July 2026
TFTC Releases Policy Statement on Generative Artificial Intelligence and Competition Law
Taiwan Fair Trade Commission (hereinafter the “TFTC”) published the “Public Consultation Paper for Generative Artificial Intelligence and Competition” (hereinafter the “Consultation Paper”) on July 18, 2025. Following the public consultation from July 18, 2025 to September 7, 2025, the TFTC officially published the “Public Consultation Report and Policy Statement on Generative Artificial Intelligence and Competition” (hereinafter the “Policy Statement”) on March 18, 2026.
The key points of the Policy Statement are summarized below.
I. Generative AI Market Structure and Barriers to Entry
In the Consultation Paper, the TFTC identified the generative AI market structure into three layers: infrastructure, models, and application deployment. It cited numerous studies pointing out that model development places extremely high demands on infrastructure—such as computing resources, massive amounts of data, and specialized talent—thereby constructing significant barriers to market entry. This entry barrier is also related to a high threshold of capital investment.
In addition, the TFTC noted that many startups, unable to afford high hardware deployment costs, often secure computing resources through investments from or partnerships with Cloud Service Providers (CSPs). This practice may allow CSPs to exert substantial influence over the operations of model developers, or demand that the latest models be made available preferentially on their own cloud platforms. This pattern of integration may potentially result in the large-firms-becoming-larger effect and create switching barriers between different ecosystems in the market.
II. Policy Statement: Four Enforcement Principles
1. The Policy Statement outlines several characteristics of the AI market:
A. Companies in the AI market are increasingly competing and cooperating with one another at the same time.
B. Competition in the AI industry is leaning more heavily toward “ecosystem competition” and “competition for the market,” as highlighted in the TFTC’s “White Paper on Competition Policy in the Digital Economy” (hereinafter the “White Paper”). Defining relevant markets, measuring market power, and assessing the reasonable justifications of the conduct at issue remain challenging.
C. AI remains in its early stages of development, characterized by prominent dynamic competition.
2. The Policy Statement establishes the following four enforcement principles:
A. Issue-Driven, Competitive Harm Verification, and Evidence-Based Enforcement
a. The TFTC will adopt an issue-driven enforcement principle, meaning directing its enforcement resources on market behaviors that are genuinely relevant to market competition (such as abusing computing power for discriminatory treatment, tying, refusal to access, or self-preferencing). Issues falling under the scope of industrial policy shall be addressed by the respective sector authorities. When implementing industrial policy raises competition concerns, the TFTC will proactively advocate with relevant industrial government authorities in accordance with Article 46 of the Fair Trade Act (hereinafter the “FTA”).
b. Once an issue is identified as a competition law matter, the specific “theories of competitive harm” involved must be further clarified. For instance, when assessing whether self-preferencing or refusal to access implemented by a market participant raises competition concern, a key question may be to assess whether such conduct is used to steer users toward prioritizing its own models and computing services, thereby making it difficult for competitors to access critical cloud services, raising competitors’ costs, and causing market tipping.
c. Given the dynamic changes in market structure, the TFTC will follow an evidence-based enforcement principle and avoid a “one-size-fits-all” analytical approach.
B. Local Nexus Enforcement Principle: The TFTC stated that enforcement should be tailored to the Taiwan market conditions, including the structure and competitive landscape of its AI industry (for example, Taiwan holds a crucial position in the semiconductor sector, encompassing a complete value chain, while remaining less competitive in cloud services and large language models). There is no need to transplant foreign regulatory models in their entirety.
C. Maintaining an Enforcement Approach Centered on Market Contestability: The TFTC emphasizes that “only by maintaining the countervailing force from potential competitors — thereby ensuring that incumbents do not abuse their advantages to restrict competition — can competition be sustained and innovation enhanced.” In other words, enforcement will take into account the “potentiality” and “probability” of market entry. Furthermore, when evaluating data and computing power in the future, attention will be paid to the “qualitative” dimension, rather than relying solely on “quantity” and “scale”.
D. Strengthening the Competition Assessment under the “Rule of Reason”: The TFTC does not intend to rely exclusively on indicators of market structure to presume illegalities of competitive practices. The TFTC will evaluate the procompetitive and anticompetitive effects of disputed conduct on a case-by-case basis. Also, the dynamic relationship between the intra-ecosystem and inter-ecosystem competition would be assessed as well. The TFTC further indicated that “innovation” is the critical factor shaping and reshaping the landscape of the AI market.
In conclusion, the TFTC indicated that competition issues arising from the emergence of GenAI remain adequately addressable within the existing framework of the FTA. The current enforcement policy will be dynamically calibrated from time to time to address anticompetitive concerns arising from future AI market development.
