Particle.news

Zoho Unveils Nathu La, an India‑Designed Server to Control Its AI and Cloud Stack

The company plans to lower AI inference costs by running its ZLLM models on purpose‑built servers that it currently deploys only inside its own data centres.

Overview

  • Zoho publicly unveiled Nathu La on June 10 as a server platform designed in Nagpur to give the company tighter control over hardware, firmware, data centres and AI workloads.
  • The system was built over roughly five years with Intel collaboration using Intel Xeon 6 processors, while Zoho engineered the motherboard, chassis and modular subsystems in‑house.
  • Zoho says it has deployed about 1,000 Nathu La servers across its Indian data centres and reports 12–18% lower power use and a 20–30% reduction in total cost of ownership versus its prior platform.
  • The platform is integrated with Zoho’s in‑house large language models (ZLLM) to shift inference onto owned infrastructure and cut operating costs for AI services.
  • Zoho has filed multiple design patents, used local EMS partners for assembly and involved more than 300 SETU‑trained engineers, and it says offering preloaded servers to enterprises or government customers is a possible future step without a set timeline.