India AI Impact Summit Insights
All sessions

Flipping the Script: Re-Coding the AI Economy for the Global Majority

Contents

Executive Summary

This AI summit brings together startups, policymakers, and thought leaders from the Global South to reframe AI development around the needs of the global majority rather than replicating Northern models. The core message: the Global South can leapfrog by building trustworthy, efficient, locally-grounded AI systems that address development priorities in health, justice, agriculture, and education—while establishing new governance standards that emphasize sustainability, linguistic inclusion, and economic sovereignty.

Key Takeaways

  1. Purpose Before Speed: The Global South can leapfrog by asking "why and for whom?" before building, avoiding repeat of the North's unsustainable, inequitable AI rollout.

  2. Language & Data Are Sovereignty: Representing all 7,000+ languages and localizing datasets is not a nice-to-have—it's foundational to trustworthy, inclusive AI and freedom from outsourcing intelligence.

  3. Governance + Innovation Are Partners: Regulatory frameworks (public-private partnerships, risk-based approaches, consent-based data trusts) enable rather than stifle innovation when centered on development outcomes.

  4. Efficiency as Competitive Advantage: The Global South can build more efficient, sustainable AI (edge AI, SLMs, renewable-first infrastructure) at lower cost, capturing massive market share in inference-driven applications.

  5. AI Solves Real Problems Only With Interdisciplinary Buy-In: Lasting impact requires finance ministers, health ministers, farmers, and judges at the table—not just technologists—to translate AI into food security, health equity, justice, climate action.

Key Topics Covered

  • Startup Pitches & Technical Innovation

    • Legal tech (Runstack): AI-powered dispute resolution
    • Compute infrastructure (Turyam): Energy-efficient GPUs
    • Embedded AI (Craftify.ai): IoT/edge device development
    • Counter-drone systems (Armory): Defense AI
    • IoT/AI infrastructure (Ubic Edge): Prompt-based device automation

  • AI Literacy & Capacity Building

    • Multi-stakeholder approaches to AI fluency (not just awareness)
    • Foundational understanding of data, algorithms, and deployment
    • Moving beyond tech-for-tech's-sake toward purposeful innovation

  • Digital Infrastructure & Data Gaps

    • Internet connectivity disparities (35% in developing nations vs. 80% in advanced economies)
    • Language representation in AI models (7,000 languages globally; models trained on handful)
    • Data center capacity inequities (Africa: 18% population, <1% global data center capacity)

  • Sovereign AI & Language Models

    • India's multilingual LLM initiatives (Bhashini, Serum)
    • Representation of Indian languages (20+ constitutional languages, 100+ dialects)
    • Digital public infrastructure as model for AI data infrastructure
    • Data trusts and consent-based public datasets

  • AI Governance Frameworks

    • Policy postures: knowledge and capability-building vs. prohibition-heavy regulation
    • Public-private partnerships as governance mechanism
    • Pluralistic governance (not one-size-fits-all)
    • Africa's continental AI strategy ($60B fund, embedded ethics)
    • Risk-based frameworks in Latin America

  • Sustainability & Green AI

    • Data center energy consumption (400+ terawatt-hours in 2024; 1.5% of global electricity)
    • Carbon footprint of LLM training (equivalent to 300 transatlantic flights)
    • Small Language Models (SLMs) and edge AI as sustainable alternatives
    • Second-mover advantage for Global South in renewable-first infrastructure

Key Points & Insights

  1. Dispute Resolution at Scale (Runstack)

    • India's court case resolution time: 600 days → target 60 days via AI-powered litigation automation
    • 90% automation of litigation argument flow; preemptive dispute tracking on "day minus one"
    • Deployed with high courts and government agencies; measurable outcomes: lenders recover twice as well; insurers gain 2-quarter forecasting advantage

  2. Energy-Efficient Compute (Turyam)

    • 5–10x performance per watt advantage over Nvidia; focus on inference (not just training)
    • Open-source models approaching proprietary model efficiency (within 10% gap at <70B parameters)
    • Sovereign AI requires owning the entire stack (GPU + software)

