Competing to Innovate: How Competition Accelerates AI Innovation

Abstract

The panel examined how the concentration of AI‑related assets (data, compute, foundational models, and distribution channels) in the hands of a few global corporations threatens the innovation ecosystem of the Global South. Panelists discussed the structural “dependency” problem, the limits of current competition enforcement, and the promise (and pitfalls) of an open‑source, locally‑tailored AI stack. They contrasted industrial‑policy‑driven approaches with competition‑law tools, debated the role of sovereignty versus ownership, and suggested concrete regulatory and market‑level remedies to foster a more competitive, inclusive AI future for India and other emerging economies.

Detailed Summary

Moderator (Payal Malik) framed the discussion around the “forgotten” but critical role of competition in AI innovation.
She highlighted historic platform‑dominance issues (search, e‑commerce, app stores, cloud) and introduced the concept of “dependencies”—the reliance of advertisers, developers, and consumers on a handful of platforms.
The moderator argued that the same pattern is now repeating in the AI stack: concentration in hardware, cloud, foundational models and distribution. She emphasized that data hegemony is the underlying driver of this dominance.

2. Concentration Risks in the AI Supply Chain

Speaker	Main Points
Augustine Peter	• Confirmed concentration risk in India and globally – ~92 % of AI‑related resources are controlled by firms from the United States and China. • Described India’s “last‑mile” position: a massive consumer base but negligible presence in early AI pipeline stages (data, model training, compute). • Highlighted vertical integration of large firms, “killer acquisitions,” and the slow procedural timeline of the Competition Commission of India (CCI) – investigations can take years, by which time the market may have shifted.
Amba Kak	• Traced market power back to the past decade, not just the rise of ChatGPT. The foundational‑model layer remains the most capital‑intensive and therefore the biggest barrier to entry (massive compute, chips, data). • While downstream applications appear open (“a thousand flowers”), big tech controls distribution, monetisation pathways, and can cannibalise downstream value (e.g., equity‑stake deals with hospitals). • Warned that reliance on a handful of hyperscalers could let them “squeeze” startups via high fees and self‑preferencing.
Shweta Rajpal Kohli	• Described the Indian startup ecosystem’s enthusiasm for AI applications but noted that the operating‑system / app‑store layer remains tightly held (default search → default “chat” interface). • Questioned the 30 % transaction fee imposed by platform app‑stores and asked whether the relationship with big tech is symbiotic or exploitative.
Kush Amlani	• Focused on the lower layers of the stack (OS, browsers, search, app stores) that directly affect consumers. These markets are already “tilted” – a “wonky house” built on uneven foundations. • Stressed that openness should be a floor, not a ceiling: open‑source AI alone does not solve structural dependencies on infrastructure or distribution. • Warned against “open‑washing” where big‑tech brands open‑source components to mask continued dominance.
Payal Malik (moderator)	• Asked about the practical impact of industrial policy (government‑led compute and data initiatives) versus competition policy in addressing sovereignty and agency for Indian innovators.

3. Sovereignty, Openness, and the Role of Industrial Policy

Amba Kak introduced the concept of “open‑weighted models” as a potential tool for U.S. hegemony (citing the Trump administration’s AI export order that encourages foreign governments to build on U.S. AI stacks). She argued that openness can be weaponised to lock other nations into a US‑centric ecosystem.
Shweta Rajpal Kohli explained India’s dual‑track approach: building strategic, sovereign compute while simultaneously allowing foreign investment. She noted recent Indian initiatives (e.g., the “Sarbam” frontier model) that aim to tailor AI to vernacular languages and local data.
Kush Amlani reiterated that interoperability, open standards, and user choice are essential for genuine sovereignty. He cautioned that merely releasing open‑source code without addressing the distribution and infrastructure layers leaves developers dependent on the same dominant platforms.

