The Schott Glass Standard in the AI Era: Sufficient or Strained?

I. Introduction

In May 2025, the Supreme Court [“SC”] said exactly what we had been waiting for years, that you can’t punish a dominant company just because their actions look anti-competitive, rather you have to actually show they injured competition. The case of CCI v. Schott Glass India Pvt. Ltd., marks a significant inflection point in the interpretation of abuse of dominance.

The judgment comes precisely at a point where the Competition Commission of India [“CCI”] has something a lot more complicated to worry about than a glass tube manufacturer. They are about to deal with AI platforms, algorithmic pricing and markets where the injury they want to prevent will only materialise on a spreadsheet years later. Therefore, the question is not whether the Schott Glass standard is the right one, but whether it’s sufficient.

II. What Schott Glass Was Actually About?

Schott was the dominant domestic supplier of neutral borosilicate glass tubing. Kapoor Glass alleged that Schott had abused its position through volume rebates, a long-term supply arrangement with its affiliate Schott Kaisha, tying of two glass tube types, and selective refusals to supply. CCI confirmed such abuse in 2012 and imposed a fine of Rs 5.66 crore on Schott for violation of Section 4 of the Competition Act, 2002. The CCI decision was later set aside by the Competition Appellate Tribunal [“COMPAT”]. The apex court affirmed this and dismissed the appeal. It clearly held that Section 4 analysis should be effects-based. It said that an effects-based analysis is mandated, not optional, under Indian law as supported by the preamble, the definition of dominant position, and section 19(4)(l). It also affirmed that Section 4(2) is not a deeming provision and even a presumption of abuse under Section 3(3) is rebuttable and there can be no automatic abuse under Section 4. Furthermore, the court took into account objective justifications about the uniform, commercially justified rebates offered by Schott and ordinary business transactions involved in the supply agreement with affiliates; the absence of coercion which defeated the tying claim. For this, the court also relied upon various European Union [“EU”] laws, including Intel v European Commission and TeliaSonera Sverige AB v Konkurrensverket.

Thus, the shifting from form-based to effects-based analysis is firmly cemented in Indian competition law. However, the complex question which arises is, what next?

III. Where the Framework Gets Strained: AI and Digital Markets

In October 2025, the CCI published its ‘Market Study on Artificial Intelligence and Competition’, its first in-depth effort to chart the competitive implications of AI. The concerns raised are not hypothetical, but rather structural and expose a genuine limitation in the Schott Glass framework. The framework was designed for traditional markets where competitive harm typically manifests through observable changes in price or output, but in digital and AI markets, harm often appears more subtly, and sometimes in forms that traditional effects analysis struggles to measure.

a) The Problem of Collusion

One of the more counterintuitive situations is the case of algorithmic tacit collusion. Consider two e-commerce firms that compete by using AI-driven dynamic pricing algorithms, each acting independently. No agreement is ever reached, no meeting of minds occurs. After thousands of pricing cycles, however, each algorithm learns that attempts to undercut its rival lead to price wars that ultimately harm both. So, they implicitly maintain high prices without explicitly colluding, increasing costs to consumers without engaging in traditional unlawful conduct.

This isn’t an abstract possibility. A 2024 study of German petrol stations found that after the two firms in local duopolies adopted AI pricing software, margins increased by 28%, while there was no similar price rise for local monopolies. These issues have also surfaced in the U.S., with the Department of Justice filing a suit against RealPage in 2024, alleging it facilitated algorithmic convergence in rental markets, while in India, 37% of responding firms believed algorithmic collusion was a considerable threat in the CCI study.

The judgment requires a specific type of conduct that leads to an ‘Appreciable Adverse Effect’ on competition to be established. The difficulty with algorithmic tacit collusion is that the evidence for it is not “conduct,” as there is no agreement or concerted practice and thus no communication or meeting of minds. Traditional competition law doctrines assume some human element of communication. With independently optimising algorithms producing supra-competitive outcomes, it becomes immensely difficult to meet the evidentiary standard.

b) The Stack Problem

A second structural problem lies in the AI value chain itself. A handful of global companies have market power in the upstream: cloud compute, large-scale training data, foundational models, and distribution channels. An Indian AI startup that builds applications using services like Azure OpenAI or AWS Bedrock is, from day one, both a customer and downstream competitor of the upstream provider.

