Planet-Scale Intelligence for Economic Growth: AgriStack, Climate AI, and Foundational Models for Global Impact

Detailed Summary

• Prateep Basu set the stage, arguing that AI must move from isolated pilots to reliable, continuously operating systems embedded across agriculture, climate, energy, telecom, finance, and public institutions.
• She stressed four intertwined system requirements: talent, elastic infrastructure, sustainable energy, and trust.
• The moderator framed AI as the next layer of digital public infrastructure (comparable to UPI for finance) that can enable “intelligence inclusion” at population scale.

2. Panelist Introductions

Panelist	Core Focus
Gaurav Aggarwal (Reliance Jio) – national‑scale AI deployment, latency, reliability, data governance.
Arun Sharma (Rezonia) – large‑scale power transmission infrastructure that underpins AI compute.
Derick Jose (Accenture) – operationalizing AI in asset‑heavy environments; converting pilots into operating systems.
Michael Tsan (Dalberg) – intersection of technology, policy, and inclusive development; climate‑smart credit, risk‑informed public spending.
Gaurav Kataria (ITC) – manufacturing and supply‑chain risk management under climate stress.

3. Key Themes Explored

3.1. Investment Priorities & Systemic Gaps (Michael Tsan)

• High‑impact sectors: geospatial intelligence for energy‑transition planning, food‑system surveillance, and climate‑resilient infrastructure.
• Scale of loss: ≈ $800 bn annual economic damage to infrastructure from climate disasters; ≈ 1 bn people facing severe food insecurity; ≈ 300 mn experiencing extreme hunger last year.
• Three principal gaps:
  1. Technical capacity & talent – ability to sense data exists, but sense‑making (analytic capacity) is lacking.
  2. Institutional incentives & decision‑making – fragmented ministries, no single empowered budget holder, procurement not fit for AI‑scale projects.
  3. Quantification of opportunity cost – governments struggle to compare AI investment vs. one‑off surveys; lack of robust economic case.

3.2. Sovereign AI Infrastructure & Affordability (Gaurav Aggarwal)

• Cost barrier: Current AI‑as‑a‑teacher costs ₹10 /min (≈ ₹2 400 for a four‑hour session) – prohibitive for most Indian households.
• Target affordability: Aim for ₹10–30 paise per minute (≈ 100× cheaper) so AI services could be cheaper than a glass of milk.
• Why massive, locally owned infrastructure matters:
  • Energy, data‑center capacity, GPU supply, and AI frameworks must be sovereign to drive down cost.
  • India’s price‑sensitivity and wide mobile‑penetration make a public‑good AI stack essential.
• Broader vision: Apply the same model that made UPI ubiquitous to AI, making it open, interoperable, and accessible across domains (energy, finance, climate).

3.3. Scaling Industrial AI – Trust, Incentives & Learnability (Derick Jose)

• Case study – Mayonnaise plant (Poland): AI recommended set‑points to reduce defects.
  • Trust building: Spent seven months benchmarking human vs. AI recommendations; operators needed to see the data.
  • Incentive alignment: Production bonuses tied to reduced rejects encouraged operator adoption.
• Three pillars of scale:
  1. Reliability & trust – system must consistently perform.
  2. Incentive design – operators must profit directly from AI use.
  3. Learnability – AI should improve over time; early performance isn’t decisive, but a high learning rate is.
• Human factor: “Agents are like children – they need to be given time to crawl before they can run.”

3.4. AI as an Augmentation Layer in a Diversified Conglomerate (Gaurav Kataria, ITC)

• Process‑first approach: Redesign business processes before inserting AI; avoid “AI for AI’s sake”.
• Key use‑cases across ITC:
  • Supply‑chain risk mitigation (COVID‑era disruptions, canal blockages).
  • Regulatory compliance (EU deforestation rules – mapping wood provenance via satellite).
  • Marketing & advertising (AI‑generated digital avatars, hyper‑local ad targeting).
  • New product discovery (identifying unmet adult‑care segment → “Right Shift”).
• Three outcome categories:
  1. Efficiency & productivity gains (bottom‑line impact).
  2. New revenue streams / business models.
  3. Societal impact (sustainability, shop‑floor safety).
• Change‑management emphasis: People, process, and technology must move together; quality is the non‑negotiable cornerstone.

3.5. From India to the World – Transferability of AI Solutions (Michael Tsan)

• Digital Public Infrastructure (DPI): India’s robust DPI (e.g., Aadhaar, UPI) is not replicated in many African contexts; limits plug‑and‑play.
• Scale‑driven business models: Certain models that work for a 1.4 bn population may be infeasible in Rwanda (10 mn) or Kenya (40 mn).
• Local ecosystem building: Need local developers, data pipelines, benchmarks, and trust to adapt Indian innovations abroad.