4. Competition‑Policy Toolbox: Ex‑Ante vs. Ex‑Post

Speaker	Suggested Instruments
Augustine Peter	• Accelerate the formation of an AI‑focused team within the CCI. • Use existing Section 33 interim measures to act quickly on anti‑competitive behaviour. • Avoid overly bureaucratic “ex‑sui” processes that delay enforcement.
Kush Amlani	• Propose structural separations at the infrastructure layer (e.g., cloud providers cannot also own dominant AI‑model marketplaces). • Adopt narrow, proportionate, technology‑neutral rules that can be updated swiftly (inspired by the EU Digital Markets Act but tailored to India).
Amba Kak	• Treat openness as a regulatory floor – mandatory baseline reuse and transparency, while ensuring that deeper dependencies (distribution, data pipelines) are also monitored.
Shweta Rajpal Kohli	• Emphasised policy coordination: industrial policy to build domestic talent and compute capacity, coupled with competition oversight to prevent anti‑competitive bundling or data‑locking.
Payal Malik	• Highlighted the need for public‑private dialogue (e.g., the recent “deep‑fake” rapid‑response rule) as a model for agile regulation.

5. The Startup Perspective

Shweta Rajpal Kohli noted that many Indian founders are eager to collaborate with global giants (OpenAI, Nvidia, Google) rather than fearing dominance. Their primary goal: access cutting‑edge models and compute.
Kush Amlani warned that 75 % of startup funding in the previous year came from the three hyperscalers (Google, Microsoft, Amazon), and that acquisition pathways often lead startups into the hands of the same dominant firms, reducing long‑term competition.

6. Audience Q&A – Key Themes

Model Access vs. Platform Control – An audience member questioned why discussions continue around foundational models when platforms like OpenAI and Gemini already provide free services in India. Panelists responded that access does not equal control; the underlying distribution and data‑ownership mechanisms still pose competition concerns.
Speed of Enforcement – Several participants highlighted the slow pace of Indian competition proceedings (average 3‑4 years from filing to final order) and argued for ex‑ante interventions to avoid “too‑late” remedies.
Regulatory Inspiration – References were made to Japan’s targeted law on operating‑system competition and the EU Digital Markets Act as possible templates, but all agreed that India needs a home‑grown, flexible framework.

7. Concluding Reflections

The panel converged on the view that competition, openness, and sovereign capacity must be jointly pursued.
While industrial policy can fund compute and talent, competition law must safeguard market entry, prevent bundling, and ensure interoperable standards.
The discussion underscored the urgency of proactive (ex‑ante) regulation to keep AI innovation inclusive, especially for the Global South.

Key Takeaways

Concentration risk is acute: ~92 % of AI resources are controlled by firms from the United States and China, leaving India and other Global‑South economies at the periphery of the AI stack.
Data hegemony fuels platform dominance: Control over large data sets underpins the market power of the few hyperscalers.
Foundational‑model layer is the biggest barrier: Massive compute and capital requirements create a choke point that limits new entrants.
Open‑source is necessary but not sufficient: Openness should be a regulatory floor; without addressing distribution, infrastructure and data‑pipeline dependencies, open‑source AI can become “open‑washing.”
Ex‑post competition enforcement is too slow: Current CCI processes can take years, by which time the market may have already tipped.
Ex‑ante tools are essential: Structural separations, rapid interim measures (e.g., Section 33), and narrow, technology‑neutral rules can provide timely safeguards.
Industrial policy must complement competition law: Building sovereign compute, talent pipelines, and vernacular models is valuable, but must be paired with strong antitrust oversight to avoid lock‑in.
Start‑up dynamics are mixed: While founders seek collaborations with big tech, the concentration of funding (75 % from Google/Microsoft/Amazon) and acquisition pathways risk reducing long‑term competition.
Interoperability and open standards are the true enablers of sovereignty: Users and developers need the ability to move data and models across platforms without restrictive bundling.
Regulatory design should be India‑specific: Lessons can be drawn from the EU DMA, Japan’s OS law, and US practices, but any framework must fit India’s unique market size, policy environment, and development goals.

See Also:

India AI Impact Summit 2026

Explorer

competing-to-innovate-how-competition-accelerates-ai-innovation