplored. Based on this policy report, it will be important to continue closely monitoring how the KFTC defines and refines its competition policy and law enforcement direction in the AI industry.