Democratizing AI Resources Equitable Access to Compute and Data for Entrepreneurship
Abstract
The panel examined what “democratizing AI” really means for India. Four inter‑linked pillars were debated: (1) open, ethical access to data, (2) affordable compute infrastructure, (3) development of foundational models & marketplaces, and (4) the broader policy, standards and ecosystem required to sustain entrepreneurship. Speakers highlighted current gaps—in data‑governance, compute availability, talent mobilisation, and regulatory frameworks—while offering concrete suggestions for short‑, medium‑ and long‑term actions that could turn India into a genuine AI‑prosuming nation.
Detailed Summary
- Amit Srivastava opened by stating the short answer to “Are we democratized yet?” is no, but the more important question is how we plan to get there.
- He identified three “pillars” of democratization: data, compute, and models/marketplaces.
- The moderator stressed that democratization is not just a technical issue; it is also an ethical, regulatory and geopolitical one.
2. Pillar 1 – Ethical & Accessible Data
| Speaker | Main Points |
|---|---|
| Amit | Data goes through five stages (creation → flow → security → processing → storage). Current consent mechanisms are a blunt “agree‑to‑terms” checkbox; there is no transparent framework for ethical reuse. |
| Dr. Subi | India’s data‑harvest landscape is already hostile—both state and non‑state actors scrape data. A national policy must define ownership, stewardship, and permissible uses (e.g., homomorphic encryption for privacy‑preserving model training). |
| Animesh | Data is more valuable than oil. Raw data is a public good, but raw‑data leakage can create security threats. A “national movement” akin to Pulse Polio is needed to raise data‑literacy and to institutionalise open‑data standards. |
| Nitin | Data‑centric AI must balance privacy vs. utility. Examples (COVID‑tracking, health diagnostics) illustrate how personal data, when shared responsibly, saves lives. He called for “trusted corridors”—government‑certified pipelines that guarantee provenance and mitigate poisoning. |
Announcements & Findings
- Over 13,000 Indian datasets have already been released under public‑domain licences.
- The DPDP (Data Protection Personal Data) law now forces apps to request fresh consent when purpose changes.
Open Questions
- Pricing of data: Should data be monetised for its creator? How to prevent “data‑as‑oil” exploitation?
- Regulatory scope: What categories of data (genomic, financial, location) require strict gate‑keeping?
3. Pillar 2 – Compute Infrastructure
| Speaker | Main Points |
|---|---|
| Amit | India’s AI Mission announced ₹280/hr GPU rentals (≈38,000 GPUs). This is only a starting point; sustainable compute needs a “highway‑level” approach—national data‑centers, shared compute pools, and policy‑driven procurement. |
| Animesh | Early‑stage startups can experiment on “classical” CPUs, but refined models need premium hardware (e.g., NVIDIA H100, A200). The cost barrier remains high. |
| Dr. Subi | Quantum computing is still nascent; access is limited to a few qubits. However, AI‑for‑quantum research could accelerate both fields. |
| Nitin | The GERD (Gross Expenditure on R&D) should rise to 1‑3 % of GDP to fund compute ecosystems. He warned that the current “compute‑as‑a‑service” model places startups in a profit‑leak loop: revenue → compute fees → net loss. |
Announcements & Findings
- The Semiconductor Mission is underway, but full‑scale chip‑fabrication and wafer‑packaging are still “years away”.
- Quantum‑AI labs are emerging (e.g., partnerships with NVIDIA’s quantum‑AI team).
Recommendations
- Build government‑backed shared compute clusters that can be accessed by startups at cost‑price.
- Encourage open‑hardware accelerators to reduce reliance on a single vendor (NVIDIA).
4. Pillar 3 – Models, Marketplaces & the “AI Stack”
| Speaker | Main Points |
|---|---|
| Animesh | Large‑scale foundation models are “equalizers”. India must develop Bharat‑GPT‑type models tuned to local languages and contexts. |
| Dr. Subi | Model development must be coupled with ethical guardrails—bias audits, watermarking, and transparent training data provenance. |
| Nitin | The “AI stack” (hardware → framework → model → app) must be modular so that innovators can swap layers (e.g., use an open‑source model on a public compute node). |
| Amit | Marketplaces need standardised APIs and interoperability (similar to the internet’s 7‑layer model). This will lower friction for third‑party developers. |
Data & Findings
- Open‑source LLMs are now freely downloadable, but compute costs dominate.
- The AI‑city pilots (Uttar Pradesh, Andhra Pradesh) are testing shared compute labs and localized model “incubators”.
