When Referee Plays The Game: Ai, Data Foreclosure, And Limits Of Indian Competition Law

India's e-commerce sector, projected to exceed $160 billion in gross merchandise value by 2028,^[1] is dominated by platforms that occupy three distinct commercial roles simultaneously. Amazon India and Flipkart operate as marketplace intermediaries hosting thousands of third-party sellers, as aggregators of granular transactional data generated by those sellers and their customers, and as direct competitors to those same sellers through private-label brands. This triple function is not incidental. It is embedded in the architecture of the vertically integrated platform model.

The Amazon Solimo case illustrates this structure with particular clarity. Investigative reporting in 2021 revealed internal Amazon documents showing that the company's India private-brands team studied sales data, customer review patterns, and product specifications of popular third-party brands listed on Amazon.in, and then used that intelligence to develop competing products under the Solimo, Symbol, and Amazon Basics labels.^[2] The strategy, described internally as leveraging “tribal knowledge,” involved identifying bestselling products, replicating their specifications, and deploying “search seeding” to ensure private-label goods appeared in the first two or three positions in search results.^[3] Amazon has denied giving preferential treatment to any seller, asserting that it identifies selection gaps based on aggregate customer preferences only.^[4]

This structure is not unique to Amazon. Flipkart, now owned by Walmart, has operated a comparable model through affiliated sellers and private labels. Reliance JioMart has entered the e-commerce space with an arguably more potent data advantage, combining telecom subscriber data from Jio with retail transaction data from Reliance Retail and payments data from Jio Financial Services.^[5] The Indian Parliamentary Standing Committee on Commerce noted in its 2022 report on e-commerce regulation that the dual role of platforms as hosts and competitors raises fundamental concerns about fairness and market access for small and medium enterprises.^[6] The competitive problem, therefore, is not about one company making better products through innovation. It is about an informational asymmetry that is structural, permanent, and embedded in the platform's business architecture. No third-party seller listing on Amazon.in or Flipkart can replicate the dataset that the platform itself generates through hosting marketplace activity. This asymmetry is the factual foundation upon which the legal analysis in this article rests.

From Data to Dominance: How AI Converts Information Asymmetry into Market Foreclosure

Raw data alone does not produce competitive advantage. The critical mechanism is the processing of proprietary behavioural data through artificial intelligence and machine learning systems. When platforms deploy recommendation algorithms, demand-forecasting models, dynamic pricing engines, and keyword optimisation tools trained on datasets that no competitor can access, the output is qualitatively different from anything achievable through conventional market research. AI-processed intelligence reveals optimal price sensitivity curves, temporal purchasing patterns, consumer preference signals at a granular demographic level, and predictive demand indicators that third-party sellers simply cannot reverse-engineer through surveys, publicly available analytics tools, or intuition.^[7]

This phenomenon constitutes what competition scholars have termed “data-driven foreclosure,” a concept distinct from traditional market foreclosure.^[8] Traditional foreclosure involves blocking a competitor's access to a physical input or distribution channel, such as refusing to supply a raw material or denying access to a transport network. Data-driven foreclosure operates differently. It denies access to an informational input that, once processed by AI, determines competitive outcomes in the downstream market. The platform does not need to refuse to deal with competitors explicitly. It merely needs to retain exclusive control over the data pipeline that feeds its AI systems, and the competitive gap emerges automatically.

Critically, this advantage is self-reinforcing. More data enables better AI models, which produce superior product placement and pricing, which generate more sales, which produce more data. This feedback loop, sometimes described in economic literature as a “data flywheel,”^[9] means that the informational asymmetry does not diminish over time but compounds. Farronato, Fradkin, and Karr, in their 2024 NBER study, found empirical evidence that Amazon's search rankings systematically favoured its own products, and that this self-preferencing was measurable through observed ranking patterns rather than relying solely on internal documents.^[10] The analytical distinction this article turns on is important for legal purposes. The relevant competitive input is not the data itself, which might be characterised as a collection of individual transaction records, but the AI-processed intelligence derived from that data, which constitutes actionable commercial knowledge. This distinction matters because the legal analysis of “data as essential facility” differs materially from the analysis of “AI training outputs as essential facility,” a point developed in further portion of the Article.

