India AI Impact Summit Insights
All sessions

Keynote by Vivek Mahajan | CTO, Fujitsu | India AI Impact Summit

Contents

Executive Summary

Vivek Mahajan, CTO of Fujitsu, presents a comprehensive vision for AI sovereignty — enabling countries like India to build independent, secure, and controlled AI infrastructure without overwhelming reliance on third-party providers. Fujitsu positions itself as a complete-stack technology provider (compute, networks, software) offering sovereign alternatives through cutting-edge hardware (2nm and 1.4nm chips), quantum computing, and open-source AI software platforms.

Key Takeaways

  1. Sovereignty is Strategic: Countries like India cannot afford to be entirely dependent on foreign cloud providers and proprietary AI stacks. Building indigenous capability across compute, networks, and software is both possible and necessary.

  2. Hardware + Software Convergence Matters: Sovereign AI requires owning the full stack from silicon (2nm/1.4nm chips with security) to open software (no lock-in) — not just adopting US-designed GPUs.

  3. Inference, Not Just Training: The near-term AI value lies in deploying fine-tuned, domain-specific models for inference at the edge (small/medium LLMs), not only large-scale LLM training.

  4. Quantum + Classical Hybrid is Next: Quantum computing (1,000 qubits live in 2024; 250 logical qubits by 2030) will complement HPC for mission-critical AI — this requires early investment and partnership now.

  5. Open Ecosystems Reduce Risk: Complete open-source stacks (no Fujitsu lock-in) lower switching costs and enable organizations to customize AI without being trapped by proprietary vendor choices.

Key Topics Covered

  • AI Sovereignty: Data ownership, security, flexibility, and independence from third-party control
  • Fujitsu's Technology Portfolio: 90-year history spanning mainframes, quantum computing, and advanced semiconductors
  • Advanced Computing Infrastructure: 2nm ARM-based Monaka servers, 1.4nm processors, quantum roadmap
  • Quantum Computing: 250 logical qubit roadmap by 2030; 1,000-qubit machine launching in Kawasaki (2024)
  • AI Exascale Supercomputers: 20 AI exascale machine launching in ~2 years
  • Network Technology: 1.6 Tb/s optical switches with focus on power efficiency and long-range transmission
  • AI Software Stack: Domain-specific models, fine-tuning capabilities, security-first architecture
  • Physical AI Platform: Convergence of compute, networks, and AI in edge devices (robots, drones, medical devices)
  • Sovereign vs. Third-Party Dependency: Contrasting building vs. renting AI infrastructure
  • Strategic Partnerships: Collaboration with AMD, NVIDIA, Lockheed Martin, robotics manufacturers, and others

Key Points & Insights

  1. Sovereignty Definition: Sovereignty means flexibility + security: owning and controlling your data, building domain-specific models, modifying/tuning them independently, and reducing third-party reliance.

  2. Three-Pillar AI Architecture: Effective AI requires compute, networks, and software working in concert. Fujitsu focuses on independence in all three rather than outsourcing.

  3. Hardware Leadership: Fujitsu ships the world's first 2nm ARM-based servers (Monaka chip, 2 months from keynote date) with built-in confidential computing for hardware-level security; upcoming 1.4nm processors will feature 256-core CPU + 128-core CPU + NPU for inference.

  4. Quantum as AI Enabler: Quantum computing (250 logical qubits by 2030) combined with HPC will drive mission-critical AI workloads; represents convergence of quantum + classical computing for hybrid workloads.

  5. Open Software Stack: Fujitsu's AI platform is completely open-source without vendor lock-in; fine-tuned for AI, data centers, and HPC on Monaka servers.

  6. Inference-Focused Architecture: NPU design targets inference (not just training), enabling deployment of small and medium language models in private/semi-private environments — critical for sovereignty.

  7. Domain-Specific Model Priority: Healthcare, defense, government, finance, manufacturing data should not live on public clouds. Fujitsu's platform (Takane LLM + Kosuchi agentic model) enables organization-specific AI without public exposure.

  8. Power Efficiency at Scale: 1.6 Tb/s optical switches reduce power consumption while maintaining long-range transmission (1,000 km) and low latency — essential for data centers in India and elsewhere.

  9. Physical AI Convergence: The future is edge-deployed AI (robots, drones, medical devices, healthcare apps) requiring integrated compute, networks, and AI security — not centralized cloud.

