Abstract
The AI market study of the Competition Commission of India ('CCI') lays out a complex AI ecosystem. This article seeks to assert that a map of this ecosystem treated as a single 'AI market' will disarm the CCI because of Section 4(2)(e) of the Competition Act, 2002, which prohibits leveraging in two different markets. This article seeks to encourage layer-by-layer market definition so that the CCI can deal with cross-layer leveraging by the standard set by both Indian and European enforcement.
Introduction
The Competition Commission of India ('CCI') released a market study regarding artificial intelligence ('AI') on 6 October 2025. The study built an AI stack consisting of multiple layers: data and computer infrastructure, machine learning/modeling, and application. The study characterized the layers as well. NVIDIA holds an impressive 88% of the graphics processing market share in India and three hyper-scalers accounted for almost 65% of processing. Even more interesting, one-third of AI companies in India are involved in developing applications while only 3% work on foundational models. The findings in the CCI study are significant. From this point onward, the pressing issue that needs further exploration is what happens following these findings if this level of detail is used as an implicit definition of 'the market'.
Evidently, simply because those layers are interrelated, defining a traditional market in this manner does not mean the regulators are free to define an entire AI technology ecosystem as such. The most common argument against defining a broad 'AI market' is the imprecision of the definition. More importantly, however, the argument becomes statutory, which may be counterintuitive: creating a single definition of 'the AI market' actually limits the CCI's ability to take regulatory action in this area, rather than enhances it. Section 4(2)(e) requires a definition of two separate markets, which constitutes the primary distinction of the argument. It is narrow, as it does not argue that the market can be defined analytically and provides an argument that contradicts a number of fundamental claims.
Substitutability instead of proximity
The Competition Act, 2002 at Section 2(t) maintains the relevant product market is defined by whether the products are regarded as substitutable by consumers, while Section 19(7) of the Act sets out what factors are relevant in that regard.
Technological linkage is not included as a relevant factor. A graphics processing unit and consumer-facing AI chatbot are complementary products at different stages in the supply chain because even expensive processors cannot influence a customer's decision to buy a chatbot. In contrast, cloud capability, model weights and the assistant interface all contribute to the development of each subsequent product in the vertical chain. Thus, they are in different markets. One side of this vertical boundary will be taken into account when calculating foreclosure in a market in that vertical chain; without the separation of the two sides it would be impossible to recognize whether a foreclosure situation exists.
The revision of the European Commission Notice with regard to market definition is frequently used to justify the contrary position. The notice recognizes that core product ecosystems exist, whereby either core or secondary products are dependent on each other based on technology and interoperability considerations; hence, the examination of previous products offered free of charge as substitutes must adopt a modified SSNIP test.
It is notable, however, that the CCI is directed to apply the same SSNIP test at Section 19(7), where it considers substitution variables (end use of product, price point and consumer preferences) that depend on the buyer's choice of alternative products, not technological product proximity.
Actual enforcement
Current enforcement points the same way. The European Commission ordered interim measures against Meta on 9 June 2026 for excluding rival general-purpose AI assistants from the WhatsApp Business API while keeping Meta AI available. It was only the second such decision under Regulation 1/2003. The Commission found Meta prima facie dominant in an existing market, the European Economic Area market for consumer communication applications, since at least January 2023, and used the AI assistants market only to show where harm would fall. Italy's competition authority, in its case against Meta over Meta AI's integration into WhatsApp, likewise anchored dominance in consumer communications apps and assessed harm in AI chatbot services.
Similarly, the Commission's inquiries into Google in December 2025 over its use of publishers' content and YouTube videos for AI rest on Google's search dominance, not on a generative AI market. In the UK, the Competition and Markets Authority chose in March 2026 to address AI tools such as Copilot through an investigation into Microsoft's business software ecosystem, rather than pursue designation of the cloud providers its 2025 market investigation found to hold significant market power.
India already has the framework
India need not borrow this method from abroad. In Umar Javeed v Google LLC, the CCI identified five relevant markets within one technologically integrated ecosystem: licensable smart mobile operating systems, Android app stores, general web search, non-OS-specific web browsers and online video hosting. It declined to define an "Android ecosystem" market; only this five-way division made its leveraging finding possible, and the National Company Law Appellate Tribunal largely upheld the order. Harshita Chawla v WhatsApp Inc and Facebook Inc followed the same logic, separating a market for over-the-top messaging apps on smartphones from one for UPI-enabled digital payments apps, into which WhatsApp allegedly leveraged its dominance.
Why breadth disarms
Section 4(2)(e) prohibits using dominance in one market to enter, or protect, another. That rule covers most of what regulators fear in AI: a firm dominant in cloud, search or messaging extending that power into models or assistants. But it has substance only if the two markets are separate.
If one market is collapsed into the other, that separation disappears. If compute, models and assistants form a single AI market, a hyperscaler that pairs its own model with its cloud offerings is not leveraging anything; it is acting within one market, so Section 4(2)(e) cannot be invoked. The CCI would be left with Section 4(2)(a), (c) and (d), and would first have to establish dominance across compute, models and applications combined, a much harder task than doing so in cloud infrastructure alone.
There is a substantial evidentiary difference. In regard to those heads, the CCI has to demonstrate that conduct occurring within a particular market is abusive in itself, whereas under Section 4(2)(e), all that needs to be demonstrated is that the dominant position in one market has adversely affected the entry into or protection of the firm's position in another market. Therefore, with the collapse of the stack the fundamental mechanism which was conceived to explore cross market leveraging disappears.
As regards the proposed ex-ante regime for digital markets in India, the definition of markets becomes irrelevant: the proposed Draft Digital Competition Bill 2024 does impose obligations on dominant firms in regard to specific core digital services.
The Bill, however, has not been enacted. Until it is, AI conduct must be assessed under Section 4, where market definition remains the starting point.
A boundary, not a rebuttal
Harshita Chawla v WhatsApp Inc and Facebook Inc is a caution against overstating this claim, not a rebuttal of it. The CCI defined the two markets correctly there and still declined to find leveraging, because the payments market was too immature for the alleged effects to be shown. That was a problem of proving effects, not of defining markets: immaturity affects whether foreclosure can be proved, not whether a second market exists. Nascent AI layers will remain hard cases for the CCI however the market is drawn, so breadth matters mainly in mature layers such as cloud.
Conclusion
Market definition works as a constraint: it requires an authority to locate power before finding abuse. Applied correctly to AI, it leads to an unattractive conclusion: for a Section 4(2)(e) case to succeed, there will usually be no single AI market at all. The CCI should therefore read its market study as a map, not a market definition. In individual cases, it should define markets layer by layer under Sections 2(t) and 19(7), locate dominance in established markets such as cloud, search or messaging, and test effects in the AI layer where harm appears, as the European Commission did with Meta. Until the Digital Competition Bill is enacted, this approach keeps Section 4(2)(e) usable. The relevant question is where leverage starts, not where an AI market begins.