5. Policy, Standards & Global Positioning
| Speaker | Main Points |
|---|---|
| Dr. Subi | International bodies I‑STAR, ISOC, ITU set internet‑ and telecom‑standards. India must be present in those forums to avoid “digital colonisation”. |
| Nitin | Regulatory sandboxes (e.g., at NIC, I‑AM Lucknow) allow startups to experiment under supervised conditions. |
| Animesh | The “AI‑city” concept must include tax holidays (until 2047) for data‑center construction, plus incentives for Indian‑owned hardware. |
| Amit | Data sovereignty is a national security issue. Policies must address DPDP compliance, data localisation, and cross‑border data flows. |
Announcements
- The central government announced tax holidays on AI‑infrastructure to 2047.
- IAM Lucknow is establishing a blockchain excellence centre to complement AI research.
Open Questions
- How to balance foreign investment (e.g., NVIDIA) with Atmanirbhar (self‑reliant) goals?
- What will be India’s stance in upcoming USTR trade negotiations concerning AI‑related services?
6. Entrepreneurship, Talent & Ecosystem
| Speaker | Main Points |
|---|---|
| Nitin | India produces 40 % of the world’s AI talent but this talent often works for foreign firms. Creating local IP is essential. |
| Amit | “McDonaldization” of data‑centers: make compute as ubiquitous as fast‑food outlets. |
| Animesh | Start‑ups need co‑founder matching, mentorship, and network capital; these can be institutionalised via Y‑Combinator‑style accelerators. |
| Dr. Subi | Upskilling teachers, professors, and policymakers is required to sustain the pipeline of AI innovators. |
Recommendations
- Government should procure shared‑compute licences for universities and incubators.
- Public‑private consortia (e.g., TAQBIT + ConstemsAI) can create industry‑academia bridges for proof‑of‑concept projects.
7. Audience Q&A – Highlights
| Question (paraphrased) | Answer Summary |
|---|---|
| Can individuals earn royalties from their personal data? | Dr. Subi: Current licences usually surrender that right; a future regulatory framework could create “data‑ownership royalties”, but ethical & enforcement challenges remain. |
| What is the role of quantum computing in India’s AI roadmap? | Animesh: Quantum computers are still a research tool; the immediate focus should be on AI‑for‑quantum to accelerate chip design and material science. |
| How do we price data and protect it from misuse? | Nitin: Introduce trusted data corridors, enforce privacy‑by‑design and data‑poisoning detection; pricing should reflect social value rather than pure market rate. |
| Should internet be treated as a utility like water? | Amit: Yes—treating broadband as a basic service lowers the entry barrier for AI adoption, especially in rural districts. |
| What incentives are needed for Indian hardware vendors? | Animesh: Provide R&D tax credits, subsidised fab‑line access, and priority procurement for Indian‑made AI accelerators. |
8. Closing Remarks
- Amit Srivastava reiterated the need for a holistic, multi‑layered strategy—data, compute, models, standards, and talent—all coordinated by public policy.
- The panel thanked the audience and announced a group photograph and a memento presentation to the speakers.
Key Takeaways
- Democratization is incomplete; India must move from “candy‑store access” to a full‑stack ecosystem.
- Data must be treated as a national public good, with transparent consent, open‑data standards, and mechanisms for privacy‑preserving reuse.
- Compute access remains the biggest bottleneck; shared, government‑backed clusters and tax incentives for data‑centers are essential.
- Local foundation models (e.g., Bharat‑GPT) are crucial for linguistic diversity and reducing dependence on foreign APIs.
- Policy engagement is non‑negotiable – India must claim seats in ISOC, I‑STAR, ITU, and USTR to shape global AI standards and avoid digital colonisation.
- Talent is abundant but under‑utilised; systematic upskilling, mentorship, and start‑up incubators will convert talent into IP.
- Quantum computing is a medium‑term lever; the immediate focus should be AI‑for‑quantum research rather than expecting quantum advantage today.
- Funding and R&D spend must rise to 1‑3 % of GDP to sustain long‑term AI infrastructure and innovation.
- A multi‑stakeholder “flywheel” model—government procurement, private‑sector hardware, open‑source software, and startup innovation—will create a self‑reinforcing AI ecosystem.
- Ethical and legal frameworks (DPDP, data‑ownership royalties, trusted data corridors) are needed to balance privacy, security, and economic value.
Prepared by the AI Conference Summarisation Team, Delhi, 24 Feb 2026.
See Also:
- ai-for-economic-growth-and-social-good-ai-for-all-driving-economic-advancement-and-societal-well-being
- ai-innovators-exchange-accelerating-innovation-through-startup-and-industry-synergy
- building-resilient-sustainable-ai-infrastructure-for-people-planet-and-progress
- democratizing-ai-compute-and-digital-data-infrastructures