Competition and AI: How Innovation Accelerates at Scale

Contents

Executive Summary

This panel discussion examines how market concentration risks in AI—particularly dominated by a small number of US and Chinese firms—threaten innovation and sovereignty in developing economies like India. The speakers argue that while open-source AI and industrial policy are necessary, they must be paired with forward-looking regulatory frameworks (ex-ante regulation) and targeted structural interventions to prevent the replication of digital platform dominance patterns from the previous technology cycle.

Key Takeaways

Concentration in AI mirrors the pattern from previous technology cycles (search, social media, e-commerce). Without early intervention, the Global South will become a "feature controlled by the platform" rather than an independent innovator.
Openness and industrial policy are necessary but not sufficient—they must be paired with ex-ante regulatory frameworks that allow rapid, proportionate intervention to preserve contestability and prevent structural tipping.
Interoperability and user choice, not just open-source code, are the real mechanisms for democratization. Regulation should target the distribution layer (operating systems, browsers, app stores) and infrastructure layer, not just foundational models.
India's government is committed to both attracting investment AND enforcing competition law—these are not mutually exclusive. Sovereignty means having the legal and regulatory capacity to shape the technology ecosystem, not rejecting foreign technology.
Awareness and responsibility across all actors (policy makers, regulators, startups, consumers) is critical now, because the regulatory process is too slow to catch harm after it occurs. Community-level choices about which platforms and tools to adopt send market signals that can influence corporate behavior.

Key Topics Covered

Market concentration risks in AI across the technology stack (hardware, cloud, foundational models, distribution)
Data hegemony as the underlying driver of platform dominance
Digital sovereignty and what it means for the Global South and developing nations
Regulatory approaches: comparing ex-post competition enforcement vs. ex-ante regulation
Open-source AI and openness as necessary but insufficient solutions ("openness is the floor")
Industrial policy vs. competition policy: whether they are antithetical or complementary instruments
Startup ecosystem dynamics: collaboration vs. competition with big tech; acquisition as pathway to success
Infrastructure dependencies: cloud compute, distribution channels (app stores, browsers, search engines)
India's strategic response: building sovereign compute capacity, supporting vernacular/Indic language models
Lessons from previous cycles: how platform dominance emerged in search, social media, e-commerce, and app stores
Remedies and legal instruments: targeted regulations, interoperability requirements, structural separation
US foreign policy and AI: using open-source models to maintain US hegemony

Key Points & Insights

Concentration is real and structural: 92% of AI resources and control originate from the US and China. India and the Global South occupy only the "consumer interface" layer, with no presence in AI protocol or foundational layer development.
Vertical integration and acquisition as competitive tactics: Large firms use acquisition ("killer acquisitions") and vertical restraints to prevent competition. The current competition enforcement framework is too slow to address this—cases take 3-4 years from filing to final order, while AI markets shift dramatically in months.
The "collaboration trap": While Indian startup founders express enthusiasm about partnerships with OpenAI, Google, and Microsoft, 75% of startup funding last year came from these three hyperscaler companies, who also control the infrastructure. Partnerships often come "without strings attached" regarding data, bundling, and revenue sharing.
Distribution layer is already captured: Operating systems, browsers, search engines, and app stores—the layer that directly impacts consumers—have already "tipped" to a handful of players over the past two decades. The AI stack is being built on these "wonky foundations."
Openness cannot be a silver bullet: Open-source models and open-weighted models are necessary for democratization and reusability, but they obscure deeper structural dependencies at the distribution and infrastructure layers. The US is strategically promoting open-source AI through export policies to maintain global dependence on the US stack.
Sovereignty requires agency, not just ownership: India's approach—combining industrial policy with competition policy—aims to give startups and users "agency" without necessarily requiring full ownership. Building indigenous compute capacity, local cloud options, and locally-trained models are essential components.
Ex-ante regulation is urgently needed over ex-post enforcement: The traditional approach of competition authorities investigating and penalizing dominant firms after harm occurs is too slow. By the time remedies are imposed, markets have already tipped and innovation opportunities have been lost. Forward-looking, proportionate regulation is needed.
The "one thing" approach: Rather than comprehensively rewriting competition law, targeted, narrow regulatory interventions (like Japan's law on operating systems) may be more effective. Key principles should include interoperability, open standards, user choice, and encouragement of open-source solutions at every layer.
India is building its own path, not copying: The Indian government is clear it will not adopt the European DMA model wholesale or follow the US approach. Instead, it is developing AI regulation suited to India's context—supporting vernacular language models, voice AI, and Indian datasets. The Saram frontier model launch represents early evidence of indigenous capacity.
Capital intensity creates concentration risk: Modern AI innovation requires massive capital expenditure (compute, infrastructure, data). This structural fact means that access to capital becomes the primary determinant of who can innovate, automatically creating concentration risks and favoring incumbent firms.

