India AI Impact Summit Insights
All sessions

Stacked for Scale: Semiconductors and Foundational AI

Contents

Executive Summary

India is positioning itself as a primary architect of the global AI hardware stack, moving beyond being merely a technology consumer. Through the India Semiconductor Mission (ISM), the government and emerging startups are building a complete semiconductor ecosystem—from chip design and manufacturing to packaging and R&D—with ambitions to create global-scale semiconductor companies within 10–15 years. The session featured policy leadership and three semiconductor startups demonstrating that India has the talent, capital, and government support to compete globally.

Key Takeaways

  1. India is building, not just consuming: The presence of 24 design startups, three scaling hardware startups at the panel, and government-backed manufacturing fabs marks a fundamental shift. India is architecting its own AI hardware stack, not licensing it.

  2. Talent + Government Support ≠ Instant Success: While India has the talent (20% of global semiconductor designers), the real challenges are supply chain monopolies (TSMC, HBM vendors), market access barriers (CUDA lock-in, hyperscaler relationships), and the missing systems/OEM layer. Policy must address these, not just talent.

  3. Moonshot thinking with 10-year timelines: Agrani (GPU), C2I (power chips), and Mangrove (SOCs) are betting on becoming ₹1B+ companies by 2035. This mirrors India's fintech and space wins. Credible team backgrounds and government backing make this plausible, but supply chain and market access remain the hard problems.

  4. R&D investment in foundational technologies is non-negotiable: Advanced nodes, materials science (diamond substrates, silicon photonics), equipment, and software ecosystems cannot be outsourced. ISM 2.0's explicit focus on R&D across all verticals is strategically correct and addresses the ecosystem's weakest link.

  5. The next 12–24 months are critical: ISM 1.0 projects entering revenue phase (C2I tape-out April 2025, Mangrove generating revenue in 2024), ISM 2.0 rollout imminent, and the first 10 manufacturing projects starting commercial production define whether this is sustainable momentum or one-off success.

Key Topics Covered

  • India Semiconductor Mission (ISM) strategy and evolution: ISM 1.0, 2.0, and future roadmaps
  • Design ecosystem: 24 design startups approved; 14 have secured VC funding
  • Manufacturing and packaging: 10 approved projects (1 silicon fab, 1 silicon carbide fab, 8 packaging units)
  • R&D priorities: Advanced technology nodes, silicon photonics, materials science, equipment development
  • Three startup case studies:
    • Agrani: Building GPUs from the ground up
    • C2I: Power management chips for AI data centers
    • Mangrove (Mine): Microcontrollers and microprocessors for edge applications
  • Marvel's role: Global infrastructure company doing cutting-edge work in India
  • India's competitive advantages: 20% of world's semiconductor designers based in India; talent density in design and software
  • Supply chain and market access challenges: Wafer allocation, high-bandwidth memory (HBM) bottlenecks, market penetration strategies
  • Ecosystem gaps: EDA tools access, systems design capability, OEM/ODM absence, software ecosystems (CUDA equivalents)
  • Geopolitical and policy considerations: Make in India provisions, government procurement, data sovereignty, technology standards
  • AI acceleration in chip design: Using AI to shrink development cycles and improve design efficiency

Key Points & Insights

  1. Rapid ecosystem maturation: India had zero semiconductor startups in 2020; by 2024 there are 24 design companies with significant VC traction. This mirrors India's successful patterns in fintech (60% of global real-time payments by 2025) and space tech (300 startups today vs. 2 in 2015).

  2. Complete stack ambition: ISM 2.0 will expand beyond design/manufacturing/packaging into R&D (advanced nodes, materials, silicon photonics, quantum), equipment manufacturing, and chemical/gas sourcing—recognizing that sustainable ecosystem requires vertical integration and control of foundational technologies.

  3. Talent is not the bottleneck: 20% of global semiconductor designers work in India. Founding teams have 25–30 years of experience at Intel, AMD, TI, IBM. The real constraints are supply chain access, market entry, and ecosystem breadth (not talent quantity).

  4. Supply chain monopolization is the critical blocker: TSMC wafer starts and HBM (high-bandwidth memory) are sold out globally. Only 3 vendors (Samsung, SK Hynix, Micron) produce HBM. Global companies must "partner upward" to access these, limiting Indian startups' independence. This is described as a "sleepless night" problem requiring government intervention (e.g., Micron allocation agreements).

  5. Market access requires government policy beyond capex subsidies: ISM is rightly tapering capex incentives over ISM 1.0 → 2.0 → 3.0, betting on ecosystem pull. But strategic procurement (defense, space), Make in India policy enforcement for quality standards, and ensuring Indian market access (₹80B GPU opportunity over 4–5 years) are needed. "We're not asking for favors; this is our technology for our market."

  6. Software ecosystem is critical but underdeveloped: CUDA dominance means any GPU company must solve firmware, middleware, and compiler ecosystems—a long-term race. India is strong in software but lacks CUDA-equivalent open ecosystems at the systems level. This requires parallel R&D investment.

  7. Systems design and OEM/ODM layer missing: India has jumped from services to chips, skipping the intermediate "electronics → products → systems" trajectory. Companies like C2I (power delivery for AI clusters) and Agrani (GPUs) need systems companies, board designers, rack integrators, and OEMs to successfully bring products to market globally.

  8. AI's impact on semiconductor design cycles is unquantified but strategic: Using AI for design automation, iteration acceleration, and reducing tape-out cycles could compress development timelines. This is a "strategic angle" Naven flagged but noted all numbers are "gut feel," suggesting R&D needed here.

  9. Population-scale AI and cost inflection points: India's diverse, geographically dispersed population creates unique challenges (aggregating volume across disaggregated markets) but also opportunities. Delivering voice-based AI services at ₹5/minute requires reimagining the entire stack—a problem India is uniquely positioned to solve.

