Keynote

Abstract

In this keynote, Fujitsu’s CTO Vivek Mahajan presented the company’s vision for “sovereign AI” – a strategy that gives enterprises full ownership, flexibility, and security over their data and AI models. He traced Fujitsu’s 90‑year legacy, highlighted breakthrough hardware (2 nm ARM‑based servers, the upcoming 1.4 nm chip, and an exascale AI supercomputer), outlined a world‑leading quantum roadmap, and described next‑generation networking and open‑source AI software stacks. Mahajan emphasized domain‑specific, secure AI platforms (Takane LLM, Kozuchi agentic model) and the convergence of compute, network, and software into physical‑AI edge devices such as robots, drones and medical instruments. The talk concluded with a call for partnerships and an invitation to the next fireside chat.

Detailed Summary

Mahajan opened by linking his topic to the previous plenary discussion on sovereignty.
Sovereignty, as defined for India and similar nations, means flexibility + security:
- Ownership & control of data
- Ability to build, tune and modify AI models without dependence on third‑party platforms.

2. Fujitsu’s Historical Context

Fujitsu is a 90‑year‑old technology company with origins in mainframe computing alongside IBM.
Notable past achievements noted:
- Development of the one‑megabit DRAM and early mainframe business.
- Recent launches: world’s first 2‑nm ARM‑based servers, and a new U.S. brand (2021) that bundles Fujitsu solutions for customers.

3. The Three Pillars of an AI Platform

Mahajan framed an effective AI platform as requiring three core components:

Pillar	What Fujitsu Provides	Why It Matters for Sovereignty
Software	Open‑source AI stack, no vendor lock‑in	Full transparency & adaptability
Compute	High‑performance CPUs, GPUs, NPUs, exascale supercomputers	Local, powerful processing reduces external reliance
Networks	Photonic & wireless solutions, ultra‑high‑capacity switches	Secure, low‑latency data movement across sites

4. Compute Innovations

4.1 Legacy Supercomputing Leadership

Fujitsu operated the world’s fastest supercomputer for five consecutive years.

4.2 Upcoming Exascale AI Supercomputer

20 AI‑exaflop supercomputer slated for ~2 years from now.
Powered by the Fujitsu Monaca chip (2 nm, ARM‑based, highly power‑efficient, with built‑in confidential computing).

4.3 Near‑Term Server Roadmap

Test servers (ARM‑based) arriving in ~2 months.
1.4 nm processor line – world’s first, featuring:
- 256‑core CPU + 128‑core CPU
- Integrated NPU for AI inferencing
Two versions: CPU‑only and CPU + NPU (targeted at sovereign AI inference workloads).

4.4 Open Software Stack

Entire stack is open‑source, ensuring no lock‑in to Fujitsu‑only tools.
Optimised for AI, data‑center, and HPC workloads.

4.5 Inference Focus

Emphasis on small‑ and medium‑size language models that can be run in private or semi‑private environments, crucial for sovereign AI.

5. Quantum Roadmap

Fujitsu positions itself among the top three global quantum players.
Roadmap highlights:
- 250 logical qubits by 2030 – the most ambitious publicly known target.
- 1,000‑qubit machine to go live next month in Kawasaki, Japan.
- 10,000‑qubit machine expected in ~3 years.
Fujitsu builds its own control systems and cooling infrastructure.
Quantum + HPC convergence: Jointly powering mission‑critical AI workloads.

6. Networking Breakthroughs

6.1 Photonic Switching

Developed a 1.6 TBps optical switch delivering:
- Low power consumption
- ~1,000 km transmission range
- Low latency for data‑center interconnects (critical for Indian data‑center builds).
Planned upgrade to 3.2 TBps.

6.2 Wireless & Open‑RAN Integration

Leveraging open‑RAN and a network orchestration stack to move AI workloads efficiently across wireless and wired domains.

7. AI Software Stack & Domain‑Specific Platforms

7.1 Platform Philosophy

Domain‑specific, secure, and flexible AI solutions (e.g., defense, nuclear, healthcare).

7.2 Key Products

Platform	Purpose	Security Feature
Takane	Large language model (LLM) platform	Built on Fujitsu’s internal security framework
Kozuchi	AI‑agentic model	Powered by the same security platform; integrates third‑party tools

Both platforms enable fine‑tuning for unique organisational needs while maintaining data sovereignty.

7.3 End‑to‑End Solution

Fujitsu offers a total solution: compute → network → AI software → domain‑specific applications (government, manufacturing, finance, healthcare).

8. Physical AI & Edge Devices

Fujitsu is developing Kozuchi Physical OS, an intelligent operating system for robots, providing “brain‑like” memory to prevent forgetting.
Vision extends to edge devices: drones, medical equipment, and even consumer health gadgets (e.g., smartphone health sensors).

9. Ecosystem Partnerships

Emphasis on collaborative development rather than a solo approach.
Notable partners mentioned:
- AMD, Intel, NVIDIA (GPU‑Hybrid Architecture)
- Lockheed Martin (defense)
- Super Micro
- Robotics manufacturers (physical AI partnership announced in October).

10. Closing Remarks & Transition

Mahajan reinforced that sovereign AI is a strategic imperative for Fujitsu globally and specifically in India.
He thanked the audience and announced the next session – a fireside chat featuring Vivek Kaneja (C‑DAC), Nitin Bajaj (Intel), moderated by Aman Khanna (Fujitsu Asia), with a brief request for speakers and audience to clear the stage.

Key Takeaways

AI sovereignty = ownership + flexibility + security; a cornerstone for Indian and global AI strategies.
Fujitsu’s 90‑year legacy underpins its current push into 2 nm/1.4 nm chips, exascale AI supercomputing, and open‑source AI stacks.
Monaca 2 nm ARM chip (with confidential computing) will debut in test servers within two months; a 1.4 nm, 256‑core CPU + NPU follows shortly, targeting sovereign AI inference.
Quantum roadmap: 1,000‑qubit machine next month, 10,000‑qubit system in three years, and a target of 250 logical qubits by 2030.
Network innovation: 1.6 TBps photonic switch (upcoming 3.2 TBps) delivering low‑power, long‑range, low‑latency connectivity for data centers.
Open software stack ensures no vendor lock‑in, enabling customers to tailor AI workloads across CPUs, GPUs, and NPUs.
Domain‑specific platforms (Takane LLM, Kozuchi agentic model) provide secure, fine‑tuned AI for sensitive sectors (defense, healthcare, finance).
Physical AI vision: Kozuchi OS for robots, extending AI capabilities to edge devices such as drones and medical gadgets.
Fujitsu leverages a broad partner ecosystem (AMD, Intel, Lockheed Martin, Super Micro, robotics firms) to deliver end‑to‑end sovereign AI solutions.
The keynote set the stage for a fireside chat that will explore practical implementations and collaborations in the Indian AI landscape.

See Also:

India AI Impact Summit 2026

Explorer

keynote-1