AI summary of understanding the Asset-Light Colocation vs. GPU Cloud Dynamics
The Bulk Business is Colocation: Anant Raj is not buying thousands of Nvidia H100s/B200s for their core business. They provide the real estate, utility-scale power architecture, shell security, and cooling. Hyperscalers or enterprise clients bring their own GPU clusters and install them in Anant Raj’s racks.
The Sovereign Cloud Exception: Anant Raj does run a small, managed services cloud platform called Ashok Cloud (partnered with Orange Business). They have historically run small pilot server deployments (<0.5 MW) where they do own the hardware stack, but management targets capping total cloud service mix at ~25% of their total MW capacity long-term, keeping the capital-intensive GPU risk contained. source
As liquid-cooling specialists, Submer is essential to making the Colocation model work for modern AI workloads:
- High-Density Requirements: Traditional data centers handle 5–10 kW per rack. AI GPU clusters (like Nvidia Blackwell architectures) require 40 kW to 100+ kW per rack. Air cooling cannot dissipate this heat.
- The Submer Role: Submer provides the modular liquid-cooling infrastructure and prefabricated mechanical, electrical, and plumbing (MEP) systems. source
- De-risking Capex: By integrating Submer’s tech, Anant Raj can offer “AI-Ready” slots to clients. The client brings the high-obsolescence hardware (GPUs), while Anant Raj provides the specialized infrastructure to keep them operational without absorbing the rapid technology risk themselves. source
Summary
The high capex (₹250 cr/MW) and 3-year technology risk apply to the entities buying the chips. Because Anant Raj focuses primarily on providing the physical infrastructure and liquid-cooling capability via Submer, they capture predictable, long-term infrastructure margins while shielding their balance sheet from direct GPU obsolescence.
Source: Submer and Anant Raj partner to accelerate sovereign, AI-ready infrastructure across India