  10. 10-year horizon is realistic but requires ecosystem patience: Government messaging (Prime Minister quoted) frames this as a 20–25 year marathon. But panelists believe 10 years to global competitiveness is achievable if all pieces align: ISM 2.0/3.0 execution, R&D breakthroughs, supply chain partnerships, and policy-enabled market access.

Notable Quotes or Statements

  • Amitesh Singh (Additional Secretary, ISM): "This is not a 100m race, it's a marathon. Countries with complete semiconductor setups have taken 20, 30, 40, 50, 60 years to reach this level. Nothing happens here in a day."

  • Rajan Anand (Panel moderator): "2015 India was not in the top 100 countries for digital payments. 2025 we were 60% of all global real-time payments. Today there's no question in the world who leads in digital payments." (Analogy for what India can achieve in semiconductors.)

  • Dimant (Agrani, GPU startup): "Building a GPU is second nature for us. We've been doing it for 30 years at Intel, AMD, IBM. Now we're going to build it for India. In 10 years, we want to be in the top five fabless companies in the world."

  • Pritham (C2I, power chips): "When you're bringing in a megawatt of power to this room, 30–40% is already lost in distribution and conversions before reaching the GPU. Power lost is power not generating tokens. That's the problem we're solving."

  • Naven (Marvel): "We're not just building this for the world; we're building solutions for the AI of the world. When AI is ready, how do we use it to turn around and design the chip in 12 months?" (On using AI for semiconductor design acceleration.)

  • Amitesh on long-term vision: "By 2030, we'll have 10 projects fully functional. This 10 can become 15, 17, or 18. We want to see them as Broadcom, Qualcomm, Nvidia-type companies eventually."

  • Shashwat (Mangrove, SOCs): "We built this chip with a total team of seven people. We're young, we're agile, and in a marathon, you're paying attention to what you're doing and improving yourself rather than keeping your eye on competition."

Speakers & Organizations Mentioned

Government & Policy:

  • Amitesh Singh – Additional Secretary, Ministry of Electronics & IT; CEO of India Semiconductor Mission
  • Ministry of Economy and Finance of Uzbekistan (Aziz Aram, Director of Semiconductor Strategies) – attending as observer

Startups (Three Panel Founders):

  • Agrani – GPU startup; founder Dimant (formerly Intel, AMD, IBM)
  • C2I – Power management chips; co-founder Pritham (formerly Texas Instruments India, built 5 founders + 45-person team from TI and other companies)
  • Mangrove (Mine) – Microcontrollers/microprocessors; co-founder Shashwat (from academia; youngest team; 7-person design team)

Global Companies:

  • Marvel – Naven (leading India operations); $70B market cap; infrastructure/data center chips
  • Texas Instruments (TI) – Mentioned as source of founding talent
  • Intel, AMD, IBM – Referenced for GPU/processor design expertise
  • TSMC – Foundry for tape-outs
  • Broadcom, Qualcomm, Nvidia – Benchmarks for future Indian companies
  • Samsung, SK Hynix, Micron – HBM memory vendors (oligopoly)
  • Google, 11 Labs – Comparison for voice/Indic language models

Investors & Ecosystem:

  • Peak 15 (venture firm) – Backed Mangrove; announced 4 semiconductor investments
  • Incor – RISC-V IP company (2nd investment)

Other Notable Mentions:

  • Saram – Indian voice/speech model company; recently achieved SOTA in Indic language speech-to-text and text-to-voice
  • CDC (mentioned in context of high-power computing) – Government computing infrastructure

Technical Concepts & Resources

  1. Semiconductor nodes: 28nm, 180nm, 7nm, 5nm, 4nm, 3nm, 2nm – technology maturity levels
  2. Silicon technologies: Silicon photonics, silicon carbide (SiC), GaN (gallium nitride), sapphire substrates, diamond substrates
  3. Memory technologies: HBM (High-Bandwidth Memory), DRAM scaling, 448 GHz serializer-deserializer (SerDes) IP
  4. GPU/chip architectures: 100 billion transistor GPUs, power delivery systems (800V → 800mV conversion, 2,000A current delivery)
  5. Packaging: Advanced packaging, thermal management, interconnects (copper and electrical)
  6. Software/frameworks: PyTorch, CUDA (Nvidia's software ecosystem), compilers, low-level system programming, firmware, middleware
  7. Design tools: EDA (Electronic Design Automation) tools – flagged as costly barrier for startups; government considering subsidies
  8. AI applications in design: Using AI to accelerate chip design cycles, reduce iterations, improve design efficiency
  9. Process technologies: Display (OLED), quantum computing, materials research
  10. Systems concepts: Data center AI clusters, GPU racks (scale measured in gawatts of compute), power distribution, thermal cooling systems
  11. Standards/protocols: Risk-5 (open ISA), data sovereignty considerations
  12. Models in Indic languages: Saram (speech-to-text, text-to-voice SOTA); 12 Indian AI model companies launching models ranging from 1B to 100B parameters

Methodology & Research Approach:

The session employed a policy-academic-industry panel format combining:

  • Government roadmap presentation (ISM structure, timelines, R&D priorities)
  • Founder storytelling (team backgrounds, technical challenges, moonshot ambitions)
  • Industry perspective (global company's cutting-edge India work)
  • Audience Q&A (international observers, product companies, investors) revealing ecosystem gaps

Key Research Implications:

  • The semiconductor design bottleneck is not talent but supply chain and systems design.
  • Parallel software ecosystem development (beyond CUDA) is essential for GPU competitiveness.
  • Population-scale AI optimization is a unique Indian problem/opportunity requiring bespoke hardware design.