In 2024, India’s Committee on Digital Competition Law (CDCL) released a report proposing a draft Digital Competition Bill (DCB) to complement the existing antitrust regime in regulating emerging technological markets. Through the report and the draft Bill, the committee called for an ex-ante legislative instrument to be applied to a variety of core digital services, including social networking websites, e-commerce platforms, cloud services, and online search engines. In theory, this ex-ante instrument would set out a list of principles to guide competition governance and deter anti-competitive behaviour in the digital economy, as opposed to the ex-post rules currently found in India and the United States, which address anti-competitive behaviour only after it occurs.

The report’s stated aim was to study the current and future potential of AI technologies; the scope of their use in the Indian economy; and their impact on competition in the digital sphere.

Two years on, the draft Bill has yet to be officially introduced in Parliament, while digital markets have rapidly proliferated and expanded—most recently through the Artificial Intelligence (AI) revolution.

Large language models, given their enormous requirements for data, computing power, and energy, have emerged as the ideal opportunity for deep-pocketed Big Tech firms—many of which have already invested billions in developing datasets, algorithms, and products on top of their traditional digital offerings. Emerging market evidence indicates, however, that commonly found market practices—such as mandatory bundling of AI services, deep discounting at the model layer, or restricting cloud interoperability—are just as likely to entrench the dominance of hyperscalers as infrastructure providers for the AI stack.

It is in this context that we must situate the Competition Commission of India (CCI) and its role in regulating digital and, now, AI markets as India’s primary antitrust authority. In the aftermath of the CDCL report, the CCI authorised a market study on “Artificial Intelligence and Competition”, the final report for which was released late last year.

Prepared through a mixed-methods approach, the report’s stated aim was to study the current and future potential of AI technologies; the scope of their use in the Indian economy; and their impact on competition in the digital sphere. Notably, though the CCI has conducted similar initiatives in the past—such as studies of the e-commerce and telecom sectors—the scope of the AI market study is significantly larger.

Rather than concerning itself solely with antitrust risks in the evolving market, it takes on a much wider mandate, including questions such as the competitive advantages for AI adopters and potential market uses of AI tools.

NVIDIA holds 88% of the graphics processing unit (GPU) market share in India—a serious technical and economic bottleneck in the global AI supply chain.

Given the relative novelty of the topic, the CCI's initiative to conduct such an expansive study is both understandable and laudable. However, the report’s analysis of observed market conditions and its subsequent recommendations leave much to be desired in terms of progressive antitrust action.

Let us see how.

How is the Market Structured?

As the study rightfully points out, the creation of AI tools today depends on a range of inputs, including sector-specific and sector-agnostic data, compute infrastructures, and foundational and contextual models.

High entry barriers in these input markets mean that most, if not all, of them display clear trends of consolidation. Per the report’s own data, as shown in its “Figure 5” (reproduced here), NVIDIA holds 88% of the graphics processing unit (GPU) market share in India—a serious technical and economic bottleneck in the global AI supply chain.

The report also highlights that just three firms—Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP)—control nearly 65% of the compute market and about 46% of the data layer.

The report’s investigation into the core focus areas of Indian startups reveals a similar picture. “Figure 4” of the report shows that about two thirds of Indian startups work in the area of developing AI application models—that is, the final stage before deployment—while only 3% work in foundational model development, 10% in the compute infrastructure layer, and 20% in the data layer.

For competition regulation, such centralised supply chains and their impact on local innovation priorities should be a cause for concern. NVIDIA’s dominance in the infrastructure layer, for instance, has come from its strategy of bundling GPUs with its proprietary software platform—Compute Unified Device Architecture (CUDA)—which provides NVIDIA with critical network effects that raise entry barriers. Similarly, the dominance of large cloud service providers, or hyperscalers, in the compute and data layers is rooted in their history as first-moving behemoths that have cornered significant shares of their respective markets.

Why is this Dominance a Problem?