3.6. Designing a Planet‑Scale Intelligence Stack (Panel Discussion)

• Voice‑first interfaces are essential in India due to low typing literacy; voice is the “new keyboard”.
• Cost‑sensitivity drives demand: Any AI service must be affordable per interaction (e.g., < ₹0.20/min).
• Infrastructure layers must be cost‑effective, energy‑efficient, and aligned with everyday human capabilities (speaking, not typing).

3.7. Durable AI for Climate‑Stressed Physical Systems (Derick Jose)

• Multi‑modal sensing (“three‑eyes”):
  • Satellite – large‑scale emissions monitoring.
  • Drone – low‑altitude, pinpoint detection of methane leaks.
  • Ground sensors – localized data.
• Knowledge‑graph‑driven reasoning: AI ingestes structured (law clauses) and unstructured (sensor feeds) data to assess compliance across federal and state regulations.
• Outcome: Ability to predict fines (e.g., $240 mn penalty for a refinery’s methane leak) and proactively mitigate emissions.

3.8. Implementation Lessons – “Problem‑Statement First” (Gaurav Kataria)

• Iterative, baby‑crawls‑then‑runs approach: Identify a clear problem, redesign the process, define data needs, select the model, and then measure outcomes.
• Data as the bedrock: High‑quality data pipelines are non‑negotiable.
• Quality obsession: Across all factories and product lines, AI systems are tied to quality assurance (vision systems, defect detection).

3.9. Closing Reflections – From Projects to National Intelligence Infrastructure (Rashmit Singh Sukhmani)

1. Shift from “projects” to “intelligence infrastructure” – AI must be a persistent layer, not a one‑off effort.
2. From analytics to decision authority – AI should augment human decision‑making, not replace it.
3. Compute‑to‑decision economics – Sustainable compute (GPU, energy) is as critical as model accuracy.
4. Coordinated, not fragmented architecture** – Chips, sensors, models, applications, energy, and institutions must be jointly designed.
5. National capability – Foundational models and the supporting stack should be owned and usable by every citizen, positioning India as a global leader in applied intelligence.

4. Announcements & Notable Numbers

Announcement / Data Point	Details
$800 bn annual climate‑related infrastructure loss.	Highlighted to justify large‑scale AI investment.
1 bn people face severe food insecurity; 300 mn experienced extreme hunger last year.	Context for planetary‑scale intelligence.
AI‑teacher cost: ₹10 /min vs. target ₹0.10–0.30 /min.	Cost‑reduction goal for mass adoption.
Potential refinery fine: $240 mn (2025) for methane emissions.	Demonstrates financial risk of non‑compliance.
Scale of AI‑ready data centers: gigawatt‑scale, necessitating sovereign energy & compute.	Rationale for a national AI backbone.

5. Open Questions & Areas of Ongoing Debate

• Quantifying opportunity cost: How to build robust, comparable economic models for AI investment vs. traditional surveys?
• Business model adaptation for low‑population contexts (e.g., African nations) – what innovations are required?
• Voice‑first UX vs. textual interfaces: How quickly can large language models become truly voice‑native for Indian multilingual users?
• Governance of sovereign AI infrastructure: Balancing openness, security, and commercial participation.
• Incentive design at scale: Translating successful operator‑level incentives (as in the mayonnaise plant) to massive, multi‑sector ecosystems.

Key Takeaways

• **AI must be treated as national‑scale critical infrastructure—open, interoperable, and owned by the public sector—to achieve affordable, inclusive impact.
• Three systemic gaps impede scaling: talent/sense‑making capacity, fragmented institutional incentives, and lack of quantified economic cases.
• Cost‑effectiveness is paramount; a viable AI service for India should cost ≲ ₹0.20 /min, comparable to a glass of milk, not ₹10 /min.
• Trust and incentive alignment are the decisive levers for moving AI pilots to operational systems (e.g., mayonnaise plant, refinery monitoring).
• Voice‑first interfaces are essential for mass adoption in India’s multilingual, low‑literacy context; AI must adapt to human communication habits, not force new ones.
• Multi‑modal sensing (“three‑eyes”)—satellite, drone, ground sensors—combined with knowledge‑graph reasoning, enables high‑value compliance and climate‑risk monitoring.
• Process‑first, data‑first methodology: Redesign business processes before layering AI, ensuring high‑quality data pipelines and clear problem statements.
• From projects to infrastructure: The sector needs to shift from isolated AI pilots to a persistent, decision‑authority‑enabled intelligence layer spanning compute, sensing, applications, and institutions.
• National capability: Foundational models, sovereign compute, and coordinated ecosystems should be cultivated as a public good, positioning India as a global leader in applied planet‑scale intelligence.

---

Prepared for the AI‑India Conference, Delhi – 24 Feb 2026.

See Also:

• building-resilient-sustainable-ai-infrastructure-for-people-planet-and-progress
• ai-innovators-exchange-accelerating-innovation-through-startup-and-industry-synergy
• democratizing-ai-compute-and-digital-data-infrastructures