The development of Artificial Intelligence (“AI”) is among the most significant revolutions impacting human behavior in recent times. Due to its rapid growth in usage and adoption in the past few years, AI is already relevant to almost every single business in operation today. The adoption of AI has therefore been of special interest to the Competition Commission of India (the “CCI”), which is mandated with the duty to protect competition among businesses in India.
In 2025, the CCI released its ‘Market Study on Artificial Intelligence and Competition’ (the “Report”)1 in an effort to explore and understand the AI ecosystem, identify the potential impact of AI and AI-powered tools on competition and provide recommendations from a regulatory standpoint to curb or prevent anti-competitive practices that arise from the use of AI. Significantly, the Report provides guidance around self-audits that businesses using AI systems should conduct to ensure compliance with competition law.
This note explores the Report and its findings on the effects of AI on competition between businesses, as well as how the Indian competition regulator views this change in the way of doing business in India.
Basic Overview of the Use of AI in India
The Report provides a helpful overview of the AI landscape in India, identifying two key segments:
Upstream AI segment
The Upstream AI segment is further divided into four layers:
| Sr. No. | Layer | Function | Major players |
|---|---|---|---|
| 1. | Data Layer | This layer relates to the collection of data from various sources, and the preparation (i.e., cleaning, labeling) and governance (i.e., security) of data. | Appen, Amazon Web Services (“AWS”), Google, Microsoft Azure (“Azure”), Scale AI |
| 2. | Computing and AI Infrastructure Layer | This layer relates to the infrastructure for processing data collected at scale, including (i) cloud infrastructure; and (ii) hardware such as AI chips and graphics processing units (“GPUs”) to train and deploy models. | Cloud computing: AWS, Azure, Google Cloud Platform Semiconductor (hardware chips): Taiwan Semiconductor Manufacturing Company, Samsung, Intel, NVIDIA, Qualcomm and Broadcom |
| 3. | AI Development Layer | This layer is where AI models and algorithms are developed, including:
|
Google, Microsoft, Meta, Amazon, OpenAI |
| 4. | Generative AI/ Foundation Model Layer | Generative AI allows users to generate new content such as images, music, text and code, based on the patterns it has learned. This is used in applications such as ChatGPT for human-like text generation, AI generated images and code.
It is basically a model ‘pre-trained’ on large amounts of data, which often uses the ‘Generative Pre-trained Transformer’ (“GPT”) architecture to generate text. Foundation models are the building blocks for a wide variety of AI applications. |
Microsoft, OpenAI,2 Meta, Google, Anthropic (globally); Observe.AI, Pixis, Ola Krutrim, InVideo, Sarvam AI, Avaamo AI, Senseforth.ai (in India).3 |
Downstream AI segment
The Downstream AI segment has three layers:
| Sr. No. | Layer | Function | Major players |
|---|---|---|---|
| 1. | AI Model Layer | Here, foundational models/ basic AI systems are fine-tuned and adapted to specific industries and use cases, including the training of models using sector specific data (for example, retail, logistics, etc.) | Google, Microsoft, Meta, Amazon and OpenAI. |
| 2. | AI Release and Deployment Layer | This layer includes the deployment and distribution of AI models to businesses, including the development of proprietary AI models in-house for specific business needs, adapting open-source AI models for specific business needs, helping teams operationalize open-source models and integrating large language models with enterprise data. | Key Indian players include Sigmoid (end-to-end model lifecycle platforms), Cumin AI (scalable batch APIs and managed vector stores), Inferless (serverless GPU platform) and produktiv.ai (governance-linked retrieval-augmented generation tools). |
| 3. | AI User Interaction Layer | This layer includes (i) customer-facing AI such as chatbots and recommendation engines that improve engagement; (ii) operational AI for real-time decision-making in relation to logistics and warehouse management; (iii) contextual AI that adapts to changing conditions (like traffic or demand); and (iv) personalization engines which customize experiences based on user preferences. | Notable Indian startups include Haptik (enterprise-grade chatbots), Yellow.ai (omnichannel virtual agents), Engati (no-code chatbot deployment), along with firms such as Sketcha and Frontitude pushing AI-based design automation. |
There is also a governance and orchestration layer that spans across the layers of the AI stack. It serves to, inter alia, coordinate different AI models, enterprise integration and workflow automation and performance optimization.
Uses in Various Industries
The Report analyzes certain specific industries, identifying the role of AI in such industries:
| Industry | Role of AI |
|---|---|
| Retail |
|
| E-commerce |
|
| Logistics and Delivery |
|
| Marketing |
|
| Banking, Financial Services and Insurance |
|
| Healthcare |
|
Potential Competition Issues
Based on the above uses of AI, the Report identifies certain potential competition issues that may arise due to the use of AI:
Algorithms to determine pricing or supply
AI and AI-powered algorithms that determine pricing can independently learn to co-ordinate prices and monitor competitors’ actions (i.e., the elements of a cartel) without human intervention, replacing explicit collusion with tacit collusion. Further, such algorithms could be used to co-ordinate matters other than pricing such as supply, bidding strategies and market allocation. Such algorithms could potentially increase market transparency to the extent that changes in prices can be anticipated or calculated by competitors (through algorithms) in real-time, facilitating collusion.
The Report specifically highlights risks posed by the following categories of algorithms:
- Monitoring algorithms: Such algorithms capture and track data relating to demand and pricing through scraping or other methods, and act as data collection tools for pricing decisions. Typically, a monitoring algorithm would require explicit human collusion to lead to any potential competitive harm.