III. Observations
1. Continuing the spirit of “big” does not equal “bad” from its White Paper, the TFTC reiterated the importance of the rule of reason, indicating that it is in no rush to characterize specific conduct as illegal per se during the early development stages of the AI industry. However, it also noted that future enforcement faces numerous challenges, including how to assess inter-ecosystem competition, competition for the market, and how to measure the role of innovation in competitive analysis—all of which rely on advancements in both theory and practice. Additionally, collaborations and alliances regarding chips, data, and computing power among upstream and downstream players in the AI value chain have become more frequent than ever, [1] which will inevitably challenge enforcement.
2. Although most industries seem to believe that the AI industry is still in its early stages and prefer that competition law not intervene too early, some insights can be gleaned from the opinions summarized in the Policy Statement. While foreign companies (primarily foreign cloud service providers and foundational model developers) generally do not believe there are anticompetitive concerns in the market, domestic Taiwanese firms express greater concern over potential restrictions on competition in the AI market, which is understandable.
3. Regarding the refusal-to-access practice, following the publication of the TFTC's Policy Statement, the European Commission recently directed an investigation into Meta for banning third-party general-purpose AI assistants from accessing the WhatsApp for Business API—essentially foreclosing competitors to ensure that only Meta's own AI assistant, Meta AI, remained accessible on WhatsApp. On June 9, 2026, the European Commission made a rare move by invoking its interim measures, ordering Meta to re-instate free access for third-party general-purpose AI assistants to the WhatsApp for Business API under the same terms and conditions that were in place before October 15, 2025, pending the completion of the investigation. [2] The action of the European Commission aligns with its position on structural regulation. While intervening with interim measures aims to prevent irreparable competitive harm, its positive and negative effects warrant further observation. The opposing views of the European Commission and Meta regarding the promotion and restriction of competition in this case are also worthy to be observed.
[1] For example, regarding unilateral conduct in the generative AI industry (see Policy Statement, Part IV, Public Comments on Potential Competition Issues Arising from the Generative AI Industry, Question 5), the TFTC summarized and disclosed four pieces of industry feedback. This included two from domestic companies and trade associations, and two from foreign foundational model developers and cloud service providers. Based on these opinions, domestic players and associations appear to believe that ecosystem integration and market behavior of large platforms could reshape the competitive landscape; conversely, foreign players believe there are no significant anti-competitive concerns as of yet.
[2] Press Release, European Commission, Commission Imposes Interim Measures on Meta to Preserve Free Access to WhatsApp for rival AI assistants (June 9, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1276?utm_source=chatgpt.com.
The key points of the Policy Statement are summarized below.
I. Generative AI Market Structure and Barriers to Entry
In the Consultation Paper, the TFTC identified the generative AI market structure into three layers: infrastructure, models, and application deployment. It cited numerous studies pointing out that model development places extremely high demands on infrastructure—such as computing resources, massive amounts of data, and specialized talent—thereby constructing significant barriers to market entry. This entry barrier is also related to a high threshold of capital investment.
In addition, the TFTC noted that many startups, unable to afford high hardware deployment costs, often secure computing resources through investments from or partnerships with Cloud Service Providers (CSPs). This practice may allow CSPs to exert substantial influence over the operations of model developers, or demand that the latest models be made available preferentially on their own cloud platforms. This pattern of integration may potentially result in the large-firms-becoming-larger effect and create switching barriers between different ecosystems in the market.
II. Policy Statement: Four Enforcement Principles
1. The Policy Statement outlines several characteristics of the AI market:
A. Companies in the AI market are increasingly competing and cooperating with one another at the same time.
B. Competition in the AI industry is leaning more heavily toward “ecosystem competition” and “competition for the market,” as highlighted in the TFTC’s “White Paper on Competition Policy in the Digital Economy” (hereinafter the “White Paper”). Defining relevant markets, measuring market power, and assessing the reasonable justifications of the conduct at issue remain challenging.
C. AI remains in its early stages of development, characterized by prominent dynamic competition.
2. The Policy Statement establishes the following four enforcement principles:
A. Issue-Driven, Competitive Harm Verification, and Evidence-Based Enforcement
a. The TFTC will adopt an issue-driven enforcement principle, meaning directing its enforcement resources on market behaviors that are genuinely relevant to market competition (such as abusing computing power for discriminatory treatment, tying, refusal to access, or self-preferencing). Issues falling under the scope of industrial policy shall be addressed by the respective sector authorities. When implementing industrial policy raises competition concerns, the TFTC will proactively advocate with relevant industrial government authorities in accordance with Article 46 of the Fair Trade Act (hereinafter the “FTA”).
b. Once an issue is identified as a competition law matter, the specific “theories of competitive harm” involved must be further clarified. For instance, when assessing whether self-preferencing or refusal to access implemented by a market participant raises competition concern, a key question may be to assess whether such conduct is used to steer users toward prioritizing its own models and computing services, thereby making it difficult for competitors to access critical cloud services, raising competitors’ costs, and causing market tipping.
c. Given the dynamic changes in market structure, the TFTC will follow an evidence-based enforcement principle and avoid a “one-size-fits-all” analytical approach.
