AI Impact Summit in New Delhi, Abu Dhabi-based technology group G42 announced a landmark partnership to build a national-scale AI supercomputer in India.
This system, with a peak compute capacity of 8 exaflops, marks India’s transition into the exaflop-scale computing era. It is set to be roughly 19 times more powerful than the combined capacity of India’s current leading AI systems, AIRAWAT and PARAM Siddhi-AI.
Key Partners & Implementation
The project is a collaborative effort involving several high-profile international and domestic entities:
- G42 (Abu Dhabi): Leading the project deployment and providing the “Intelligence Grid” infrastructure.
- Cerebras Systems (USA): Providing the hardware, specifically their Wafer-Scale Engine-3 (WSE-3) chips, which are among the fastest AI processors globally.
- C-DAC (India): The Centre for Development of Advanced Computing will host and co-manage the facility.
- MBZUAI (UAE): The Mohamed bin Zayed University of Artificial Intelligence will serve as a research partner.
Focus on Sovereign AI
A major highlight of the deal is its focus on Sovereign AI—a concept where a nation maintains strategic control over its AI infrastructure and data.
- Local Hosting: The supercomputer will be physically located in India.
- Data Residency: All data processed by the system will remain within Indian jurisdiction to meet national security and compliance requirements.
- Indian Governance: The system will operate under governance frameworks defined by the Indian government as a cornerstone of the India AI Mission.
Impact on the Ecosystem
The supercomputer is designed to be a “democratized” asset, meaning it won’t be restricted to just one company or department. Access will be extended to:
- Startups & SMEs: Reducing the high cost of entry for small businesses wanting to train large models.
- Research Institutions: Enabling breakthroughs in healthcare, agriculture, and climate modeling.
- Government Ministries: Supporting public service applications tailored for India’s 1.4 billion citizens.
Supercomputing Scale Comparison
| System | Capacity | Current Global Rank (Approx.) |
| AIRAWAT (C-DAC Pune) | 200 AI Petaflops | #188 |
| PARAM Siddhi-AI (C-DAC) | 210 AI Petaflops | Top 250 |
| New G42 Supercomputer | 8,000 AI Petaflops (8 Exaflops) | Top 10 (AI-specific) |
Broader Context
The deal follows the December 2025 release of NANDA 87B, a high-performance open-source Hindi-English large language model co-developed by G42 and MBZUAI. The new supercomputing capacity is expected to significantly accelerate the development of more “India-centric” models like NANDA.