The Legal Blindspot: Why Section 4 of the Competition Act Falls Short

Section 4 of the Competition Act, 2002 prohibits the abuse of a dominant position by an enterprise.^[11] The provision enumerates specific categories of abuse, including the imposition of unfair or discriminatory conditions under Section 4(2)(a)(i), denial of market access under Section 4(2)(c), the use of dominant position in one market to enter or protect another market under Section 4(2)(e), and predatory pricing under Section 4(2)(a)(ii). The Competition Commission of India has applied these provisions with increasing vigour in digital markets. In its 2022 orders against Google, the CCI found abuse of dominance in relation to Android's licensing conditions and Play Store policies, imposing a cumulative penalty exceeding Rs 2,200 crore.^[12] In the WhatsApp/Meta case, the CCI in November 2024 imposed a penalty of Rs 213.14 crore and prohibited WhatsApp from sharing user data with Meta for advertising purposes, finding violations of Sections 4(2)(a)(i) and 4(2)(c).^[13] The NCLAT, in its order of 4 November 2025, upheld the CCI's jurisdiction to examine data-sharing practices as a competition concern and affirmed the penalty, though it set aside the five-year data-sharing ban on the ground that it risked collapsing WhatsApp's business model.^[14]

These cases demonstrate that the CCI is neither hostile to digital-market enforcement nor reluctant to treat data practices as potential abuses. However, the existing statutory framework and jurisprudence present three specific doctrinal difficulties when confronted with AI-driven foreclosure. First, the definition of the relevant market becomes deeply uncertain. Section 2(r) of the Competition Act defines relevant market with reference to relevant product market and relevant geographic market. When competitive advantage depends not on a conventional product market but on access to proprietary training data that feeds AI systems, the traditional methodology of market definition struggles to capture the locus of competitive harm. The CCI's investigation into Amazon and Flipkart, initiated in January 2020 following a complaint by Delhi Vyapar Mahasangh,^[15] and its August 2024 findings of preferential treatment of select sellers,^[16] were framed in terms of vertical agreements and seller discrimination rather than data-market competition as such.

Second, proving abuse becomes elusive. The conduct in question, namely training AI on proprietary data and algorithmically optimising private-label placement, is not a discrete, observable exclusionary act of the kind that Section 4 was designed to capture. It is an emergent property of the platform's architecture. Unlike predatory pricing, which can be measured against costs, or a refusal to deal, which involves an identifiable denial, AI-driven self-preferencing operates through continuous algorithmic optimisation that is invisible to external observers and potentially even to the platform's own management in any granular sense.^[17]

Third, causation presents a fundamental evidentiary challenge. There is currently no accepted methodology in Indian competition jurisprudence to disentangle whether a private-label product ranks higher in search results because it is genuinely superior in quality and price, or because the platform's AI was trained on data that competitors can never access. The Supreme Court's January 2025 transfer order, directing all pending petitions challenging the CCI's Amazon and Flipkart investigation to the Karnataka High Court,^[18] signals judicial recognition of the case's complexity and significance, but the underlying doctrinal gap remains. The CCI's investigative framework lacks tools for algorithmic auditing, data access obligations, or AI-specific remedies. The Competition (Amendment) Act, 2023 enhanced penalty provisions and streamlined settlement mechanisms but did not introduce any provisions addressing algorithmic competition, data-driven market power, or platform-specific obligations.^[19]

Data as Essential Facility: An Untested Doctrine for the AI Era

The essential facilities doctrine, which holds that a monopolist controlling an infrastructure essential for competition may be required to grant access to competitors, originated in United States antitrust law with the Supreme Court's decision in United States v Terminal Railroad Association of St. Louis,^[20] where railroads were compelled to share terminal facilities. The doctrine was elaborated in MCI Communications v AT&T,^[21] which established a four-part test: the facility is controlled by a monopolist, the competitor is unable practically to duplicate it, access has been denied, and providing access is feasible. However, the US Supreme Court significantly curtailed the doctrine in Verizon Communications v Trinko,^[22] expressing scepticism about judicial imposition of sharing obligations and refusing to recognise the essential facilities doctrine as an independent basis for liability under Section 2 of the Sherman Act.

In European Union jurisprudence, the doctrine has followed a more expansive trajectory. The European Commission's 2017 decision in the Google Shopping case,^[23] upheld by the Court of Justice of the European Union in 2024, established that self-preferencing by a dominant platform constitutes an abuse of dominance even without an explicit refusal to deal. Academic commentary has pushed the analysis further. Inge Graef's foundational work on “data as essential facility” in the EU context argues that aggregate data held by dominant platforms may satisfy the essential facility criteria: competitors cannot practically duplicate the dataset, the platform controls exclusive access, denial of access forecloses meaningful competition, and sharing is technically feasible.^[24] The Yale Law Journal's 2022 “Essential Data” framework similarly contends that data accumulation by platforms can create competitive bottlenecks that warrant compulsory access obligations.^[25] Guggenberger has extended this reasoning to argue that certain platforms have become “essential” infrastructure in digital economies, warranting regulatory obligations analogous to common carriers.^[26]