These observations are insightful because, as the report later notes, “hyperscalers are the primary providers of infrastructure, so the majority of AI developers and startups, including new and upcoming players, may have to enter into contractual arrangements in order to get access to infrastructure, including high-end computing capabilities”.

Incumbents and even well-capitalised startups have regularly indulged in “predatory innovation” that hinders rather than promotes market competition.

The bargaining power asymmetry between the two sides means that, without adequate oversight, these contracts can allow hyperscalers to indulge in “rentier capitalism” and charge monopoly prices to further entrench their dominance. The CCI’s recent investigation into Google’s contracts with Indian phone manufacturers has revealed this precise trend of exclusive arrangements, as well as forced bundling of Android with its suite of mobile applications.

Against this backdrop, the market study introduces a range of loosely categorised competition issues—not just those that exist in the AI industry itself (such as entrenchment of dominance or unilateral conduct), but also those that emanate from AI use in other domains (such as algorithmic collusion and selective pricing). This distinction is critical: the former emerges from specific anti-competitive conduct by already dominant Big Tech firms, whereas the latter has more to do with the market practices of deployers and adopters of algorithms, which need not always be affected by Big Tech’s market power.

Without explicitly acknowledging this core difference, the report ends up crossing the historically blurred line between promoting competition and fostering innovation.

In its discussion of entry barriers for startups, for instance, it includes challenges such as a dearth of skilled AI experts, absence of funding pathways, intellectual property restrictions, and lack of training data. While some of these could be linked to Big Tech’s dominant position—and the report successfully draws this link in a few cases—many have little to do with competition, let alone the power of hyperscalers. A similar conflation appears in the section on competitive advantages, where the report muddles the competitive benefits of using AI in a domain with the business value AI might generate for its adopters.

Although often seen as overlapping, competition and innovation are not simply two sides of the same coin, and digital markets are particularly instructive on this point. Incumbents and even well-capitalised startups have regularly indulged in “predatory innovation” that hinders rather than promotes market competition—for instance, by restricting data portability in ways that increase switching costs for users and lock them in.

Experience from data and platform markets also reveals how dominant firms— aware of their market power—chart their innovation trajectories to extract as much consumer surplus as possible, even when this leads to a deteriorating user experience (what activist Cory Doctorow has popularly referred to as “enshittification”).

Worse still, the report measures a politically charged topic like “innovation” using simple business metrics—such as projected market size and customer adoption—while paying little attention to the cultural, economic, and ecological dimensions of the concept. This means, on one hand, understating the many unrealised costs currently accompanying the use of AI tools, including their dependence on natural resources, the vastly prevalent digital divide, and their visible effect on factors such as cognitive development, labour rights, and social cohesion.

By measuring the value of innovation solely through the economic returns it could generate for adopters and the consumer welfare it could unlock for end users, the study adopts a market-led conception of innovation over one rooted in creating large-scale public value.

What Should Be Done About Market Consolidation?

While these instances of oversight might seem trivial or only relevant at a theoretical level, their implications become clearer in the recommendations put forth by the report.

The report’s preference for lightly governed, market-led innovation and its silence on the limitations of the existing ex-post regime leave much unexplored.

Having framed the goal of regulating competition as synonymous with incentivising innovation—itself measured through traditional metrics of consumer welfare—the study’s proposals put forward a policy regime that allows for free competition until harm is proven, rather than predicting and addressing it before it sets in. Some of its suggestions include compliance-linked self-audits and transparency disclosures by dominant enterprises that design or deploy AI models, national computing infrastructure and open data repositories to reduce entry barriers, and compliance-related advocacy by the CCI.

In effect, the recommendations sidestep questions of competition law altogether in favour of dealing with principles, processes, and investments—all of which are necessary but insufficient without the surrounding legal apparatus.