Notable Quotes or Statements

"If the base operating system and app stores are controlled still by a handful of companies and the infrastructure that runs AI models sits also on these handful hyperscalers, the global south risks becoming a feature controlled by the platform." — Opening speaker

"A competitive opensource AI stack built on affordable compute, local cloud options and locally trained model is essential for innovation." — Opening speaker

"75% of total startup funding came from three companies Google, Microsoft and Amazon who also happen to be the hyperscalers that control the infrastructure in this market." — Amba (AI Now Institute)

"It's woken up way too late and digital markets have already tipped as we said earlier... what's needed is exanti regulation in this space." — Kush (Mozilla)

"Openness is the floor but also like don't let it become the thing that makes you look away from from these more structural dependencies." — Amba

"Open weighted models could just as well be a tool for reinforcing and spreading US hegemony." — Amba, referencing Trump's AI export order

"We're going to do both. We will do everything it takes to build up our own infrastructure our own AI infrastructure... at the same time... India is going to do AI the India way and that will not be the US way it will not be the China way." — Shwara (Startup Policy Forum India), paraphrasing government position

"If you're not at the table, you're on the menu... I don't think you want to sit at that table, right? Like I think what you're saying is India's way and the way of countries in the global majority can be to build their own tables right." — Kush

"It's my job... to speak to founders and say you know what this is great I'm supporting you... but you know what watch out because if all of us don't do that then we're not doing the right service." — Shwara

"There is no match. A number of uh uh number of uh lot of efforts have been made but in the case of competition authority... any issue is raised it takes long time." — Augustine Peter (former CCI member), on India's regulatory capacity gaps

Speakers & Organizations Mentioned

Panelists:

Augustine Peter — Former member, Competition Commission of India (CCI)
Shwara Rajpal Kohli — Founder-Director, Startup Policy Forum India
Amba Kak — Co-director, AI Now Institute (US)
Kush Amlani — Head of Competition Law and Public Policy, Mozilla

Other Individuals/Officials Referenced:

Narendra Modi — Prime Minister of India
Ashwini Vashna — Minister (India)
Secretary Krishnan — Government of India
Sam Altman — OpenAI
Mark Carney — Reference to his World Economic Forum speech
Chair Khan — FTC (Federal Trade Commission)
Judge Meta — Reference to Google search decision
Mr. Vinodal — Mentioned as having led committees on competition policy

Companies/Platforms:

OpenAI, Google, Microsoft, Amazon, Apple, Meta, Nvidia
Mozilla (Firefox)
Zoho (Indian software company)
Salesforce, Uber (mentioned as context for panelist experience)
WhatsApp

Government/Regulatory Bodies:

Competition Commission of India (CCI)
Federal Trade Commission (US)
UK Competition and Markets Authority (UKCMA)
European Union regulators

Geographic/Regional References:

India, Global South, United States, China, European Union, Japan, UK

Technical Concepts & Resources

Foundational/Large Language Models (LLMs) — The core of AI concentration; barriers to entry include compute costs, data requirements, and capital
Small Language Models (SLMs) — Alternative approach that may be more viable for resource-constrained regions
Vernacular/Indic Language Models — Indian government strategy to build models for Hindi, Tamil, Telugu, and other Indian languages
Voice AI — Identified as a key frontier for India
Frontier Models — State-of-the-art AI models (e.g., Saram, referenced as an Indian frontier model launching at the summit)
Open-Source/Open-Weighted Models — Models released with weights available for reuse (e.g., Meta's Llama)
Digital Public Infrastructure (DPI) — India's approach to providing sovereign infrastructure (e.g., UPI payments system model applied to AI)
Interoperability — Ability for different systems to work together; identified as critical remedy
Vertical Integration — When a single firm controls multiple layers of the supply chain (hardware, cloud, models, distribution)
Killer Acquisitions — Strategic acquisitions of potential competitors to eliminate threats
AI Overviews — Google's AI-generated summaries in search results; cited as example of market tipping without adequate competition review
Data Hedgemony — Control over data as the underlying driver of platform dominance
Ex-Ante Regulation — Proactive regulatory intervention before harm occurs (contrast: ex-post enforcement after harm)
Digital Markets Act (DMA) — EU regulation referenced as a model (though not to be copied wholesale)
GDPR and DPDP — Data privacy frameworks mentioned as precedent for India's approach to regulation
Structural Separation — Regulatory remedy preventing a firm from competing in multiple adjacent layers

Key Policy/Regulatory References:

India's startup India initiative
Japan's law targeting operating systems
US Trump AI export order (promoting open-source models as US policy tool)
India's Saram frontier model (launching during the summit)

Production Notes:

This transcript contains significant repetition and some audio quality issues (indicated by "the picture. the picture. the picture." type artifacts)
Some speaker attributions are ambiguous in the original; identified speakers based on context
Discussion moves between abstract competition theory and concrete examples from previous platform dominance cases (Google Search, app stores, etc.)
Tone is balanced but advocates for proactive regulatory intervention; speakers acknowledge trade-offs between innovation, investment, and competition/sovereignty