The question for Indian law is whether and how this doctrine can be accommodated within the existing statutory framework. Section 4(2)(c) of the Competition Act prohibits “denial of market access” as an abuse of dominance.^[27] Could this provision be interpreted to encompass denial of access to data essential for competing in algorithmic markets? The doctrinal stretch required is substantial: from physical infrastructure such as rail bridges and port terminals to intangible data assets, and from “data” in a general sense to “AI training data” specifically. No Indian court or the CCI has yet treated data, let alone AI training data, through the essential facilities lens.

Counterarguments are significant. The Areeda critique, influential in US antitrust scholarship, cautions that mandatory sharing obligations may reduce incentives for innovation by diminishing the returns from investment in data infrastructure.^[28] Administrability problems are real: defining what data must be shared, at what level of aggregation, and under what conditions poses immense practical difficulties. Furthermore, mandatory data sharing creates tension with data protection obligations under the Digital Personal Data Protection Act, 2023,^[29] which imposes purpose limitation and consent requirements that may be difficult to reconcile with broad compulsory access mandates. This article does not argue that India should adopt the essential facilities doctrine uncritically. Rather, it contends that the question is ripe for adjudication and that the CCI needs a principled test to evaluate when, if ever, denial of access to AI training data constitutes an abuse of dominance under Section 4.

Lessons from the EU and US, and What India Should Build Instead

The European Union's Digital Markets Act, which entered into force in 2022, provides the most developed regulatory response to the problems identified in this article.^[30] Article 6(5) of the DMA explicitly prohibits designated gatekeepers from engaging in self-preferencing, that is, from treating their own services more favourably in ranking, indexing, or display than comparable services offered by third parties. Article 6(2) bars gatekeepers from using non-public business-user data to compete with those users.^[31] Enforcement has been vigorous. In March 2025, the European Commission fined Apple €500 million and Meta €200 million for breaching DMA obligations.^[32] The Commission's first review of the DMA, with a consultation that closed in September 2025, is evaluating whether AI should be designated as a core platform service, a move that would bring AI-integrated marketplace functions within the scope of the DMA's obligations.^[33]

The United States offers no equivalent legislative framework. The FTC's September 2023 complaint against Amazon,^[34] filed in the Western District of Washington under Section 2 of the Sherman Act and Section 5 of the FTC Act, proceeds under traditional monopolisation theory. The complaint alleges self-preferencing in search results, exploitative fee structures, and exclusionary logistics practices, but does not frame any claim in terms of AI-specific data foreclosure or training-data access obligations. The case, scheduled for trial in October 2026, thus relies on conventional ex post enforcement tools without AI-specific analytical frameworks.

Neither model can be imported directly into India. The DMA is an ex ante regulatory framework tied to the EU's institutional architecture, involving European Commission enforcement and per-infringement fines of up to 10 per cent of global turnover. India's CCI operates under a fundamentally different institutional model of investigation, adjudication, and appeal, and the Competition Act remains an ex-post enforcement statute.^[35] The US approach, which does not contemplate ex ante data obligations at all, offers no model for the kind of structural intervention that AI-mediated markets may require. India could, however, adapt specific elements: the DMA's data-use prohibitions offer a regulatory template for preventing cross-use of non-public seller data; the empirical work by scholars such as Jurgensmeier and Skiera on measuring self-preferencing on digital platforms^[36] could inform the CCI's investigative methodology. What India must build from scratch is an AI-auditing methodology suited to the CCI's investigative capacity, and a data-access framework that is reconciled with the DPDP Act's consent and purpose-limitation requirements. The design principles for an Indian-specific response must be technically enforceable, privacy-compatible, innovation-preserving, and suited to the CCI's existing adjudicatory capabilities.

Toward a Framework: What India's Competition Regime Needs Now

The analysis in the preceding Article points to four contributions that are needed to close the regulatory gap. First, a legal test for when AI training data becomes “essential” under Indian competition law, building on the Section 4(2)(c) prohibition of denial of market access but adapted to the distinctive features of intangible, non-rivalrous, and continuously generated data assets. Second, an economic methodology to measure AI-driven foreclosure, one capable of quantifying the competitive impact attributable to data access rather than product quality, and suitable for deployment in CCI proceedings. Third, a regulatory framework for data access obligations in AI-mediated markets that is reconciled with the privacy architecture of the Digital Personal Data Protection Act, 2023, possibly through mandatory anonymisation protocols, data intermediary models, or purpose-limited access regimes. Fourth, a comparative analysis of how the Indian Competition Act, the EU DMA, and US antitrust law can or cannot address algorithmic foreclosure, identifying both transferable elements and irreducible jurisdictional differences.

