StopDownOpen the app

Partnership · AI & Deep Tech

Cerebras partners with G42 to build India's largest AI supercomputer

By ·

Date
Company
Cerebras Systems
What it does
Builds massive AI processors
Kind
Partnership
Amount
$5.55 billion
Founded
2015
Sector
AI & Deep Tech

What they do

Cerebras builds giant chips using entire silicon wafers for high-speed AI training

What happened

India and UAE executed a partnership to deploy an 8-exaflop cluster in India

Why it matters

The project brings sovereign compute power to Indian startups and researchers onshore

The details

  • UAE President gifted a Cerebras chip to PM Modi marking the Condor Galaxy India partnership.
  • The deal involves deploying 64 Cerebras CS-3 systems to build one of India's largest AI clusters.
  • The partnership signals growing importance of AI infrastructure in India-UAE cooperation.
  • The project aligns with domestic ambitions by bringing sovereign AI compute onshore.
  • It reduces dependence on foreign cloud infrastructure for startups and researchers.
  • PM Modi and UAE President announced the project during a visit in January 2026.
  • A term sheet was signed between G42 and C-DAC at the India AI Impact Summit.
  • G42 will handle installation, deployment, operations, and maintenance of the system.

The bigger picture

  • The project addresses data sovereignty by hosting infrastructure and data within India.
  • It provides affordable access to frontier-scale computing power for local entities.
  • The capacity is almost 19X compared to India's two flagship AI supercomputers.
  • It offers greater control over data residency, security, and national control.

About the business

  • Cerebras builds a single massive processor called the Wafer-Scale Engine (WSE).
  • The company uses an entire silicon wafer instead of individual chips.
  • They solve integration challenges by adding spare cores to reroute around flaws.
  • Cerebras went from selling hardware boxes to building giant cloud supercomputers.
  • The technology claims up to 20X faster AI training and inference speeds.
  • It trains models with up to 24 trillion parameters without code splitting.
  • The WSE-3 features 44 GB of on-chip SRAM for rapid data movement.
  • The chip houses over 4 trillion transistors on a single piece of silicon.
  • It claims aggregate bandwidth of 214 petabits per second.

What happens next

  • The facility will help train large AI models much faster.
  • It will offer affordable access for research and commercial use.
  • The project will likely be used for drug discovery and disaster management.
  • It aims to simulate smart energy grids and foster research.

The deal

Type
partnership

Founders

  • Andrew Feldman, CEO
  • Gary Lauterbach, Co-founder
  • Michael James, Co-founder
  • Sean Lie, Co-founder
  • Jean-Philippe Fricker, Co-founder

Read it as a card in the app

Same day

All startup news on 16 May 2026 →