  10. Strategic Non-Alignment: As a Japanese company with Japan-made technology, Fujitsu offers a geopolitically and technologically neutral alternative to US semiconductor/AI dominance for countries prioritizing sovereignty.

Notable Quotes or Statements

  • "Sovereignity is being flexible and being secure, right? You want ownership of your data. You want to control that data..." — Defines AI sovereignty for the Indian context.

  • "This is what you need for AI, right? So, all the key areas are a lot of open source software that we have fine-tuned..." — On the open-source commitment of Fujitsu's stack.

  • "We are a Japanese company. Our technology is made in Japan. And that's where what we find ourselves at a very interesting point because we are a choice to lot of American companies uh as an alternative." — Emphasizes geopolitical positioning.

  • "If you don't have a platform that helps you deliver that you're never going to be sovereign you're never going to control the AI business" — Warns against fragmented, point-solution approaches.

  • "Robots tend to forget... what we're working on some intelligence work and research that will so that robots can continue to remember..." — On physical AI challenges in edge deployment.

Speakers & Organizations Mentioned

Speakers:

  • Vivek Mahajan — CTO, Fujitsu (primary speaker)
  • Tokitaan — CEO, Fujitsu (partnership announcement with NVIDIA)
  • Jensen — NVIDIA (implied, co-announcer of physical AI partnership)
  • Vivekanandan — Executive Director, CDAC (mentioned for follow-up session)
  • Nitan Bajaj — Director Sales and Marketing, Intel (mentioned for follow-up session)
  • Aman Khanna — Vice President, Asia Group (moderator mentioned)

Organizations:

  • Fujitsu (Japanese multinational technology company)
  • AMD, Intel (CPU competitors; also partners)
  • NVIDIA (GPU partnership on physical AI)
  • IBM (historical mainframe competitor)
  • Lockheed Martin (defense partnership)
  • Super Micro Computing (physical AI partner)
  • Nokia (wireless/network competitor/partner)
  • CDAC (Centre for Development of Advanced Computing, India)

Technical Concepts & Resources

Semiconductor Technology:

  • 2nm (Monaka chip) — Current-generation ARM-based processors for Fujitsu servers; Japan-made; power-efficient data center focus
  • 1.4nm — Next-generation processors with 256-core CPU, 128-core CPU, NPU variants
  • ARM Architecture — CPU instruction set (not x86/Intel-based)
  • NPU (Neural Processing Unit) — Hardware accelerator for AI inference

Computing Platforms:

  • Monaka Servers — Fujitsu's flagship 2nm ARM servers shipping in ~2 months (from keynote date)
  • 20 AI Exascale Supercomputer — Launching in ~2 years
  • Confidential Computing — Hardware-built security layer for data protection

Quantum Computing:

  • 250 Logical Qubit Roadmap — Target by end of 2030 (ahead of competitor timelines)
  • 1,000-Qubit Machine — Launching in Kawasaki, Japan (~1 month from keynote)
  • 10,000-Qubit Machine — Live in ~3 years
  • Quantum + HPC Convergence — Hybrid classical-quantum for mission-critical AI

AI Software Platforms:

  • Takane — Large language model (LLM) platform by Fujitsu
  • Kosuchi — Agentic AI model platform with security integration
  • Open Software Stack — No vendor lock-in; open-source AI, data center, HPC-focused tools

Network Technology:

  • 1.6 Tb/s Optical Switch — Long-range (1,000 km), power-efficient, low-latency
  • 3.2 Tb/s Switch — Next-generation upgrade
  • Open RAN — Open Radio Access Network (partnership for network orchestration)

Physical AI / Edge Deployment:

  • Kosuchi Physical OS — Intelligence layer for robots/edge devices with memory persistence
  • Edge Devices: Robots, drones, medical devices, healthcare apps on iPhones

Historical Reference:

  • 1 Megabit DRAM — Early Fujitsu innovation
  • Mainframe Business — Fujitsu pioneered alongside IBM
  • World's Fastest Supercomputer (5 years) — Historical achievement (timeline not specified)

Context & Significance

This keynote aligns with India's broader AI sovereignty push and reflects geopolitical tensions around semiconductor and AI supply chains. Fujitsu's positioning as a non-US vendor offering complete stack independence resonates with government and enterprise concerns about:

  • Data localization and privacy
  • Reducing US technology dependency
  • Building indigenous AI capability
  • Long-term cost control via open software

The emphasis on inference over training, domain-specific models, and edge deployment also reflects practical enterprise AI maturation beyond centralized LLM services.