These are not academic exercises. The CCI's Amazon and Flipkart investigation is at an advanced stage, with penalty proceedings pending following the August 2024 findings. The NCLAT has upheld the CCI's jurisdiction to treat data-related practices as abuses of dominance in the WhatsApp/Meta decision.^[37] Reliance JioMart is entering e-commerce with unprecedented cross-sector data drawn from telecom, retail, and financial services. The Indian Parliamentary Standing Committee on Commerce has flagged the competitive risks of platform self-preferencing and data asymmetry.^[38] Meanwhile, the Digital Competition Bill, which would empower the CCI to address digital competition through ex ante obligations, remains paused and has not yet been introduced in Parliament.^[39]

The regulatory gap is widening as the market grows. The tools to address AI-driven data foreclosure, whether through statutory amendment, CCI guidelines, or judicial interpretation of the existing Section 4 framework, must be developed now. India cannot afford to wait for a fully formed foreign model to emerge. The problem is present, the institutional need is urgent, and the doctrinal foundations, while incomplete, are sufficient to begin building a principled response that is fitted to India's legal and market conditions.

Endnotes:

1. Bain & Company, 'India E-Retail Market Estimate' (2024), cited in 'India Probe Finds Amazon, Walmart's Flipkart Breached Antitrust Laws' (CNBC, 13 September 2024), https://www.cnbc.com/2024/09/13/india-probe-finds-amazon-walmarts-flipkart-breached-antitrust-laws.html. ↑

2. 'Amazon Copied Products, Rigged Search to Push Own Brands' (Reuters, 14 October 2021), https://www.reuters.com/investigates/special-report/amazon-india-rigging/. ↑

3. ibid. ↑

4. ibid. Amazon's statement to Reuters noted that it “identifies selection gaps based on customer preferences at an aggregate level only and shares this information with all sellers.” ↑

5. Parliamentary Standing Committee on Commerce, Report on the Promotion and Regulation of E-Commerce in India (172 Report, 2022), https://prsindia.org/policy/report-summaries/promotion-and-regulation-of-e-commerce-in-india. ↑

6. ibid paras 5.8–5.12. ↑

7. Jacques Crémer, Yves-Alexandre de Montjoye and Heike Schweitzer, Competition Policy for the Digital Era (European Commission Report 2019) ch 2 https://data.europa.eu/doi/10.2763/407537. ↑

8. ibid ch 4; see also Feng Zhu and Qihong Liu, 'Competing with Complementors: An Empirical Look at Amazon.com' (2018) 39 Strategic Management Journal 2618, https://dash.harvard.edu/entities/publication/06b55865-1bab-4ccb-95c6-a17b6ff9100a . ↑

9. Lina Khan, 'Amazon's Antitrust Paradox' (2017) 126 Yale Law Journal 710, 780–85, https://ssrn.com/abstract=2911742. ↑

10. Joel Waldfogel, 'Amazon Self-preferencing in the Shadow of the Digital Markets Act' (NBER Working Paper No 32299, April 2024, https://www.nber.org/papers/w32299. ↑

11. Competition Act 2002 (India), s 4. ↑

12. Competition Commission of India, Google LLC & Ors, Cases No 39 and 40 of 2018 (CCI Orders, 2022). ↑

13. Competition Commission of India, Order in the matter of Suo Moto Case No 01 of 2021 (WhatsApp LLC / Meta Platforms Inc) (18 November 2024). ↑

14. WhatsApp LLC v Competition Commission of India, Competition Appeal No 1 of 2025 (NCLAT, 4 November 2025). ↑

15. ^. Delhi Vyapar Mahasangh v Flipkart Internet Private Limited & Amazon Seller Services Private Limited, Case No 40 of 2019 (CCI, Order dated 13 January 2020 directing investigation). ↑

16. Competition Commission of India, Investigation Report (DG Report, August 2024) in Case No 40 of 2019; 'India Probe Finds Amazon, Walmart's Flipkart Breached Antitrust Laws' (Reuters/CNBC, 13 September 2024), https://www.cnbc.com/2024/09/13/india-probe-finds-amazon-walmarts-flipkart-breached-antitrust-laws.html. ↑

17. Lukas Jurgensmeier and Bernd Skiera, 'Measuring Self-Preferencing on Digital Platforms' (arXiv:2303.14947, 2024), https://arxiv.org/abs/2303.14947. ↑