Currently, the CCI uses the Competition Act and its ex-post method for curbing anti-competitive conduct in digital markets. Emerging experience, including the CDCL report, highlights the inadequacies of this approach. A first-mover advantage and software’s low reproduction costs mean that successful entrants in digital markets can scale rapidly and leverage practices such as tying and bundling, self-preferencing, and strategic acquisitions to achieve and sustain dominance.

In an ex-post enforcement regime, the antitrust authority usually steps in only after it becomes aware of anti-competitive conduct and has conducted a thorough investigation. By then, however, the process of market consolidation is largely complete, and dominant players become practically immune to fines and other behavioural remedies.

The intertwined relationship between today’s AI and the erstwhile digital markets means that these limitations of the law are just as relevant for the report’s mandate—particularly since many of the instances of market consolidation and competition issues it identifies also have their origins in similar structural factors.

As a relatively late entrant to the global AI scene, India has the opportunity to chart a new path—informed by the experiences of others but rooted in principles of public interest.

A typical example is NVIDIA’s grip over the GPU market, which it owes in large part to its early entry and its proprietary programming tool, CUDA—a de facto industry standard among programmers that works solely with the company’s chips. Similarly, Big Tech’s use of exclusive partnerships with startups to maintain dominance is an issue the report explicitly lists, and it is, through and through, a question of merger review and digital competition law.

The report’s preference for lightly governed, market-led innovation and its silence on the limitations of the existing ex-post regime leave much unexplored. This is starkly apparent at a time when countries across the world—including from the Global South—have been pushing for an ex-ante approach to competition regulation in the digital economy.

By tracking and forbidding Big Tech firms from certain market-distorting practices—such as bundling of AI models into current offerings, exclusive and opaque arrangements with vendors and manufacturers, and potentially anti-competitive mergers and acquisitions (M&As)—the antitrust regulator can plan for trends of monopolisation before they set in, while also nudging private actors to serve the needs of public interest innovation.

Conclusion: Reviving Rules and Bill

The report in question is the CCI’s first major attempt at understanding market dynamics in the Indian AI ecosystem, and it provides a revealing insight into ownership structures and the complexities of competition in this market. Its identification of issues such as high entry barriers, switching costs, self-preferencing practices, algorithmic cartelisation, and price discrimination is essential for framing future policy proposals.

However, the report ignores a wealth of emerging antitrust discourse and locates its diagnosis in antiquated understandings of market power, competition, and innovation—prioritising traditional metrics of consumer welfare and pricing over structural factors such as technical lock-ins, bargaining power asymmetries, and data-driven consolidation. By relying on a market-led innovation model and light-touch regulatory methods, the report fails to advocate for more modern and suitable policy instruments like the draft Digital Competition Bill mentioned at the outset.

Despite its imperfections, the ex-ante Digital Competition Bill proposes much-needed modifications to India’s antitrust law, such as the identification of narrower markets in the digital economy and the recognition of specific anti-competitive principles. The Parliamentary Panel on the Standing Committee on Finance has also recently articulated the “critical regulatory gap” left by the stalled legislation. The CCI’s announcement last year of plans to develop an evidence base for the law is an encouraging step—but it will be crucial to ground this exercise in a structural understanding of digital competition, rather than one rooted in narrow definitions of consumer welfare and market-led innovation.

As a relatively late entrant to the global AI scene, India has the opportunity to chart a new path—informed by the experiences of others but rooted in principles of public interest. Doing so, however, requires wresting away the tight control exerted by Big Tech firms over the AI supply chain, and moving towards infrastructures that are open, inclusive, affordable, and accountable to the many communities that rely on them.

Abhineet Nayyar studies the political economy of digital technologies at IT for Change, Bengaluru.

Acknowledgement: This is an expanded and revised version of an article that was first published in ProMarket, titled “India’s AI Market Regulation Risks Falling on Dated Ideas.”

The author thanks Sadhana Sanjay and Isha Suri for their feedback in structuring and refining this article.