Competitive harm in this structure typically doesn’t involve price increases or volume limitations; rather, it involves phased lock-in. Once a startup builds a product from a specific model or infrastructure, the cost of switching is prohibitively high, which involves retraining models, re-deploying data, and rebuilding APIs. Potential rivals who get worse access or are deliberately slowed just never materialize as competitors, which leads to a prospective foreclosure and a lost rivalry. Traditional Schott Glass analysis focusing on actual foreclosure of an incumbent rival doesn’t really capture these structural long-term exclusionary processes in vertical markets.

c) The Self-Preferencing Problem

There stems a problem of hidden self-preferencing from the relationship between layers. In Matrimony.com v. Google, the CCI ruled that Google abused dominance by favouring its services within its search rankings. However, self-preferencing takes on a more complicated and hidden form within AI systems. A firm which develops the basic model and its downstream applications can indirectly achieve favourable results through choices such as model fine-tuning or differing access and data availability, which allow their services to be better optimised than competitors who may not have the same access.

This is not conduct that is likely to immediately manifest in price effects or a decrease in output. The harm here is potential, and thus difficult to demonstrate in a short-term, effects-based standard. Competitors are not necessarily foreclosed; they simply never emerge as effective potential downstream competitors. The true injury within AI markets may be harm to competition that never gets off the ground.

The relevant concerns within digital markets involve platform and ecosystem dominance, long-term foreclosure of downstream markets, and structural benefits within vertical structures. Without modifications, an effects-based standard is likely to be either too heavy-handed or ineffective in dealing with the complexities of the AI market.

IV. What India Should Do About It

The Supreme Court has given us a principle that is sound and should be retained. What the CCI requires are proportionate and practical tools and means to apply the principle effectively to the complexity of AI and digital markets.

a) Public Digital Effects Guidelines

With existing powers under Sections 36, which empowers the CCI to regulate its own procedure, and Section 64, which empowers the Commission to make regulations consistent with the Act, the Commission need not await further legislation to provide clarity. Digital Effect Guidelines should lay out clearly how to perform effects analysis in the digital sphere, define which counterfactuals apply in a volatile market definition, what kind of quality degradation or reduction in innovation qualifies as harm, how to assess consumer switching costs, and when there is a justification for intervention in forward-looking potential foreclosure before price effects become visible. Such guidelines would bring much-needed predictability for businesses and consistency in enforcement.

b) Establish clear safe harbours for open and transparent AI systems

Over-enforcement is an important challenge when addressing competitive harms by new technologies. An open-source, open Application Programming Interface [“API”], transparent, interoperable, and research-ready AI stack has demonstrably lower competitive risks than closed proprietary models. Based on the CCI’s own recommendations in the 2025 AI Study for a ‘light-touch’ approach and self-auditing for AI models, specific safe harbours for such transparency frameworks will give Indian businesses the certainty to invest in and develop such technology, and focus the CCI on areas that represent genuine monopolistic practices.

c) Build institutional capacity and inter-agency coordination

Competition enforcement in the digital space requires an array of skills in econometrics and computer science that go beyond traditional legal economics. It is important to create specialized economic units within the CCI, employing data scientists proficient in algorithms and economics, and well-versed in the platform economy. Coordination gaps with other sectoral regulators like SEBI and RBI also need to be addressed more robustly; while Section 21A already permits collaboration, formal operational working groups or MoUs with regulators must be put in place to avoid “enforcement blind spots.”

These steps can enable effective functioning even in the face of emerging digital technologies and the competitive harms they generate, while ensuring the Indian market climate encourages innovation and investment, in line with the spirit of Schott Glass.

V. The Bottom Line

Schott Glass has truly revolutionised Indian competition law by marking the end of an era in which firms with a commanding position could be punished for the form of their conduct, rather than its outcome. However, it has also posed difficult questions, namely, how does the effects-based approach translate into a tool for concrete digital effects guidelines, restricted rebuttable presumptions for systemic platforms, a well-calibrated safe harbour for open and transparent AI, and a specialised institutional unit with reinforced inter-regulatory cooperation.

About the Author

Ahan Garg is a second-year law student at National Law University, Jodhpur.

Editors

Abhimanyu Vyas, Senior Editor