18. Supreme Court of India, Transfer Petition Order (Amazon.in / Flipkart v CCI) (6 January 2025); also see, Bhavini Mishra 'Amazon, Flipkart pleas against CCI probe order transferred to Karnataka HC' (Business Standard, 06^th January 2025) https://www.business-standard.com/companies/news/sc-transfers-amazon-flipkart-petitions-on-cci-probe-to-karnataka-hc-125010600909_1.html. ↑

19. Competition (Amendment) Act 2023 (India). ↑

20. United States v Terminal Railroad Association of St Louis 224 US 383 (1912). ↑

21. MCI Communications v AT&T 708 F 2d 1081 (7th Cir 1983). ↑

22. Verizon Communications v Law Offices of Curtis V Trinko 540 US 398 (2004). ↑

23. European Commission, Decision in Case AT.39740, Google Shopping (2017), upheld by the Court of Justice of the European Union (2024) https://ec.europa.eu/competition/antitrust/cases/dec_docs/39740/39740_14996_3.pdf. ↑

24. Inge Graef, EU Competition Law, Data Protection and Online Platforms: Data as Essential Facility (Kluwer Law International 2016) https://ssrn.com/abstract=3635378; Inge Graef, 'Rethinking the Essential Facilities Doctrine for the EU Digital Economy' (2019) Journal of Intellectual Property, Information Technology and E-Commerce Law, 2019, https://research.tilburguniversity.edu/en/publications/essential-facility/. ↑

25. Abrahamson, Z. (2014), 'Essential Data', Yale Law Journal, 124, p.867, https://heinonline.org/hol-cgi-bin/get_pdf.cgi?handle=hein.journals/ylr124§ion=25&casa_token=wHhMVYnSSucAAAAA:tcmd8sdUsvbkVQPc9X7xLQA-hEP9HDxGxW-ODsYahiby7Rcp3EIIDZcUW1-mdYZHSIf2mbwI_A. ↑

26. Nikolas Guggenberger, 'Essential Platforms' (2021) 24 Stanford Technology Law Review 237, https://law.stanford.edu/publications/essential-platforms/; Nikolas Guggenberger, 'The Essential Facilities Doctrine in the Digital Economy' (2021) 23 Yale Journal of Law & Technology, https://yjolt.org/sites/default/files/23_yale_j.l_tech._301_essential_facilities_0.pdf. ↑

27. Competition Act 2002 (India), s 4(2)(c). ↑

28. See Phillip Areeda, 'Essential Facilities: An Epithet in Need of Limiting Principles' (2016) https://www.semanticscholar.org/paper/ESSENTIAL-FACILITIES:-AN-EPITHET-IN-NEED-OF-Areeda/6c268e049f8396881b25d2f0b398c55ea41e2728. ↑

29. Digital Personal Data Protection Act 2023 (India), ss 4-6. ↑

30. Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 September 2022 on Contestable and Fair Markets in the Digital Sector (Digital Markets Act), http://data.europa.eu/eli/reg/2022/1925/oj. ↑

31. DMA, arts 6(5), 6(2). ↑

32. European Commission, DMA Enforcement Decisions (March 2025): Apple fined €500 million, Meta fined €200 million, https://ec.europa.eu/commission/presscorner/detail/en/ip_25_1085. ↑

33. European Commission, First Review of the Digital Markets Act, Consultation on AI as Core Platform Service (consultation closed 24 September 2025). See 'Will the EU Designate AI Under the Digital Markets Act?' (TechPolicy.Press, 26 September 2025), https://www.techpolicy.press/will-the-eu-designate-ai-under-the-digital-markets-act/. ↑

34. FTC v Amazon.com Inc, Case No 2:23-cv-01495 (WD Wash, filed 26 September 2023). ↑

35. Competition Act 2002 (India), ss 19, 26, 27. ↑

36. Lukas Jurgensmeier and Bernd Skiera (n 17). ↑

37. WhatsApp LLC v Competition Commission of India (n 14). ↑

38. Parliamentary Standing Committee on Commerce (n 5), 2022, https://sansad.in/getFile/rsnew/Committee_site/Committee_File/ReportFile/13/159/172_2022_7_14.pdf?source=rajyasabha. ↑

39. 'CCI Seeks NCLAT Clarification on WhatsApp Ad Data Sharing' (Medianama, 20 November 2025), noting that the Digital Competition Bill remains paused, https://www.medianama.com/2025/11/223-cci-nclat-whatsapp-ad-data-sharing-users-india/. ↑