- Parallel algorithms: Parallel algorithms (or ‘hub and spoke’ algorithms) involve several competing entities using a common, third-party algorithm (i.e., the “hub”), which sources data from each entity (i.e., the “spokes”) to devise their respective pricing algorithms. Such collusion may lead to alignment in prices, since the hub (i.e., the common third-party provider) which creates the pricing algorithm for each spoke, has access to the data and pricing patterns of the other spokes.
- Signalling algorithms: Such algorithms respond to market conditions using sophisticated statistical models to set and align prices, often above a specific equilibrium.
- Self-learning algorithms: Such algorithms operate autonomously and use deep-learning to dynamically respond to market changes. No explicit collusion or human intervention is required by such algorithms.
Problematic conduct by dominant entities
The Report also identifies potential exclusionary and exploitative conduct by dominant entities:
- Self-preferencing: Self-preferencing (i.e., the practice of favoring one’s own goods or services over others) may be facilitated by AI-powered tools. For example, a dominant entity’s AI-powered platform may be manipulated to display the dominant entity’s products at the top of search rankings, or to promote its own AI solutions.
- Predatory pricing: AI-driven predatory pricing involves offering prices at levels below cost with the intent to oust competitors, using AI systems that may monitor competitor and consumer behavior in real-time to adjust prices. Such predatory pricing would be at a significantly higher level of speed, scale and precision compared to ordinary predatory pricing.
- Tying and bundling: Tying (i.e., the practice of making the purchase of one product subject to another) or bundling (i.e., providing multiple products together to consumers) may be further exacerbated by AI. AI-enabled algorithmic targeting may target less price-sensitive consumers with bundles at a higher price, whereas discounted bundles may be provided to more price-sensitive customers.
- Pricing discrimination: By leveraging real-time consumer data using AI (such as browsing behavior, location, device type), dominant firms could identify and segment customers, and tailor prices according to the customer’s willingness to pay and implement price discrimination more effectively.
Pricing practices
AI allows personalization of pricing and targeted marketing and search results to a very high degree since it enables factoring in a large volume of data regarding consumers and their characteristics. The Report also highlights the possibility of ‘selective pricing’, where enterprises may nudge users of competing platforms/products to their offerings by selectively identifying users and pricing offerings accordingly.
While such practices could increase efficiencies, competition authorities may be concerned about such practices in markets where there is insufficient competition, particularly where such practices erode consumer trust in online markets or increase search and transaction costs for consumers.
Entry barriers
The Report highlights certain entry barriers to the AI industry such as (i) data availability; (ii) talent availability; (iii) cost of cloud services and access to cloud services; (iv) access to funding; (v) cost of computing facility and access thereof; (vi) technological complexity; and (vii) intellectual property rights of vendors.
Reduced transparency
In cases where smaller entities such as start-ups rely on AI models and infrastructure provided by a select few larger players, the Report notes that the lack of visibility into the operation of such systems, including opaque APIs, black-box algorithms and unclear pricing structures, may lead to uncertainty and dependency, and could also impede the ability of smaller players to compete effectively.
Network effects
Markets for AI models or generative AI have broad downstream application in diverse industries. This could create network effects4 where the value of the AI model increases with the number of downstream applications for the model, or with the volume of data collected. The Report cites social media platforms as an example of how platforms may benefit from network effects of a growing user base.
Recommendations and the Way Forward
The Report recommends the following in order to minimize the risk of anti-competitive harm on account of the use of AI:
- Competition law based self-audits: Self-audits and competition compliance programmes should be conducted by businesses deploying AI systems to prevent anti-competitive harm. This includes documenting AI-based decision-making processes, designing and testing algorithms to prevent anti-competitive outcomes, periodic review of algorithms and AI-driven pricing strategies, implementation of safeguards using third-party tools and enforcing measures to prevent the sharing of commercially sensitive information. There is also additional guidance on specific points around which self-audits should be conducted.
- Transparency measures: Entities using AI systems should communicate the following clearly to relevant stakeholders: (i) the use and purpose of deployment of AI in decision-making, (ii) parameters for AI based decisions and (iii) other information which may increase transparency in this regard in clear and intelligible language (without disclosing any proprietary information).
- Removal of entry barriers: The Report recommends reducing entry barriers in the AI industry through access to better infrastructure, promotion of open-source frameworks and access to high-quality data to compete with AI-driven innovation and development of a skilled workforce.
The above recommendations represent preliminary steps towards understanding and regulation of AI. In particular, requiring self-audits as a first level of check may be beneficial in preventing anti-competitive conduct. Companies deploying AI systems in their decision-making process as well as AI service providers may be required to be vigilant and routinely ensure that their systems and policies are compliant with Indian competition law on the advice of counsel.
1 Prepared with the assistance of the Management Development Institute, Gurgaon.
2 Based on reports in the public domain, Microsoft is one of the investors in OpenAI.
3 While the Report does not specify OpenAI and Anthropic (known for their GPT and Claude foundation models, respectively) as ‘major players’ for this layer, at present these two players command a significant share of the market for generative AI models.
4 Network effect is a phenomenon where the number of users of a service increases incrementally over time as (i) more and more users flock to the service or (ii) more providers are available on the other side of a platform (for example, a platform with more business users/ app developers/ sellers tends to attract more consumers and vice versa).
This insight has been authored by Simran Dhir (Head of Competition Law Practice), Prerana De (Principal Associate) and Ritik Mohapatra (Associate). They can be reached on sdhir@snrlaw.in, preranade@snrlaw.in and rmohapatra@snrlaw.in, respectively, for any questions. This insight is intended only as a general discussion of issues and is not intended for any solicitation of work. It should not be regarded as legal advice and no legal or business decision should be based on its content.
© 2026 S&R Associates