  3. Embedded AI Automation (Craftify.ai)

    • Agentic AI reduces IoT/firmware development cost by 70%; 20-person teams → 2-person teams
    • Hardware-optimized code generation for microcontrollers and edge devices
    • Deployed on Nvidia Jetson, ESP32, others; significant timeline acceleration

  4. Defense & Counter-Drone AI (Armory)

    • Real-time radio signal processing for drone detection (5 km range) and jamming (3 km range)
    • India allocated $2B for counter-drone systems; 100 crores already ordered
    • Context: $500 drone can damage multi-billion-dollar assets

  5. IoT-to-Deployment Infrastructure (Ubic Edge)

    • Prompt-driven ("Hindi + English") device automation eliminates coding barrier
    • 25,000+ deployments scaled from 15–20 use cases to 5,000+ via agentic AI
    • Bridges gap between tech creators, system integrators, and field engineers

  6. Language Representation in AI

    • 7,000 global languages; current models trained on handful → underrepresentation of Global South populations
    • India's Bhashini initiative and Serum LLM target multilingual model parity with Gemini/OpenAI
    • Serum benchmarks show competitiveness with legacy closed-source models
    • Language inclusion = democratic infrastructure requirement

  7. Sustainable AI by Design

    • Global South can build "low-carbon AI future" from the start, avoiding North's industrial-era pollution mistakes
    • China's DeepSeek demonstrates cutting-edge models possible with lower compute; Global South should learn and improve
    • Edge AI + hardware optimization + SLMs + procurement discipline avoid "move fast and break things" mentality

  8. Governance as Enabler, Not Inhibitor

    • Friction can be a feature; purpose-driven AI slows intentionally where needed
    • Regulatory schemas: speeding tickets, licensure, prohibition—companies already self-regulate via policies
    • Multiple regulatory approaches coexist: Africa (flexible), Latin America (binding), India (mosaic of existing laws)

  9. Pluralistic Global Standards

    • Avoid "gold standard" convergence around dominant Northern models
    • Layered governance: shared norms (accountability, transparency, oversight) + diverse local frameworks
    • Legitimacy comes from diversity and pluralism, enabling resilience
    • Middle powers (not just frontier AI builders) should define accountability in resource-constrained contexts

  10. Infrastructure Gaps & Leadership Mindset

    • Africa: 600M without electricity; 0.x% global compute; 1% data center capacity; median age 19.7
    • Challenge: old rules applied to new transitory technology; ICT-only leadership insufficient
    • Solution: whole-of-society budget realignment toward talent, education, infrastructure, electricity, connectivity, use-case solving (food, health, education, justice, climate)

Notable Quotes or Statements

  • Reggie Townsen (SAS): "Friction is sometimes a feature... We have to get comfortable with the idea that sometimes you slow down in pursuit of purpose... What's the purpose [and] until someone describes for me that we're going to do something other than create a digital god, I just assume we slow down where it makes sense."

  • Ambassador Philip Thego (Kenya): "No one in this era should be forced to outsource their intelligence... For me that's about sovereignty and in most cases for people like us it's about freedom and self-emancipation. So we cannot have this new colonialism kind of in terms of intelligence."

  • Rachel Adams (Global Center on AI Governance): "Global majority governments are already leading the way in defining responsible AI in its own terms... we're not seeing global south countries localizing norms from the global north. What we see is them starting from a very different standpoint."

  • Kazim Rizvi (CORE AI, India): "Sustainability by design will only happen if there is regular access at different levels of compute... we have to balance it together—resource optimization AND access and democratization."

  • Reggie on AI Literacy: "We also need poets. We also need entertainers and fishermen and farmers. The goal is not to teach people AI, it's great if people want to become AI engineers and data scientists. We need more of those, but we also need [everyone to have] a foundational understanding."