B. Local Nexus Enforcement Principle: The TFTC stated that enforcement should be tailored to the Taiwan market conditions, including the structure and competitive landscape of its AI industry (for example, Taiwan holds a crucial position in the semiconductor sector, encompassing a complete value chain, while remaining less competitive in cloud services and large language models). There is no need to transplant foreign regulatory models in their entirety.
C. Maintaining an Enforcement Approach Centered on Market Contestability: The TFTC emphasizes that “only by maintaining the countervailing force from potential competitors — thereby ensuring that incumbents do not abuse their advantages to restrict competition — can competition be sustained and innovation enhanced.” In other words, enforcement will take into account the “potentiality” and “probability” of market entry. Furthermore, when evaluating data and computing power in the future, attention will be paid to the “qualitative” dimension, rather than relying solely on “quantity” and “scale”.
D. Strengthening the Competition Assessment under the “Rule of Reason”: The TFTC does not intend to rely exclusively on indicators of market structure to presume illegalities of competitive practices. The TFTC will evaluate the procompetitive and anticompetitive effects of disputed conduct on a case-by-case basis. Also, the dynamic relationship between the intra-ecosystem and inter-ecosystem competition would be assessed as well. The TFTC further indicated that “innovation” is the critical factor shaping and reshaping the landscape of the AI market.
In conclusion, the TFTC indicated that competition issues arising from the emergence of GenAI remain adequately addressable within the existing framework of the FTA. The current enforcement policy will be dynamically calibrated from time to time to address anticompetitive concerns arising from future AI market development.
III. Observations
1. Continuing the spirit of “big” does not equal “bad” from its White Paper, the TFTC reiterated the importance of the rule of reason, indicating that it is in no rush to characterize specific conduct as illegal per se during the early development stages of the AI industry. However, it also noted that future enforcement faces numerous challenges, including how to assess inter-ecosystem competition, competition for the market, and how to measure the role of innovation in competitive analysis—all of which rely on advancements in both theory and practice. Additionally, collaborations and alliances regarding chips, data, and computing power among upstream and downstream players in the AI value chain have become more frequent than ever, [1] which will inevitably challenge enforcement.
2. Although most industries seem to believe that the AI industry is still in its early stages and prefer that competition law not intervene too early, some insights can be gleaned from the opinions summarized in the Policy Statement. While foreign companies (primarily foreign cloud service providers and foundational model developers) generally do not believe there are anticompetitive concerns in the market, domestic Taiwanese firms express greater concern over potential restrictions on competition in the AI market, which is understandable.
3. Regarding the refusal-to-access practice, following the publication of the TFTC's Policy Statement, the European Commission recently directed an investigation into Meta for banning third-party general-purpose AI assistants from accessing the WhatsApp for Business API—essentially foreclosing competitors to ensure that only Meta's own AI assistant, Meta AI, remained accessible on WhatsApp. On June 9, 2026, the European Commission made a rare move by invoking its interim measures, ordering Meta to re-instate free access for third-party general-purpose AI assistants to the WhatsApp for Business API under the same terms and conditions that were in place before October 15, 2025, pending the completion of the investigation. [2] The action of the European Commission aligns with its position on structural regulation. While intervening with interim measures aims to prevent irreparable competitive harm, its positive and negative effects warrant further observation. The opposing views of the European Commission and Meta regarding the promotion and restriction of competition in this case are also worthy to be observed.
[1] For example, regarding unilateral conduct in the generative AI industry (see Policy Statement, Part IV, Public Comments on Potential Competition Issues Arising from the Generative AI Industry, Question 5), the TFTC summarized and disclosed four pieces of industry feedback. This included two from domestic companies and trade associations, and two from foreign foundational model developers and cloud service providers. Based on these opinions, domestic players and associations appear to believe that ecosystem integration and market behavior of large platforms could reshape the competitive landscape; conversely, foreign players believe there are no significant anti-competitive concerns as of yet.
[2] Press Release, European Commission, Commission Imposes Interim Measures on Meta to Preserve Free Access to WhatsApp for rival AI assistants (June 9, 2026), https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1276?utm_source=chatgpt.com.