Speakers & Organizations Mentioned

Panel Leadership & Moderators:

  • Rushali Savan (SAS; AI Ethics, Governance, Social Impact)

Panel Members:

  • Reggie Townsen (VP AI Ethics, Governance, Social Impact; SAS)
  • Rachel Adams (CEO, Global Center on AI Governance; Senior Fellow, University of Cambridge)
  • Ambassador Philip Thego (Special Envoy for Technology to President of Kenya; UN High-Level Advisory Board on AI)
  • Kazim Rizvi (Founding Director, The Dialogue; Founder, CORE AI [Coalition for Responsible Evolution of AI], India)

Startups Showcased:

  • Runstack (litigation/dispute resolution AI; clients include High Court, Government of Karnataka)
  • Turyam (GPU compute; 5–10x efficiency vs. Nvidia)
  • Craftify.ai (embedded AI/IoT development; 70% cost reduction)
  • Armory (counter-drone AI; 100 crores in Ministry of Defense orders)
  • Ubic Edge (IoT/device automation; 25,000+ deployments)

Investors:

  • Ankur Capital Fund (32.7 crores across two startups)
  • Piper Erica (34 crores investment)
  • Antler Innovation Fund 1 (21.8 crores across startups)

Government & Policy:

  • Government of Karnataka
  • Ministry of Defense (India)
  • UN Secretary-General (AI governance panel announcement)
  • African Union (2024 Continental AI Strategy; $60B pan-African AI fund)

AI Models/Initiatives Referenced:

  • Serum (India's multilingual LLM; benchmarked against Gemini, OpenAI)
  • Bhashini (India's language AI initiative)
  • DeepSeek (China; efficient compute example)
  • Nvidia Jetson (edge device)

Technical Concepts & Resources

Models & Frameworks:

  • Large Language Models (LLMs) vs. Small Language Models (SLMs)
  • Generative AI vs. broader AI landscape
  • Foundation models (multilingual, parameter-efficient)
  • Open-source models (within 10% efficiency gap at <70B parameters)

Infrastructure & Compute:

  • GPUs, TPUs, CPUs, FPGAs
  • Data center capacity & energy consumption (400+ terawatt-hours/year; 1.5% global electricity)
  • Edge AI & inference-driven architectures
  • Hardware-optimized code generation
  • Renewable-first data center design

Data & Language:

  • Multilingual datasets (20+ constitutional Indian languages; 7,000 global languages)
  • Data trusts & consent-based public datasets
  • National AI data repositories
  • Sectoral data trusts (healthcare, finance, agriculture)
  • Training data representation & bias mitigation

Governance & Policy:

  • AI Life Cycle framework (data → models → deployment → action)
  • Risk-based regulatory frameworks
  • Public-private partnerships (PPP)
  • Adaptive public procurement
  • Multi-stakeholder governance (government, industry, academy, civic orgs)
  • UN Resolution on AI (safe, secure, trustworthy AI for sustainable development)
  • Global Digital Compact
  • Digital public infrastructure (India's Aadhaar, UPI, ONDC as models)

Sustainability Concepts:

  • Carbon footprint of LLM training (~300 transatlantic flights per model)
  • Efficiency metrics (performance per watt)
  • "Low-carbon AI future" vs. replicating Northern industrial-era mistakes
  • Tiny ML approaches for context-specific tasks
  • Mobile-first deployment

Sector-Specific AI Applications:

  • Legal tech (litigation discovery, arbitration)
  • Healthcare (crop disease detection)
  • Defense (drone detection/jamming)
  • IoT/embedded systems (firmware, microcontrollers, robotics, drones, AI cameras)
  • Insurance claim forecasting

Referenced Reports & Initiatives:

  • From Capability to Constraint (Global Center on AI Governance report; QR code shared)
  • Africa's 2024 Continental AI Strategy
  • India's DPDP Act (Digital Personal Data Protection)
  • EU AI Act
  • Bhashini initiative (India)
  • CORE AI/The Dialogue (India think tank)

End of Summary